On the Development of a Rod-Less Pendulum Clock

Foreword

My interest in clockwork has been subliminal for most of my life. My first experience was when I was
seven years old. My grandfather had a wind-up train with a circle of track on a green painted piece of
plywood. I coveted it so much I was having dreams about it and its little puffs of smoke magically
coming out of the stack. All trips to play outside went by way of the garage where it sat on sawhorses,
begging me to play with it. Alas, I was instructed never to touch it.

I pressed and pressed, and finally my grandfather went to his bedroom, collected the alarm clock beside
his bed (a large golden brass one with two large bells on the top), and sitting me down with a couple of
screwdrivers said: “I’m tired of hearing about the train. When you can take this clock completely apart
and put it back together so that everything works, you can have the train.” No help was forthcoming, and
he continued on with his day.

I was forced to take meal breaks that day, and was the proud owner of the train by the next morning, and
tore down and built up the alarm clock several times thereafter. More times, I think, that I played with the
train.

Another brush with clockwork was as a young aerospace designer. I had built an escapement that
oscillated a dumbbell to rate limit the deployment of an extendable mast for a spacecraft, which I now
know to be a Foliot, or as I thought at the time, just a great “energy eater”.

It was not until being recommended to The Longitude1 in my early 40’s that I discovered a deeply
neglected interest in precision clockwork. I have spent my adult life creating devices for precision
motions, various mechanisms for spacecraft and large scale dynamic spacecraft models that only slightly
pre-dated our supreme confidence in finite element modeling of large structures in a weightless
environment.

I definitely could relate to Harrison’s mind-set, at least the part that was his tilting at a windmill
undaunted. I’ve since gone to Greenwich and watched his fantastic clocks running until the docent
shooed me out. I am still in awe of the accomplishment and the tools to which Harrison was limited.
Sadly, at the time I had no idea of the significance of the Shortt clock and others I may have glanced at on
my way to see Harrison’s. Shortly thereafter, I got Philip Woodward’s MORT2 and after devouring it in
quick bites, was instantly, insatiably caught up in the state of the art, and his W5, and had the great
pleasure of watching his amazing movement as interpreted by David Walters in (D)W5 in the same
month.

I’ve had an extremely pleasurable time with the Horological Science Newsletter back-issues, and realize
that I’ve only scratched the diminishingly thin patina on the surface of research available to me, and in
fact, as of this writing I am only up to 1999, but have read 2009. I must say that the HSN’s contributors
are a fine bunch of scientists, and that the state of the art is consummately presented, and the math so
beautiful and rendered understandable (though not always entirely by me), for the most part supported by
data of test run durations that stagger my mind.

1
I’ve read of several ingenious methods of removing, mitigating or compensating for circular error of
compound pendulums in HSN; too many to cite, and wanted to propose something original if at all
possible. As I embarked on creating something of interest to this tough room, I am fully expecting
responses such as “so and so beat that to death in 1680! Next!” That’s OK, if so.

In the face of far more accurate electrical means, I find the free pendulum clock to be the ultimate
expression of mechanism, and its optimization a profound riddle in mechanics and physics.

I had the pleasure of presenting an outline of the following work over dinner at David Walter’s house,
[after experiencing, photographing and filming my third (D)W5] with John Kirk in attendance. It was a
pleasurable night of tech talk, over a wonderful single malt and Bolognese; talk that would leave most
regular folk yawning and snatching surreptitious glances at wristwatches far more accurate than any I will
likely create.

Why attempt a precision clock in the first place? Dallas Cain put it perfectly when he wrote: "the exercise
has therapeutic benefits"3 I could not agree more. My life as a mechanical designer has to date always
had a capitalistic bent. This is for the pure joy of it. I hope you like my take on the problem.

Mike Everman, Santa Barbara

everman@bell-everman.com

On the Development of a Rod-Less Pendulum Clock

It should be noted that any hardware work I’ve done on this project can be found in videos on
youtube.com, under the Author “belleverman”. There are several videos in a chronological progression
that support this text. I am also writing this under the assumption that the reader is already familiar with
the “circular error” of a simple or compound pendulum, and why realizing a practical pendulum that
follows a cycloid path is a good thing.

Creating a perfectly cycloidal path for a pendulum bob as most will attest is no small task, and from what
I can gather, not done well yet for large swing angles. For me, the challenge of friendly competition with
centuries of very creative inventors is quite motivating. Certainly several designs and corrective methods
have gotten very close. My quest began with a simple question: Can it really be that hard? The simple
answer is yes, which is the base requirement for a thought experiment of any merit.

Some would argue that the following walk through various cycloid generating mechanisms, rife with Q
killing rollers and wrap bands, with pinch points that would not suffer a single skin cell in the wrong
place, and that the idealized cycloid pendulum bob is a point mass; but I “press on regardless”. I will
eventually think of a hundred ways that mine is a bad approach, but so far have been able to answer each
of my own objections in turn, sparing the project from the waste bin. I have however altered my original
intention toward a precision regulator, to that of unique and maybe-never-tried clock of respectable
accuracy.

My impetus here is that true or reasonably good isochrony would eliminate some difficulty elsewhere in
my clock project, so it’s worth shooting for, or getting near. Of one thing I’m sure; a more complete

2
analytical treatment and equation of motion would have to come from more capable mathematicians than
me. I can only get so far with the graphical and spreadsheet solutions that are my forte, and clearly I’m
not shy about sharing my early blatherings in a field new to me! I am admittedly a neophyte, and this
science is new to me, so please forgive my lapses in proper terminology.

I began with mapping out some four-bar and slider-cam arrangements, eccentric pivots, and tautochrone
ramps with rolling bob; coming quickly to the conclusion that a cycloid could or I should say must, be
created using cylindrical and flat surfaces that can be made with highly exact profile and finish. While a
grinding rig could be made fairly easily to cut this tautochrone path directly as a track, this approach’s
inelegance and preparatory work was less than appealing. As another resource I wanted to bring to bear,
one of my patents deals with wrap band flexures with large excursions (for a flexure that performs
straight line motion)4, which are similar but different in a critical way to a Rolemite, which I see has had
some interest here in HSN as a pendulum pivot means5. Bill Ellison in HSN 1994-5 reported on some
experiments and put out a call for what the opposing surfaces look like in order to use a Rolemite to make
an isochronic pendulum. I’ll not be using a Rolemite, but it certainly could be used, and the surfaces turn
out to be very simply a known cylinder and a flat plane, just like one would have hoped for. But, I’m
getting ahead of myself.

First out of the gate was to of course begin at the beginning: understand how is a cycloid generated, and
discover if it can move from schematic representation to something real in a practical manner, eschewing
the pitfalls of Huygens’ “cheeks” or “chops”, tricks of evolutes or other complex curvatures difficult to
generate in a manufacturing sense.

Thanks to Christiaan Huygens we know that the proper cycloid for a given simple pendulum length, Lsp,
is generated with a point on a circle, which is rolling “under a shelf”, and whose radius, Rc= Lsp /4, Figure
1, (re-illustrated owing to countless references, with my own nomenclature). Of course what follows is
applicable to any pendulum period desired, but I will be focusing in a hardware sense on a seconds
beating pendulum, so Rc=248.41mm.

Figure 1, Cycloid generation basics

3
The basic element

What I consider the basic element is the simplest possible reduction of the schematic in Figure 1 to
hardware, where a precision ground arc segment roll or shoe of radius Rc is lightly and magically held to,
and rolls under an actual shelf, with rod extending down to a bob whose center is also on R c. This makes
a pendulum that is now half the length of the seconds beating pendulum it replaces, or 496.82mm, which
sweeps a rather larger angle than its predecessor for a given amplitude.

The method of holding this shoe to the underside of the shelf can be magnetic (Fig. 2a) or by way of a
simple roller at the pivot point, Pc, running on top of another shelf (Fig. 2b). I originally spent a good
deal of time imagining various configurations for the magnetic method, with my major concern being the
elimination of extraneous forces in the X direction, which would directly turn into unwanted torque about
Pc. I can imagine numerous effects degrading the consistency of the magnetic field, and I’m certain I
would get mired with niggling issues with this approach. The mechanical roller has far fewer potential
“gotchas”, if one can accept sources of friction that directly degrade the pendulum Q.

Figure 2a and 2b, The Basic Element with two support schemes

Both approaches above meet the main constraint, that the cycloid is generated with purely circular and
planar surfaces which can be made with great exactness. This would be something to see, I think, and no
matter how much further you’ll see me taking the reduction in pendulum length in what follows, 2b will
eventually be the subject of a hardware build.

The subject of reducing the length of the pendulum while remaining a seconds-beating one is not a
requirement, rather it is about making the cycloidal path perfectly with no mucking about. Length

4
reduction to the point of conceptual “Maximum Rediculosity” is an interesting, perhaps not practical
outcome of this exercise, yet it is the subject of the rest of this paper because it is faaaascinating to me.

While any prototypes will display a reasonable effort toward thermal insensitivity, the primary thrust of
the models will be show a degree of isochronism as an initial impulse decays, the instrumentation used to
prove it one way or another, and the fun of the exercise no matter how quickly they ultimately ring down.

Reducing the length by half again, or taking it to “Length Rediculosity Phase One”

The basic element said something to us about what needs to happen to the rod and bob, namely: as the rod
segment immediately above the bob goes through angle α, its pivoting center (the center of the cycloid
generating circle, Pc) must translate a distance ∆x = Rc • α. Simple enough as identities go. A further
critical observation, obvious as it may be, is that that center moves in a straight horizontal line.
Understanding this, I then attempted folding the now half meter seconds pendulum rod at the virtual pivot
point Pc, and considering how it must act if rolling now on top of a shelf, coincident with the bob center
(at zero degrees) instead of under a shelf, and how on Earth to make the pendulum do what we want?

Figure 3, Folded Basic Element schematic

Schematic Figure 3 has the pendulum length now equal to Rc (248.41mm). This again elicited attempts at
straight line mechanisms to eliminate a linear slider for the pendulum pivot Pc, with the simple
requirement that the pendulum rod must still rotate about Pc, but with an angle opposite in sign of the
supporting rod and shoe, symmetrical about a vertical line.

An “aha moment” came when I considered two support rods and shoes, spaced apart, connected by a link
between their pivot centers, which would perform the straight line motion needed , and provide a pivot
attach point for the pendulum (Fig. 4). Looking ahead, the synchronizing links have been put in a
configuration such that they can be pure tension members.

5
 Figure 4, Dual Support Folded Basic Element

It should be noted that each supporting rod assembly must have a balance mass above its pivot point (Fig.
5), so there is no restoring torque toward vertical, save for that provided by the pendulum.

Figure 5, Dual support physical assembly

While the physical synchronizer links are accomplished with flexure bands, most likely of .001 to .0005”
full-hard 301SS feeler stock, it is also possible to be straight links and jeweled pivots; the physical
embodiment of the schematic links in Figure 4. I believe the bands are far easier to realize than tiny
jeweled links, of which several are required, needing unreasonably equivalent lengths. This one would be

6
fun to build and watch go; just don’t push it too far! It may in fact be as far as I should take the
progression, but let’s continue on anyway.

Reducing the support legs to zero length

Why would we do such a thing? Because we can remove a length of material (the support legs) that must
be thermally compensated, and simplify the assembly in general. Or just for the fun of it, as you prefer.
As the flexure pivot is configured using flexure bands, we are presented with an opportunity to reduce the
length of the support legs by adding a step up ratio between the pendulum pivot and the support leg
pivots, adding no additional rolling or band flexure contact points. If this ratio is 1:2, then the support leg
lengths can be reduced by half, and now sweep an angle that is 2α. The pivot system still now translates
the required distance ∆x, with the support plane now moved up half the distance to the pivot, Figure 6. A
larger step up ratio allows us to delete the legs completely, moving the pivot support plane to the
underside of the r2 rolls. Judicious material choices for r1 and r2 can make this a thermally stable section
of the system, compensating for the expansion of r3.

Figure 6, Applying a ratio to reduce support lengths

Pendulum pivot radius r1 acts against support pivot radius r2, creating ratio r1/r2. The radius of the support
roll that actually runs on the support plane is r3, so

, Eq. 1

where the alphas of course cancel. As an example, if we drive r1 and r3 to a radius of 50mm, the radius of
r2 is sought, so

Really, we’ve not reduced the support legs to zero, but have reduced their effective length to r 3, in this
case 50mm, removing the need for physical support rods. A further advantage is that the support rolls are

7
now circular and inherently, or I should say easily balanced, eliminating the balancing weights above the
support pivots, though of course these rolls must still be quite well balanced individually.

While this example gets us a reasonable flexure bend radius at r2, the support rolls are rather large and
ungainly (100mm diameter). If the pendulum pivot were pushed larger, r1 equal to 75mm for instance,
becoming more of a “T” with arc segments at 9 and 3 o’clock, and r2 left at 10mm, then r3 can be of a
more reasonable size,

giving us a pendulum system that looks like this:

Figure 7, Cycloidal seconds pendulum, closer to realistic, not to scale

The physical embodiment of this is simplified over the design with support legs in Figure 5, though I
expect it would have greater frictional losses, and still a great deal of what I term “parasitic bits” moving
to and fro.

8
Pendulum length reduction, taking it to “Maximum Rediculosity”

At least one sleepless night had me tossing about, certain that things could get simpler (if not more
practical, but I’d exchanged faith for practicality long ago), and by simpler, I mean less parts and effort to
see something run. It dawned on me one could perform the same rod length reduction-and-elimination
logic to the rod that holds the bob; but could it go as far?

Applying this logic to Figure 7, you can readily see that if you halve the rod length, you must merely
double r1, or halve r2 (referring to schematic Figure 6), and keep going from there. In truth, I’ve had to
come back here and fill in this blank between Figures 7 and 8 with some logical step, since it was an
intuitive leap that took me straight to the rod-less solution, and the subsequent pajama and toe stubbing
victory dance. Suffice to say, with the benefit of hindsight, if one makes R2 equal to R3, then the
equation tells us that R1 becomes 248.42mm, or call it Lsp/4, which is equal to Rc. By definition then, if
this curve is in fact Rc, then the CG of the bob mass must be on it, and the rod is gone!

Figure 8, Rod-less Assembly

It should be noted that the support rollers now have an irrelevant diameter, and it should now be obvious
that Figure 8 is moving things up and out properly as it moves. A property of idler rollers is that they can
be of any diameter, creating no reduction or step-up between two surfaces, while reversing the surface
translation of one to the other, which was the angular sign change or “mirroring” that I needed with the
“folding” step in Figure 3. So if the pendulum’s CG were still on the radius Rc, any diameter rolls at any
spacing would suffice between it and a flat surface, and the translation per angle α would be proper. It’s
only the flexure band that may care if they are two of the same diameter, and it does.

While it’s a simple matter to design the weights under the the CG to be proper, I’ve shown it made with
weights that are basically the same part at half thickness on either side as a simple conceptual start on

9
balancing. Again, while remotely possible to off-load the bob mass magnetically, it will likely be fraught
with stray field problems and I do not think I’ll try.

A lower friction concept would be to provide a shoe between the support plane and the pendulum radius
that has air-bearing surfaces, radial above, planar below, while keeping the flexure bands and rollers in a
cavity within to enforce the translation-per-rotation. While air-bearings are neat, viscous damping should
be considered, and while I know my way around an air-bearing, it’s not exactly the least expensive thing
to whip up.

It should be noted that the flexure bands in all of these are purposefully wide in order to take a slight bit
of shear (perpendicular to the plane of the pendulum), with the idea that rate can be slowed by tilting the
entire assembly out of the vertical plane; so the geometry must be made such that it has a bit of a gaining
rate initially. It is not my preferred method of rate change, but the clock will certainly be sensitive to
level.

Subsequent work

I have done some hardware work, and am presenting whatever I can before the publishing deadline. Of
course there is far more to add in order to have an actual timekeeper and if one of these pendulums is in
fact all that, then I hope I don’t completely blow the mechanical escapement! Hopefully, some of the
onus of escapement error at least can be removed with one of these methods.

I have made a breadboard of the rod-less concept, Figure 9, using a segment of a thin section ball bearing
outer race for the arc segment, a hardened and ground plank, two bearing outer races for support rolls, and
a .001” band of feeler stock. The band acts much like a tractor tread to constrain the motion to a straight
line, and to keep the support rolls from squirting out. There are several spot welds for this band, some
affixing it under the arc segment, some on the outer quadrant of each support roll, and a few affixing the
band to the base plank at the mid-length.

Figure 9 Basic Cycloid Arc Bearing, Rc=3.375”+, without and with bob attached

The bob rests on two pins, on the inside of the arc segment. Its aggregate CG is situated approximately
.010” below the Rc tangent point, so later I can shim it up in stages through the sweet spot and watch the
period and isochronism. A sensitivity analysis that I should perform is for the change in period based on

10
the CG’s vertical adjustment, placing it on a curve that is parallel, but not coincident with the cycloid
curve defined by Rc. It certainly changes the period, but is no longer on the “prime cycloid”, and will
have a slightly different circular error correction. “Slightly”, when we are talking about splitting
millionths of an inch is something we might know analytically, but within constraints of reasonable
roundness, smoothness and support roll stiffness, I can only hope that its trend is somewhat the right
direction in the end. I do hope that it’s close enough that this becomes the principal method of changing
the period to hit a specific number, rather than tilting the assembly. I am currently at 50.0xx cycles per
minute with this breadboard using the following electromagnetic drive, and it is exhibiting a steadily
declining rate that has not yet flattened out as of this writing. If I can adjust it to exactly 50cpm, the gear-
train becomes a simpler matter than some transcendental number, and this breadboard can become an
actual clock.

Breadboard impulse drive

As I have the great fun of designing a fully mechanical escapement for this pendulum, filling notebook
pages with big X’s, one after another, I’ve in the meantime put an electromagnetic drive to it that
functions quite well. I just wanted to make it go, and while this is old hat for most reading this, hearken
back to your first sustaining pendulum and you'll know how I feel. Just marvelous! I was up at all hours
to take what data I could or just go out to the garage and see if it was still happy.

Figure 10, Electromagnet drive setup (dark module is power relay, white is delay)

11
 Figure 11, Sensor and paper flag with gap at middle

Drive parts are:

(1) Optical sensor "light on, closed" when occluded, positioned at zero
(1) Solid state adjustable delay-on-make relay, sensitivity set to .1 to 1sec
(1) Solid state relay to take the low current sensor output and fire the delay relay which is 24v and in-
line with the coil
(1) LCD counter
(1) Electromagnet from a 24vdc contact relay
(1) LED to indicate when the coil is firing
(1) 24vdc power supply

Here is its sequence of operation, considering that period is 1.2 sec, (.3 sec to either extreme from center).
Also given that the flag is always occluding the sensor, closing the circuit other than at that “deadband”
window about zero.

1. Pendulum moves from any direction through center.

2. Half the flag gap after center, sensor encounters flag edge, switching power relay on, which
powers the delay relay.
3. Pendulum travels to full amplitude and begins traveling back to zero with nothing happening
yet.
4. At some point determined by the delay, the delay relay switches the coil on, pulling the bob
toward zero as it is approaching zero.
5. The sensor hits the flag gap and the delay relay is shut off automatically (a bit before reaching
zero).

An electronics guy would make short work of a real circuit for this, but this was a very simple, and
admittedly very low fidelity way to make it go.

12
Since it takes .6 seconds to travel from zero to one extent and back to zero, then a delay of .6 seconds
would do nothing; it will have gone out and back to zero before the delay can activate the magnet.
Setting it to less than .3 seconds will retard, firing the magnet before full travel, which pulls it back, so the
period would be reduced. Setting it between .35 and .5 or so insures that it's pulling when the bob wants
to be pulled. The higher the number, the less time it's under power before the flag gap shuts it off. I've got
roughly a 1.2 sec period, so the delay is set to fire the electromagnet about .4 sec (unverified) after
passing zero (and half of the sensor flag gap), on the bob's way back to zero.
Yes, it's not sinusoidal or proper in the slightest. It directly affects the period, and the flag gap width
complicates the simplification above. It is certainly advancing the rate set up this way; approximately 1
cycle per minute more than a single sided impulse.

It was initially varying a few beats per minute, mainly I think because there's nothing particularly
precision about the delay relay, the parts have intentionally not been precision cleaned or polished, and
primarily because the electromagnet, pulling the bottom of only one of the bob disks was setting up a bit
of rocking out of plane, and subsequent walking of the bob along the arc. I glued the bob to the arc
segment temporarily, and have seen a much more consistent period that is varying a few milliseconds per
cycle. While not impressive horologically, I’m thrilled just the same at this crude stage.

Further work

On the hardware side, I’ll be changing the flag and coil location so that I can have a pulse that is centered
about zero, which only works as a unidirectional set up, in an effort to reduce impulse error. I will also
move the electromagnet above the bob, and center it on the oscillation plane.

As for my instrumentation approach, I’m planning for Bryan Mumford’s MicroSet, but for now, and until
Christmas, will use what’s at hand. One highly appropriate direct angular feedback method is putting
20µm optical encoder tape on a curved surface below the bob, and mounting a read head below and at
zero that will give me a non-contact arc-wise resolution from 1µm down to 5nm. Currently, it’s my
wristwatch, my LCD counter and a whiteboard. My family is shaking their collective heads at the
expanse of non-sensical numbers, and my curious excitement over them.

Conclusion

Most of what I’ve proposed has some unknowns that are frankly beyond my ability to analyze, like:

Are there moments in the short pendulum rod, Rc, or the bob itself, which sweeps a larger angle than the
simple pendulum rod and bob it replaces, that detract from the desired goal? This generally applies to
whether the bob’s moment of inertia, while certainly robbing energy from this system as it oscillates
about its center, is doing so sinusoidally, that is, harmonically. It may be an over-simplification, but I’ve
taken to thinking of parasitic masses as being OK if harmonic, and their deviation from sinusoidal
velocity profile a telling measure of how they detract from any potential isochronism.

Are my roller carriages and rods in fact exhibiting harmonic motion and how do or don’t they affect the
moment of inertia of the pendulum? Obviously, it’s important to know if these ancillary bits must be
extremely light, extremely low moment of inertia or not. I have likened the rollers to a foliot; masses that
do not follow the cycloid, but may fortuitously be a harmonic energy sink. I suspect that the support rolls
may in fact be moving harmonically, which says to me that they are merely an energy drain on my

13
escapement, not affecting the frequency. It would be interesting to see what would happen if I were to
test the period, having changed the mass of the rollers or their moment of inertia, or both. As a design
parameter, I’ve been assuming that I’d like the rolls to be less than a percent of the bob mass, with
nothing solid mathematically to go by as yet. It’s a large enough percentage to be physically possible.

It was at this point, satisfied that my “parasitic bits” (the support rolls or disks) were minimized, I began
to wonder how bad a situation I am in with the bob’s moment of inertia, with my mechanism having to
rotate the bob back and forth on its axis with no positive attribute of this motion that I can see. My
mind’s eye sees this motion as the greatest detractor of any potential isochronism.

Some references had me torn as to whether allowing the bob to be free about its axis, or for it to rotate
through the angle α as my breadboard does. One reference told me it does not matter either way, which
simply cannot be right. To my way of thinking, the (perhaps unattainable) goal is to get the CG of all
molecules following the cycloid without any rotations. I believe I must embrace this and that the only
way to make the bob simulate a point mass is to allow it to be rotationally free as in the model depicted in
Figure 12. Unfortunately nothing can be done about the arc segment’s moment of inertia other than to
minimize it as much as possible, and depending on its magnitude, it may drive me to some corrective
means.

Figure 12, Rotationally Free Bob (ball bearings not shown)

Can the rollers run over a speck of dust with no issue, provided it is not at or near the extremes of travel?
A few quick amplitude tests showed decent period consistency while somewhat dirty, but at very low
amplitude, I could see the bob literally bouncing back from a bump somewhere in one of the eight pinch-
points! Maybe I should abandon isochrony and put a dog hair in the right place to get the period I want.

Lastly, I need to determine the right combination of support roll/disk diameter and material properties at
all line contacts, given a bob mass, with the goal of least mass and rolling contact losses.

14
What’s not different about this pendulum is that it is subject to all of the thermal and barometric issues
one would normally encounter on a conventional pendulum. It is heartening that I need only address the
thermal growth of the arc radius, Rc, as one would address the length of a normal pendulum rod, but with
dramatically different means. Stabilizing the support rolls is irrelevant, and the support plank must
merely remain flat. As with a conventional pendulum, mine allows supporting the bob at its center so the
CG does not shift with temperature.

I expect to make a higher fidelity version of this with sapphire disks for support rolls, and ULE or
Zerodur for the Rc arc segment. Depending on the result of “bob rotating” or “bob free” study, I will
determine whether the bob itself needs special treatment, though I anticipate that someone will point me
to a mathematical case for this, rendering the question moot before I get around to building Figure 12. I
will build the next one that way anyway, as I expect that “free” is the answer. Thermal changes to the
bob’s moment of inertia would be irrelevant, and it could grow or shrink about its CG willy-nilly, with
what has to be an irrelevant change in windage. A side note about “all things windage-like” is the
eventual plan for grooves on all rolling surfaces either side of the guide band, so these pinch points leave
somewhere for the air’s boundary layer to go. I imagine losses just from “squeegeeing” and re-filling this
sticky layer of surface air. It’s got to add up!

As all of these ideas are competing with jeweled, flexural or knife-edge pivots which are of ludicrously
low friction, and since all of my proposed ideas involve some form of rolling Hertzian stresses, losses due
to these stresses must be minimized by material choice, finish and cleanliness. Harder, stiffer, smoother
and cleaner is the mantra. In any case, I cannot expect a very high Q with any of these ideas, which is
arguably (and I’ve read only some of the arguments) not necessarily a bad thing.

“So much to do and so little time.” I am reminded of a comment made to me recently: One looks at
one’s watch not because we care exactly what time it is, but to know how much we have. I would love to
know the source of that beautiful observation. My response was that I want to know exactly how much I
have!

References:
1
“The Illustrated Longitude: The True Story of a Lone Genius Who Solved the Greatest Scientific
Problem of His Time,” Dava Sobel and William J. H. Andrewes, Jan. 2003
2
“My Own Right Time: An Exploration of Clockwork Design,” Philip Woodward. Sept. 1995
3
“The Ultimate Pendulum: Higher Q or Better Drive? Part One,” Dallas Cain. Horological Science
Newsletter, NAWCC Chapter 161, 1995-1
4
“Macro Motions with Macro Flexures with Emphasis on Vacuum Compatibility”
Everman, M. R. (Bell-Everman, Inc.) ASPE Spring topical meeting proceedings, May 2006
http://www.aspe.net/publications/Spring_2006/spr06abs/1875.pdf
5
“SUSPENSION SYSTEM DESIGNED TO PROVIDE ISOCHRONOUS MOTION OF THE
PENDULUM,” Bill Ellison. Horological Science Newsletter, NAWCC Chapter 161, 1994-5

15