Their basic principles of operation incredibly seem to have been omitted, apparently without exception, from explanations of mechanical calculator principles. (Although, I have not seen that wonderful, costly French book.) If anyone knows why, I'd love to be clued in!

I'm quite exhausted, but I couldn't resist reading and doing some study. I'm too tired to write a good message, but wanted to share my enthusiasm (not sure I'll succeed!).

First off, these machines operate on a principle quite different from all others. Machines like the Friden and Monroe, as well as Odhner, Brunsviga, Facit (just about sure), Curta, Egli, etc. all start and stop their accumulator dials. They accelerate a dial rather suddenly. move it the appropriate number of digits, and then stop it suddenly. If you hold down the Plus key/add bar on the electrically-driven machines, the accumulator dials start and stop during each machine cycle. (This also happens in the Curta, but its accumulator dials are so tiny that the start-stop movement really doesn't matter.)

All these machines (fairly sure!) set aside time later in the cycle to take care of carries.

Well, the Marchant is uniquely different. Instead of accelerating the dials to full speed regardless of the digit being accumulated, they move the dials at a speed proportional to the digit being entered. A 1 is entered 1/9th as fast as a 9, although both a 1 and a 9 take the same amount of time to be entered.

If you hold down the plus (add) key on a Marchant, the dials move steadily (and very fast! Fairly sure they run at 1,300 cycles/minute, a speed that would tear up the stop-start machines in short order) (exceptions being 1) the Curta, which might work that fast, but please, no motor drives for Curtas! and 2) the Monroe PC1421, which was probably the ultimate in sophisticated, modern mechanical design.

In the Marchant, carries are done by 10:1 reduction gearing (watt-hour meter dial style, in principle) from a dial to its neighbor to its immediate left. Of course, this means that if a dial contains a 5, its neighbor to the left would be misaligned by half a digit. Yes, there are snail cams, followers, and differentials to keep the dials aligned.

The body of the machine contains what could be called a 9-speed transmission for each of 10 (or 8) columns. (Do you see why the parts count reaches 7,000?) This is the really clever part.

Get this! There are three driveshafts. For each machine cycle, one turns 1/12 of a revolution, another, 1/4 of a revolution, and the last, 1/2 of a revolution.

Keep in mind that what counts (pun) is the number of gear teeth by which you move an accumulator dial for each cycle. For a digit 7, you move the accumulator dial drive gear 7 teeth.

OK; with that in mind, the 1/12 rev. shaft carries two gears, one with 12 teeth, and the other with 24. These gears take care of digits 1 and 2. (The 24-tooth gear moves by 2 teeth per cycle.)

The 1/4-revolution shaft carries gears with 12, 16, and 20 teeth, to give movements of 3, 4, or 5 teeth per cycle. Similarly, the 1/2-rev/cycle shaft carries gears with 12, 14, 16, and 18 teeth, for 6, 7, 8, and 9 teeth per cycle. Although this might seem obvious when explained, i.m.h.o. it is a really clever scheme.

To make practical use of such an elegant scheme, the complexity mounts considerably. The whole accumulator moves down into mesh with the main body of the machine, and the dials look really shy hiding down in their holes (If, indeed, you can even see them; it depends where your eye are).

Carries are radically different from the schemes used in the "start-stop" machines (basically, these detect a transition between 9 and 0, and set a mechanical one-bit memory. They then use these bits to make the drive increment the appropriate dials.)

In the Marchant, the 10:1 gear reduction actually does the carries, but things aren't really complete and resolved, until the accumulator rises, and cam followers contact the snail cams to align the dials. If this sounds almost analog, it is, almost!

Quite surprising, and equally elegant, is that apparently, division is basically short-cut!! In such division, the leftmost few digits of the partial remainder in the accumulator are compared with the number in the keyboard, because it might require fewer cycles to avoid counting down to overdraft. (It's easy to demonstrate on a hand-operated machine.

Sorry to be somewhat vague...)

I'd heard rumors that there's a magnitude comparator in Marchants. It isn't explicitly an isolated mechanism (not quite!); it's more or less "built in" to the existing mechanism. But, nevertheless, it does consider the relative values of the leftmost 3 digits or so of the divisor vs. the partial remainder in the accumulator.

This magnitude comparator has an analog "feel" to it, being driven partly by the snail cam followers.

There are several clutches in these machines; nothing simple like the one-clutch Friden, for instance. The main drive gear train is unexpectedly complicated. Indeed. it puts a new perspective on the Friden, which seems simple!

As I now understand it, when you do one of the most basic operations, a single add cycle, here is (approximately!) what happens. Your digit keystrokes position little dials to confirm what you entered, as well as a stack of thin, small cams, one stack per digit.

When you push the plus/add key, you start the motor, and a clutch closes to "sample" the cam stack to choose one of the gears I mentioned earlier. This sets up the gear ratio that corresponds to the number.

Once this is done, the "transmission" is locked into this ratio. The next phase is moving the accumulator "into the dip", to use Marchant lingo. (Ernie must be having fits by now, reading my generic names! Marchant's terms differ...) Going into the dip releases locks that kept the accumulator dials in position, and puts the accumulator into mesh with the "transmission"s output gear.

Now, it's time to trip another clutch and cycle the three shafts (1/12, 1/4, and 1/2) to enter the keyed-in number to the accumulator. Once this is done, the whole set of events has to be undone!

The accumulator comes out of the dip, the dials get over their shyness, and magically align to show the new sum, carries and all. (The followers for the snail cams move into contact with those cams, more like sensing fingers.) The "transmission" is reset to neutral.

If you think this is bad, division gets downright hairy. Apparently, typical division stops subtracting >before< (!) an overdraft occurs, but maybe not always? The mechanism that does this, the magnitude comparator (more or less implicit in the mechanism) is quite amazing and really has something of an analog "feel". The snail cam followers are in contact, and as the dials rotate, the follower movements have really interesting consequences. Division seems to be really complicated in detail.

Although the machine does a fancy sequence just to add a number, where it really shines is in semiautomatic multiplication ("mult"). It is so fast that you can serially enter the digits of the multiplier, and it gobbles them up like a starving Brontosaurus. I personally remember trying consecutive 9s, and finding that the delay was really minimal.

It's a considered opinion, and I'm not an M. E., but the control linkages and cams seem to be more complicated than they have to, but maybe I'm wrong. The internal details when all is taken into account become formidable, especially in division.

Are the basic principles elegant? Imho, beyond question; they have, for me, been a lifetime inspiration. The full details seem somewhat less so, but the complexity and sophistication of the design are truly notable.

There's hope of scanning the illustrations in the training manuals, doing a lot of cleanup, and publishing them with descriptions on Ernie's pages. It will take a while, but will definitely be a labor of love (unless someone wants to pay me to do it!). It wouldn't (at present, at least) be reasonable to Web-publish most of the contents of these manuals; really few people could use the detail. However, the basics cry out for publicity.

Now to get a couple of hours' sleep before I get ready for work..

.February 4, 1998.

In the ongoing process of learning to speak Marchant, an enterprise which Ernie Jorgenson is generously helping along, a few more items have come up. The older Marchants (with the Proportional Gears) had no fewer than SIX clutches, if I understand matters right! Later machines eliminated one clutch. Seems it's time to talk about single-revolution clutches; these are nothing new, and what I'm describing is not specific to any company's designs.

In this context, what we're talking about is a special kind of clutch, which like others, connects or disconnects the driven member from the driver. However, these are single-revolution (or, in some cases, half-revolution) clutches. If the driver side is rotating, the driven is not, to begin with. As well, these do not generally depend upon friction; once engaged, they lock up quite solidly.

A dog (that's a mechanical term) "blocks" the clutch to keep it out of engagement. When something (such as a keystroke) moves the dog, the clutch locks up (no slipping), and the driven member moves in sync with the driver. If the dog doesn't move back into position to block (part of) the clutch, it will continue to stay locked up, and the driven member turns exactly with the driver.

When the clutch should disengage, the dog moves back to get in the way, and when that part of the clutch comes in contact. the clutch quickly disengages with the driven member in a well-defined position. (One advantage is that the drive motor can coast after its power is cut when the machine cycle(s) is/are complete. The machine generally must not overshoot into the beginning of another cycle; when the driven member has done its thing, that's it, and it stops quickly. The drive motor, however, really needs to coast.)

This kind of clutch doesn't allow any significant drag when disengaged.

By one method or another, the driven member continues moving a bit after the process of disengaging starts, so that the parts that came into contact to drive the load are now separated by some small distance, or are otherwise relatively disconnected. (One kind of single-revolution clutch, the so-called "wrap spring" clutch, lets the spring ride on the cylinder inside it, but it absolutely has to have good lubrication. Such clutches were used in Flexowriters, and worked really well.)

These clutches are used in many places, from the huge punch presses that stamp out useful parts from sheet-steel strip stock, to the clutches in the Selectric typewriter and teleprinters, as well as, of course, mechanical calculators.

==========================

Division in the Marchants amazes me. If I ever understand it well, I'll have reason to be proud. To learn it well, it seems you really need a good manual, a properly-adjusted machine with covers off, and a handcrank. (All calc. techs. need handcranks!) The basics are the same as on other machines. BUT... The basic accumulator design makes it really "hairy" to sense either a real overdraft, or an *impending* one. That latter is what I find almost mind-blowing! When I first heard that the Marchant has a magnitude comparator in it, I figured that someone was kidding, or I misunderstood. (Maybe I still do misunderstand.)

Anyhow, seems that Marchants sense the decreasing magnitude of the number in the accumulator, and apparently compare something like the leftmost 2 or 3 accumulator digits with the corresponding digits in the keyboard (latter is the divisor). (Sorry if this repeats material in my last message!). Apparently, even >before< an overdraft actually occurs, the mechanism trips, and off it goes to shift the carriage and resume its wonderfully smooth, 1,300 cycle-per-minute subtraction.

Whether I have it all straight yet (and I do hope I'm not misleading enthusiasts!), it is definite that the snail cams on the leftmost nonzero accumulator dials do cause their cam followers to eventually trip the division mechanism, so it can do what it has to, to start a new quotient digit.

Because these cams have a smooth profile (no bumps, or just one really big one, really) and the dials are moving at constant speeds, the tripping process is, it seems to me, Hold on to your hats...

NOT a digital process, but an ANALOG one!!!

I don't think I'm wrong on that one. (Not even a quantized analog, such as the positioning of the "golf ball" in a Selectric, either.)

I finally had a delightful voice chat (old style POTS, not Internet voice (yet)) with Ernie. I was already aware in a general way that the div. trip mechanism is a critical adjustment. He said that if the adjustment isn't "sensitive" enough, div. simply won't trip, and apparently the machine not only creates an overdraft, but keeps right on subtracting essentially forever. On the other hand, if it's too sensitive, it trips too soon, and has to run for tens of cycles instead of 9 at most to reduce the dividend. (It doesn't give a wrong quotient, however.)

Numerical examples of the above, dividing 20 by 3:

Fails to trip:

20 - 3 = 17 - 3 = 14 - 3 = 11 . . . = 2 - 3 = [all 9s] -3 = 999...996

minus 3 = 999...993, and it goes on forever.

Trips too soon: (Ernie: Reasonable example?)

20 - 3 = 17 - 3 = 14 - 3 = 11 (premature trip, causes carriage shift too soon)

Because of the shift, the 11 now looks like 110 to the keyboard, as to speak.

110 - 3 = 107 - 3 = 104 - 3 = 101.

You get the idea. It take a while; the counter dials do have carries, so it works out OK.

Do realize that because the dials are all moving at constant speeds, the sudden change in accumulator contents does NOT happen in a Marchant. To represent the changing accumulator contents as above is an artifice.

Fairly sure Marchants do not do "short-cut division". (Sorry if it's too late to explain what that is! too concisely, it's a process of comparing the more-significant digits of the remaining dividend in the accumulator with the divisor, and figuring out that it might be quicker to shift and >add< instead of mindlessly subtracting. (Works great on a Curta. I'd be hard-pressed to give numerical examples, but I suspect that any quotient digit of 5 or greater is more quickly derived by >adding< (with proper position of the carriage) than by ordinary subtraction. Do any fanatical enthusiasts (I'm only about 40% fanatical) know if any mech. calcs. ever did short-cut division?

Compared to another make of machine a few people know I'm associated with historically, the control mechanism (lots of links, latch-like things, many shaft-driven cams (some with memorable shapes) and yet others that perhaps defy classification) of Marchants seems quite complicated. After all, to control 5 clutches properly is just a bit of a do. Even one push of the Add bar sets off an internal sequence of remarkable sophistication and complexity, which I'm only part way through figuring out.

Egad, it's 'way past Bed Time. Sorry for any remaining typos.

Smooooth-sounding language, that... Seriously, forgot to say that if my musings are of interest, they're all public domain. Just be forewarned that the Marchant-specific part is not especially reliable! Give me a few weeks...

|* Nicholas Bodley *|* Electronic Technician {*} Autodidact & Polymath |* Waltham, Mass. |* Amateur musician *|