My explanation turned out to have a gaping hole in it. First, background:
-- A wire carrying current has a "self inductance", which is fixed per unit length. -- The self inductance of a wire is larger for a ferromagnetic wire than a nonmagnetic wire, so the effects described here are larger for ferromagnetic bearings than for stainless steel (nonmagnetic) bearings. That sounds good. Now the bad part: Imagine a ring, connected to a current source, with a sliding contact (i.e., a ball bearing) running around the inside of the ring, going counter clockwise (the "positive" direction, in common usage). As I observed, as the contact moves, the current on the ring "behind" it must decrease, while the current on the ring "in front of it" must increase. Current "resists" decreasing or increasing, so we'd expect the current change to "lag behind" as the contact moves. That would tend to make the current "follow" the contact. However, there's a second effect, which I neglected: As the contact moves past a particular point on the ring, the current in that (tiny) portion of the ring REVERSES. Of course, current in a wire resists changing and that includes short wires, too. This effect will tend to force the current "behind" the wire to decrease more rapidly than it might otherwise do. That's not so good -- it's acting against the motor. An analysis of the two effects seems to indicate that they exactly cancel. We'll now show the analyses of those effects which led to that conclusion. * * * * We'll start with the back EMF at a sliding contact. Imagine a wire, fixed at one end, with a sliding contact. Assume there is a constant current, "I", through the wire, running from the fixed terminal to the sliding contact. It looks something like this: Fixed Terminal O------------------------------- ^ | Sliding Contact Moving at velocity "v" As the contact slides along the wire at velocity "v", there will be a voltage induced between the fixed terminal and the contact. Call that voltage V. Then, if the self-inductance of the wire per unit length is "\rho", we'll have: V = - \rho * v * I as I show, far below, in footnote [1]. Now if we bend the wire into a ring, the current, I, divides into a portion going down one side and a portion going down the other side. The *total change* in current at a particular point as the sliding contact (i.e., ball bearing) passes that point on the wire will still be I, however, so the voltage induced by the movement of the contact around the ring will still be V = \rho * v * I. If we have "k" contacts, instead of just 1, then the current going through any particular contact will be I/k rather than I. The voltage induced by each one, being a purely local affair, will therefore be \rho*v*I/k. But since there are k of them, the total induced voltage will be V(total) = - k * \rho * v * (I/k) = - \rho * v * I and in fact it doesn't make any difference how many balls there are in the bearing. The "induced EMF" will be -\rho*v*I. Now, that voltage points BACKWARDS around the ring, versus the direction the balls are rotating, and in fact that's the WRONG DIRECTION if we want to induce a useful B field. Furthermore, the voltage change when we go all the way around the ring must sum to zero. So, what's providing the "balancing" voltage? If it were just resistance of the wire, then there'd be a current going wrong-way round the ring caused by this voltage, and the motor would stop dead. But there's a second induced voltage here. Let's assume we have one contact, rotating around the ring in the positive direction. Let the current going from the fixed terminal to the sliding contact in the positive direction be I1, and the current going from the sliding contact to the fixed terminal be I2. Current I1 will be decreasing as the contact moves, and current I2 will be increasing, and that, with the inductance of the ring material, will induce a voltage on the two segments, which will clearly be in the positive direction. As I'll show below, in footnote [2], that voltage is equal to W = + \rho * v * I and it exactly balances the "contact voltage" and we are back at square 1. *************************************** Footnote [1] *************************************** Here's my rather sloppy derivation of the induced contact voltage V, which comes from macroscopic changes in current in microscopic pieces of a wire as the contact slides past: Let l = length of wire, from the terminal to the sliding contact point (that's "ell" not "one") Let L = *total* inductance of wire = \rho * l Let V = voltage induced by the sliding of the contact Let i = current through a tiny piece of the wire past which the contact moves Then we have: dL = \rho dl { tiny inductance of the "added" tiny piece of wire when the contact moves dl units } di = change in current in "dl" as contact passes = I { Note that the "infinitesimal" change di, which is the change in current in the "newly added" piece of wire "dl", is actually the fixed macroscopic value "I". } dl = v * dt { Change in length is velocity of the contact, times the time which elapses } And then: V = - dL * di/dt // inductance * rate of change in current = - \rho * dl * I/dt // di -> I, dL->\rho * dl = - \rho * v * dt * I/dt // velocity = dl/dt = - \rho * v * I // and we cancel the dt's and so the voltage induced by the sliding of the contact is V = - \rho * v * I I played kind of fast and loose with the infinitesimals there, though, so I'm not sure that's all 100% kosher. Formula looks reasonable, however, and it appears to match what we get from conservation of energy considerations, given that the power here is going into the B field around the wire. ****************************************** Footnote [2] ****************************************** Here's my derivation of the portion of the EMF which comes from microscopic changes to the current in macroscopic parts of the ring. Make the radius of the ring 1, so we can ignore it. Let: omega = resistance of the ring per unit length rho = inductance per unit length I = total current from terminal to sliding contact I1 = current going from the terminal to the contact in the positive direction I2 = current going from the sliding contact to the terminal, *also* in the positive direction. Note that if the ring voltage is positive, then I2 < 0 < I1. theta = angle at which the contact is currently located Note that if the contact is moving in the positive direction, then theta is increasing (dtheta/dt > 0) R1 = resistance of the ring from the terminal around to the sliding contact, going positively R2 = resistance of the ring from the sliding contact to the terminal, going positively again L1 = inductance of the segment of the ring between the terminal and the sliding contact, going positively L2 = inductance of the segment of the ring from the contact to the terminal, going positively Then we have a bunch of stuff: 1) I = I1 - I2 R1 = theta * omega R2 = (2 pi - theta) * omega L1 = theta * rho L2 = (2 pi - theta) * rho I1*R1 = -I2*R2 2) I1 = (1 - 2pi/theta)*I2 3) I2 = theta/(theta - 2pi) * I1 Plugging (3) in to (1) we get I1 = (1 - theta/2pi)*I and plugging (2) into (1) we get I2 = - theta/2pi * I and if we subtract I2 from I1 we get back I which is nice, means we may not be totally lost. Taking derivatives, dI1/dt = - 1/2pi * I * dtheta/dt dI2/dt = - 1/2pi * I * dtheta/dt Note that they're equal, and both decreasing, but I1 starts large and goes to zero, while I2 starts at zero and goes down, as the contact goes around the ring. Now the induced voltages are as follows, where I'm using W instead of V because I used V up above already: W1 = - L1 * dI1/dt = rho * theta * 1/2pi * I * dtheta/dt W2 = - L2 * dI2/dt = rho * (2pi - theta) * 1/2pi * I * dtheta/dt and the total induced voltage by this mechanism is: W = W1 + W2 = rho/2pi * I * dtheta/dt * [2pi - theta + theta] = rho * I * dtheta/dt = rho * v * I where v=dtheta/dt, with our assumption that the radius of the ring is one. (Of course the ring radius doesn't affect the conclusions; it depends only on our choice of units.) If we have multiple sliding contacts, I'm not certain but I think the currents from them "superpose" linearly and the induced voltage is the sum of the induced voltages. Since with k contacts the current to each is divided by k, the result is that, once again, the number of balls in the bearing doesn't matter; the induced voltage is the same. **************************************************************** ****************************************************************