Everything was either utterly simple or completely change the world forever, yet I can't find anything he's actually done. A few publications for circuits in the 1970s and a book or two of basically republished data books and he's been coasting on that for decades?
And what's his obsession with PostScript?
He has a curious way of being very binary, yes. And punctuating heavily. With periods. Not sure if he fancies himself in a Shatner style. Or what.
I remember reading, for example, his articles on "magic sinewaves" back in Electronics Now. It sounds fantastic (remember -- his own advice, even -- about "too good to be true"?), but he never seems to discuss the practical challenges you'd face, using them in a real implementation.
For example: offhand, it seems to me, you're building a waveform basically as PWM, but with the edges lined up just so, to control harmonics carefully. The solutions are simple enough, for a given setup (amplitude and frequency), and for relatively few edges. The solutions get progressively worse as number of edges rises -- you're solving for the roots of an order-N polynomial, a problem well known to be unstable (poorly conditioned), and having unfavorable complexity (i.e., as N rises, how long does the computation take?). Moreover, in any real system, you will need three things: 1. many switching edges, so you can keep the filter to a practical size; 2. amplitude control, so you can regulate the output; 3. quantized edges (i.e., clock synchronized), so that the sequence can be generated from a counter, not requiring complicated hardware (clock multipliers, or vernier, or variable analog delays, say).
So, now that others have picked up the idea and actually done the hard work, there are real results available on the subject. Example, plots of edge timing versus amplitude, for a given number of pulses during the quarter cycle. It's a nonlinear trajectory, kind of obviously because of the trigonometric polynomial being solved, but less obviously in that you would naively expect the pulses to grow outward from their centers, and that's about that. Well, as it happens, that's the small-signal equivalent, and it's close enough for large signals (that is, near 100% amplitude) to serve as a starting point for refinements. Since it's just a polynomial, we can iterate Newton's root finding method to get arbitrarily close to the local minima, which because the starting point was close enough, it's very likely to be the true minima.
That makes real time computation of the level (as a function of amplitude, tweaks in frequency, and quantization) practical, on modestly powerful MCUs, or FPGAs, for audio frequency content.
Alternately, you can forget about it entirely, which is mostly what's happened. You don't gain much by timing the edges impossibly tightly, when you have, say, a thousand edges per cycle. For one, that's a huge pain to solve for (and with so many parameters to solve for, it's also that much more likely to go numerically unstable). For a 60Hz inverter, that's 30kHz, an entirely boring PWM frequency. If the motivation is reduction of waste power, then one needs at least a modest switching frequency to begin with, because filter inductor Q generally goes as sqrt(f). This combination already makes it at least intractable, if not outright impossible. Further, if we're optimizing for size, we need to push the switching frequency up that much higher still, and then we might look to higher performance technologies like GaN power transistors to address the switching losses.
His predictions and arguments on other subjects are also here and there, in line with what you should expect for such matters.
Hydrogen fuel cells still aren't going anywhere, even with adsorption storage to try to drive down its astronomically poor specific energy density. Downside being the extreme weight and cost gain in the process. Possibly a catalyst can be developed which is relatively cheap and allows rapid, efficient and reversible bonding of hydrogen to, say, a hydrocarbon framework. Without the framework reverting to, say, graphite while it's empty.
We are, however, getting closer to solar fuels. Chemical processes are known which can convert CO2 into small alcohols and such (which can be used directly in engines), they're just awful end-to-end efficiency. Some use electrical power (paired with a PV array, you get solar fuel as such), some with traditional chemical (heat driven) methods, some with direct photolytic action.
I think he followed solar too, mainly complaining that it wasn't anywhere close to break-even. I haven't checked if his opinion has changed on that in recent years. I think also as time went on, he changed his tune from economic breakeven (which is relatively rapid with today's installs -- a few years) to thermodynamic breakeven (i.e., you don't see solar panel factories with solar panels on the top and a negative overall power consumption to the grid). He is probably still correct about that, but we'll see if that continues to change at the same rate as the first instance...
Tim