Yes, it's absolutely time to decide whether MNT can work. At this point, we don't need many breakthroughs, mainly a lot of engineering. The objections to the theory are a lot weaker than the work in support of it. "If there's no math, it's probably a myth"--and *all* the math so far is saying that MNT should work.

In the past decade, MNT has gone from a far-distant projection of what-should-be-possible to a set of concrete proposals for a straightforward manuafacturing system based on a specific and easily analyzed method of chemistry. So let's see some real attention--including criticism--to the proposals.

Assuming there's no mysterious showstopper that no one has discovered yet, how long will it take to develop? Depends largely on policy decisions and funding model. The cost and difficulty will decrease rapidly. I think by 2010 it'll be obvious that MNT can work, and it'll cost under $1 billion and five years to develop. At that point, if we can't fund it here, it'll happen somewhere else.

But how can we fund it, when the scientific establishment is currently dead-set against it? We need to improve the standards of discussion. Demand to see the math. Don't assume that credentials imply credibility. Realize that the current refusal to look at MNT is politics, not science.

And more importantly, how can we prepare for it if we don't believe it can happen? The final stages of development could happen very quickly--there won't be time to make sensible policy at the last minute. But without good policy, things could get very nasty in several different ways. We're currently betting our future that Smalley, Roco, and a few other prominent MNT deniers are right; this is not smart.

Chris Phoenix
Director of Research
Center for Responsible Nanotechnology
http://CRNano.org

The proponents of MNT have studiously ignored the problems of sensing at the nanoscale.

In contrast, those who have actually attempted nanoscale work have found that nanoscale sensing with amplification to the macro scale is a ball-busting monster, a heartless bastard that will break your spirit and leave you swearing impotently at your inability to see what you need to see.

Any serious attempt to develop MNT requires the development and use of volitional sensory feedback loops operating entirely at the nanoscale. But those feedback loops do not, and in principle cannot, exist.

We are only at the beginning of understanding the multitude of feedback loops inherent to systems biology. But we do understand that these feedback loops are non-volitional mechanisms, the result of evolutionary development (which is itself non-volitional) over billions of years. There is no way, even in principle, to greatly speed this process of development. And without a starting point of MNT self-reproduction, there is no way to even begin.

I do not expect the MNT crowd to grasp this line of reasoning. They have self-selected themselves from a group unable to comprehend such arguments.

In contrast I expect the molecular biology crowd, and the systems biology crowd, to understand this line of reasoning and extend it. But that can only happen if their attention can be called away from real work to consider the MNT fantasy in any detail. So far that hasn't happened.

Edited by author 11-12-2003 08:36 AM
I think the problem with the discussion of the feasibility of molecular nanotechnology is that there's a confusion between ends and means. The end in question is the creation of nanoscale machines capable of doing useful things, particularly assembling other nanoscale objects. We know without doubt that this goal (one could call it radical nanotechnology) is achievable, because, as Drexler pointed out, biology offers an existence proof.

But there is likely to be more than one means by which this end can be reached, and the proposed route based on carbon mechanochemistry that is associated with the vision of MNT is only one of them. Other routes to a radical nanotechnology would include the isolation and reassembly of biological machines (such as molecular motors) in artificial configurations - this is the program of bionanotechnology. Another route would involve exploiting the same principles as cell biology - the use of self-assembly, molecular recognition, molecular shape changes and Brownian motion - but using synthetic materials.

The question, then, is not whether MNT is possible, but whether the specific MNT vision represents the best or most practical route to achieving a radical nanotechnology. My own view is that it does not. The difficulty is that the MNT vision considers the inescapable features of the nanoworld - strong surface forces, constant Brownian motion, what is, in the presence of water, a viscosity dominated environment, and a general lack of stiffness in the structures one has to work with - as problems to be engineered around. Biology, on the other hand, has evolved mechanisms that don't merely overcome these problems, they actively exploit them.

I think the MNT vision is too coloured by our experience of engineering objects on the macroscale. As we understand more and more about how cell biology works, we'll understand how different the nanoworld is to the macroworld, and that biology offers a model for a much better way of engineering things on the nanoscale.

Richard, by restating the problem as you have I think you vitiate both the Drexler/Merkle/Freitas MNT proposition and the skepticism about its possibility.

You blessed peacemaker, you.

I agree strongly with your final paragraph.

I also think it's important to separate fact from fiction, so that we don't waste time, effort, and money pursuing unrealizable fantasies.

Attabuoy, it is simply not true that we have ignored the problems of sensing at the nanoscale. Nanosystems considered a few types of sensors, and Nanomedicine Chapter 4 (scroll down for TOC) analyzed the physics of 64 types of nanosensors in eight categories, including chemotactic sensor pads, displacement and force sensors, and acoustic sensors.

I'm not sure what you mean by "volitional" sensing--surely not "intelligent." Perhaps "deliberate" in the sense of engineered and causal, usable for controlling machinery.

But this is not hard; there's nothing magical about it. The simplest sensor is a mechanical interlock: A can't happen unless B is in the right place. But these exist in biology, for example in the receptors in neurons which can be blocked or opened by specific chemicals.

And a mechanical interlock is all you need to implement digital logic. Note that digital logic and mechanical interlocks, like many physical and chemical processes at the nanoscale, can be reversible, and that reversible systems can be allowed to proceed down incorrect paths as long as those paths don't lead to an irreversible state transition.

Please give a citation or a brief physics explanation for why volitional sensory feedback loops cannot exist at the nanoscale. The rest of your post rests on this claim, and this claim is unsupported and questionable: it is not a "line of reasoning" that I am unable to grasp, but a simple assertion that contradicts what I know of physics.

Richard, you list "inescapable features" of "the" nanoworld. But they are not in fact inescapable. For example, Brownian motion can be escaped simply by fastening everything down. The stiffness of molecules can vary by orders of magnitude. You even mention water, which is very easy to escape.

Your claim that we only try to avoid these features is also wrong. Surface forces can be very useful, for example in bearings; nanomechanical designs work with it, not against it. Thermal noise can also be useful; Nanosystems exploits it in at least two places, to overcome static stiction and to unjam sorting rotors.

It is interesting to explore which machine paradigms can be extended to the nanoscale and which cannot. We know some things already. Electromagnets cannot. Levers can. Electrostatic motors can. Digital logic can. In the long run, is this more efficient than biology? Time will tell. But it's worth noting the things available to dry nanomachines that are simply not available to biology. Diamondoid chemistry. Petawatt per cubic meter power density. Fast digital logic. These more than make up for things in biology that are not available in dry nanosystems, such as solution chemistry, and... um... what else?

Most of the underlying tricks used by biology can also be done with machines. We're giving up very little by switching to a dry environment. Now, to build a biological system, it may be better to use wet chemistry. But MNT is not about biological systems. It's about solving engineering-type problems--and for this, an engineering approach is probably best. Biology is as much out of place in building a digital computer as MNT would be if forced to work in your biocentric world (which is a very limited subset of the possible nanoworlds).

Internal combustion engines and digital computers will never be matched by biology. We're now proposing a way to build nanoscale structures that biology can't match. You can't get diamondoid nanodevices without MNT. The question is, how valuable will such things be? The answer is, for many purposes, irreplaceable, and for most applications, at least competitive. And that's not even getting into the advantages of general-purpose manufacturing that are far more accessible with MNT than with biology.

What I'm arguing is that MNT can easily produce radical products that biology simply can't. When you get to extremes like terapascal materials, the two are just not comparable. This extra degree of radicality lets you play tricks that biology can't play--and still use most of biology's tricks. And a technology based on these extreme capabilities will probably out-compete a bio-based technology for most applications.

Chris

Edited by author 11-13-2003 12:27 PM
Chris, I think you overestimate the ease with which these features of the nanoworld can be overcome. Most trivially, of course you can get rid of water if you machines work in ultra-high vacuum. But I think everyone agrees that some of the most important applications of nanotechnology will be in medicine, in which it is indeed difficult to get away from water. Even just in the open air, in a humid summer day in NYC or Shanghai most surfaces will be covered with a monolayer at least of adsorbed water molecules.

The issue of Brownian motion is more fundamental. I'm not just talking about the translational motion of particles, which of course you can stop by nailing the particle down, as it were. It's all the internal motions of a structure in response to its bombardment by surrounding molecules. The law of equipartition of energy can't be broken so easily. Let's be concrete for a moment. Suppose you make an axle from a nanotube, say 10 nm between its supports. At room temperature, knowing the bending modulus of a nanotube, equipartition says the root mean squared random flexing at the centre is about 1 nm, about 10% of the diameter of the tube. So you've got an effective tolerance of about 10%. Now if you go to a machine shop and ask them to make a similar macroscopic piece, a good machinist will get you a tolerance of 1/10 of a thou, which represents a tolerance of about 0.001%. Interestingly, this degree of precision in machining was achieved in the middle of the 19th century, and it was basically this that permitted mass production. This tolerance problem means that the engineering you'll be scaling down to the nano-level isn't the marvelous precision of watch mechanisms, but something more like a medieval water-mill.

OK, you can quibble about my assumptions, you can make designs which squeeze a bit more effective rigidity out of your materials, and I know Drexler and Merkle and others have been very ingenious in doing this. But, then, ask yourself the question why in MNT you have to work so hard to get round these problems when biology just takes floppy bits of jelly and it all works with no problems. The reason is that biology depends on Brownian motion to make it work - conformational change in macromolecules driven by Brownian motion is a fundamental principle underlying most biological nanomachines, from ion pumps to molecular motors and beyond.

I think the mindset that one needs to get rid of is the one that says, "if biology is so successful despite using such unpromising materials, then how much more successful MNT will be when we can use better ones, like diamandoid". Instead, one needs to understand that biology is so successful at the nanoscale precisely because of the apparently unpromising materials it uses. These materials and operating principles are optimised for the nanoworld.

You yourself illustrate my point exactly. "Internal combustion engines and digital computers will never be matched by biology". No, they won't, and with good reason. The way in which thermal diffusion and mechanical rigidity scale with size mean that you will find it very hard to run any kind of heat engine on the nanoscale; the temperature difference between hot source and cold source will diffuse away before you can extract any work from it. That's why all biological nanomotors are isothermal - they exploit, not temperature differences, but chemical potential differences. And in fact in terms of power density they aren't that far off petrol engines. As for digital computing, well, I'm not sure - i'd like to see a rigorous comparison of computing power between a nanoscale digital computer and the magnificently subtle analogue chemical computing systems that are involved in cell signalling.

But I think we won't truly know how best to do radical nanotechnology until we seriously (and by this I mean experimentally) try it. Let the best research program win!

First, let me say I completely agree with your final statement, "Let the best research program win!" To which I would only add that this excellent idea cannot be tried without at least rudimentary exploration and funding of MNT. It would not cost much to do a little bit of mechanochemistry simulation, a little bit of molecular mechanics and molecular dynamics to test nanoscale diamondoid, just one point-by-point peer review of Nanosystems. Why is this not being done? Do you agree that it should be?

Let me also say that I'm not trying to denigrate biology here. There's no doubt biology is flexible and impressive in many ways. But there's also no doubt that MNT can do things biology can't. If it can work at all, it's probably worth doing. All the evidence so far says that it can, in fact, work. I'm aware of no argument against the possibility of MNT that can withstand even a basic application of scientific theory. And the arguments against its efficiency never include math, and so don't worry me, since the math that's been done says that MNT should be extremely efficient.

About water: I thought you were talking about viscosity relative to the internal workings. It's not hard to keep water out of the innards of an immersed MNT-built device. And swimming will usually be on the micro, not the nano scale, and can use structures and motions very similar to those of bacteria. Nanomedicine has a whole section on swimming.

It sounds like instead of Brownian motion, you mean thermal noise. Yes, this is a concern. Just as corrosion and wear are concerns in almost any metal machine part. And yes, you're right, we do try to design devices that work in spite of it. (Though as I noted earlier, we're quite willing to build thermally flexible systems when it suits us.)

The advantage of fighting thermal noise is that it makes it possible to build 2D and 3D carbon polymers (diamondoid). I think high vacuum is another requirement; there may be a chemistry that can build diamondoid underwater, but I doubt it, and life has never discovered it--and life took "an evolutionary instant" to invent magnetite shoes for snails. Being restricted to 1D polymers, life had no reason to develop stiff-arm chemistry. But diamondoid is just too good to pass up.

You criticize this thinking: that diamondoid is obviously better than biomaterials. First, can we agree that it's obviously better in a lot of products? Airplanes can be made of wood, but not spaceships. Cars don't include any natural materials anymore, except as a luxury. Material properties matter a lot in products--and diamondoid is the best in a lot of ways. Especially nano-structured diamondoid.

Second, I'll disagree with your statement that "[biological] materials and operating principles are optimised for the nanoworld." They may be optimized for doing chemistry underwater. But they are certainly not optimized for doing chemistry in a vacuum. And vacuum is just as much a nanoworld as water is. That's what a lot of MNT criticism doesn't grasp. It starts from a water-centric view and says, "MNT doesn't work well here, therefore it doesn't work." Even Smalley, with his statement that ordinary chemical reactions require five to fifteen atoms in a cubic nanometer--that's simply not true! Except underwater. MNT is not wet chemistry. It is not subtractive engineering. And it can't be compared to either of those. Learn it on its own terms--vacuum workspace, diamond strength and stiffness, programmable mechanical chemistry--and you'll find that it is very powerful in its domain. *Then* you can compare the power of MNT with the power of biology, relative to what you want to build with it. Biology is far from the clear winner for most manufactured products.

I didn't mean that biology will never build an internal combustion engine; that's obvious. I meant that you will never fit 400 horsepower of biology under the hood of a car. Even if the biomotors themselves have that power density, they require too much support structure. MNT probably won't use either bio or IC; for many applications, it can convert electricity at a power density of petawatts per cubic meter, or store power in mechanical springs with energy densities near chemical fuel. Where it does use chemistry, early versions may be similar to fuel cells, and later versions may extract power directly from mechanochemical operations--isothermally.

As for computers, yes, neurons are very impressive, and may even do analog computation more efficiently than digital computers can. But neurons are not general-purpose; it'd be very hard, and very wasteful, to design a neural net that implements a standard programming language. If you want to write in C, you simply have to use digital circuitry. Likewise, it's still quite hard to design a protein to fold the way we want it--much less predict the folding of a natural protein. If you have an application where molecular shape is important, designing it in diamondoid will be much easier than designing it in protein. (This advantage is in addition to the material properties--which as I said are not directly a basis for selecting the manufacturing technology, but are a basis for selecting the product space.) You may say that molecular shape doesn't have to *be* important, or that it can be evolved as our immune system evolves antibodies. Well, when that turns into a technology, we can compare it with MNT. So far it's not even art--it's magic, though the rational drug designers are learning to make it an art.

One more comment, on tolerance. With the new rules of MNT, tolerance has two definitions. There's the manufacturing tolerance, which is effectively infinite: the atoms go where you want them, and stay there. Then there's the operational tolerance, which as you note is problematic for some designs. But this is not an indictment of MNT. Every technology has its quirks. The question is, given that MNT could do things that no other technology can, is it worth working past the quirks, which after all are relatively minor compared with the quirks of other technologies? With the information available now, it seems obvious to me that the answer is probably Yes, it's worth learning to design bulky parts to overcome thermal noise.

Chris

I just thought of a possibly useful analogy for the difference between MNT and wet/biological chemistry. Imagine a medieval marketplace, with everyone jostling and shouting and business being transacted almost at random but with different people having affinities for different products that are scattered all over the town square. This is wet chemistry, where the babble of words and barter are the chemical interactions and reactions. Now imagine a monastery with an array of monks sitting silently copying texts. This is much more programmed, less interactive, less chaotic. And many fewer words are produced--but much more accurately, and in more enduring form.

The marketplace and the monastery both involve people interacting. They have their (very different) functions and (very different) rules. Both are productive. Both pass on information via words. But only one can produce books.

To continue the analogy: it's true that the marketplace makes use of powerful and elegant principles that have no place in the monastery. And that there's a lot of overhead in writing out a book by hand. But this does not mean that monasteries were unproductive, or not worth building.

Chris

Chris, I'm grateful to you for taking my comments seriously. Let me first restate two things, lest they get lost in the detail. Firstly, I am not arguing that what I call radical nanotechnology is not possible, just that there may be more than route to it, and some routes have probably not even been thought of yet. Secondly, I'm not arguing that the MNT project is impossible in principle or contrary to some basic laws of nature; what I am suggesting is that it may not necessarily be the most efficient or practical approach for making nanodevices that will make our lives better.

Now to the argument about whether biomaterials or synthetic materials are superior, because this goes to the heart of the matter. I agree absolutely that people don't build many aeroplanes out of wood any more, and that the disadvantages of biomaterials are all too obvious to those of us whose knees hurt a bit more than we would like after we've been running. But the point is that biomaterials were evolved to work at the nanoscale, and all kinds of awkward kluges are needed to make structures from them that work at the macroscale. You misunderstood me about chemical computing, and your misunderstanding is quite instructive. I wasn't talking about neurons - I meant the cell signalling systems that even bacteria use, the sort of thing that E.Coli uses to be able to swim towards food sources. It is the basic toolbox of nanoscale chemical computing that bacteria evolved that has been put together to make our macro-scale nervous systems. The tools work spectacularly well on the nanoscale, but it probably isn't the place you would start from to make a control system for a 2m robot if you were doing it from scratch.

As to my water-centred view, I plead guilty. I'm interested in making things that do useful things in the world we live in, and that world is very definitely watery. I'm quite prepared to accept that MNT would work very well at 3 K in outer space. That's just somewhere I don't want to live myself.

It's getting late on my side of the atlantic and I want to get to my local before closing time, so I may come back to comment on science politics and the funding of MNT tomorrow.

I quite like your marketplace analogy, because it emphasises the complexity that can emerge from many autonomous units interacting by simple rules. But we know how much more efficient a marketplace is at directing economic activity than the most sophisticated central planning.

There is no doubt that "wet" nanotechnology, a synthetic version of biology, is possible. I am yet unconvinced that "dry" nanotech is possible. I am aware the covalent mechanochemistry has been done. However, its not clear to me from the papers I read if this was done at room-temperature or a cryogenic environment. It seems to me that since we live in an aqueous environment with moderate temperature and pressure conditions, the most useful nanotechnology would be one designed to operate in this environment. Biology certainly does. "Dry" nanotech is not clear to me.

Sensing and feedback loops on the nanoscale is one problem. Scaling up seems to me another. Besides, biology does make materials harder than wood. Seashells and coral are certainly hard structures and spider silk has high tensile strength. I think that many structural materials (for houses to sky-rises) could be made using "wet" or bio-memetic nanotechnology.

The proponents (and critics) of "dry" nanotechnology need to present a detailed technical argument why it is possible.
Such a presentation would be very useful for those of us trying to decide what we want to do or place our money over the next 20 years.

Regardless of what kind of nanotech proves to be possible, we want to be sure that those of us with an immortalist, tranhumanist bent get our hands on it first. Thats why answering the question as soon as possible is very critical.

1) Contact info
2) Reply to Kurt and Richard about environments and efficiency
3) Reply to Kurt about detailed technical arguments and feasibility

1) In case anyone reading this wants to email me, my email address is cphoenix at CRNano.org. I'm Director of Research at the Center for Responsible Nanotechnology. (A major part of our mission is education about nanotech, and a major focus is dry molecular nanotech since that's where we believe the biggest changes will come from.)

2) Humans mostly live near sea level. But the most useful jet airliners are designed to operate around 30,000 feet, where the air is thin and cold. It's worth doing that for the increased efficiency. And keep in mind, both of you, the difference between a manufacturing system and its products. Calculations show that it shouldn't be too hard to build nanoscale diamondoid machines stiff enough to do diamondoid mechanochemistry at room temperature. But even if "dry" nanofactories have to be run in a vibration-controlled, liquid-nitrogen-cooled environment, their products can still work just fine in the messy real world. Built with atomic precision and diamond stiffness, it's easy to keep water on one side of a seal and vacuum on the other. (Richard, I think this paragraph answers your comment about 3K and outer space?) And most mechanical interactions with the environment will be at micro or macroscale, where things are less weird and even organisms get mechanical.

Kurt mentions that biology can make materials harder than wood. True, but less relevant in applications where it's important to have maximum material performance--like aerospace. Or any system where you want to save mass (which includes any application where transport costs for feedstock or product are significant). AFAIK, biology can't touch buckytubes. So yes, houses can be made of biomaterials. High-rises... depends how high. There are some amazing building proposals floating around these days.

Richard likes my marketplace analogy "because it emphasises the complexity that can emerge from many autonomous units interacting by simple rules. But we know how much more efficient a marketplace is at directing economic activity than the most sophisticated central planning." This is one of the reasons I thought the marketplace was a good analogy: free trade and Brownian motion both have useful emergent properties. But note your qualifier: *economic* activity. By contrast, armies are run hierarchically. I don't think you'd argue that an army should be run by unguided people satisfying their selfish desires; they'd be cut down by the first drilled troops that came along. So, it depends what you want to do. If you want to do machinelike stuff, it may be best to use machines all the way down; otherwise you'll have a very awkward and probably inefficient interface at some point, in addition to having to accommodate two radically different design paradigms.

Richard said: "Firstly, I am not arguing that what I call radical nanotechnology is not possible, just that there may be more than route to it, and some routes have probably not even been thought of yet." This is true. But one reason to talk about radical nanotech is to ask what it can do. I'll go out on a limb and claim that the *lower* bound of what dry MNT can do is probably higher than the *upper* bound of what bionanotech can do, in a wide variety of contexts and applications. And for general-purpose manufacturing (which is where the really exciting stuff starts to happen), MNT will probably be developed a lot sooner--one decade, rather than (I'm guessing) three.

3) To answer your question, Kurt, as far as I know, the mechanochemistry that's been done has all been cryogenic. But that does not constitute evidence that room-temperature mechanochemistry can't work, and anyway it doesn't matter much; the products will be more than valuable enough to be worth cooling the factory if necessary.

You mention scaling without saying what you think is hard about it. I don't know that anything is. Making lots of small things work in parallel: not hard, if they all have simple interfaces--and they can, since every aspect can be explicitly designed. Fault recovery: not a big deal in a mechanical paradigm, as long as you can detect the fault, and that's one of the things the simple interface is for. Mechanical fastening: not hard. Changing behavior at various sizes: not hard--that's what levels of abstraction are for. (Another benefit of simple interfaces.)

You said: "The proponents (and critics) of "dry" nanotechnology need to present a detailed technical argument why it is possible. Such a presentation would be very useful for those of us trying to decide what we want to do or place our money over the next 20 years."

Well, that's exactly what Nanosystems is! Granted, Nanosystems is *too* detailed for most readers; it's not accessible. Part of my job (educating about nanotech) is to translate the Nanosystems work, and the subsequent decades of papers, into an accessible format.

Chris

To Richard, re: suitability of method vs. scale

If I understand you right, your comment about biomaterials being evolved to work at the nanoscale does indeed go to the heart of the matter. I think you're saying that mechanical systems work well at the macroscale, and biomaterials work well at the nanoscale, and they don't work well in each other's domain. If this is what you meant, it's indeed a deep point. And there are two ways to interpret the facts.

You look at macroscale organisms, and see kludges, such as (am I right?) wrapping long-distance neurons in myelin. And you look at proposed nanoscale machines, and see a need for kludges, such as making axles incredibly thick to withstand thermal noise. And nanoscale biochemistry works "spectacularly well." And macroscale machines obviously work well. So your conclusion is reasonable.

I look at macroscale organisms, and I see machines. I'm not being reductionist--the machines are implemented with all sorts of subtleties at the lower levels; they emerge; but what emerges is machines: levers, digital signaling (note that short-distance neurons can signal without spiking, but I don't think long-distance neurons can), lenses, pumps and pipes. But when I look at the nanoscale, I see kludgy machines. A ribosome or enzyme is very bumpy. Is this bump here to put a steric constraint here, or to extract 1/10 of an electron charge from that region over there? Or just because that's where the beta sheet ended, and the protein chain has to do an awkward turn to reach the next one? Or two functions at once?

You may see elegance in the fact that this kind of thing works. I don't. It's incredibly ad-hoc at every level. Naturally--it evolved. It's like a dancing bear: it's impressive that it can dance at all. Enough evolution, with enough degrees of freedom, and you'll eventually find something that works. And how does it work? Look at ATP synthase: a protein that physically rotates in a socket to allow a proton to pass, bending another protein like a camshaft, to push two molecules together. It's a machine! And yes, it's implemented with all sorts of subtleties and mysterious bumps on the molecules. But have you seen Drexler's neon pump? It's basically the same thing--but without the subtleties. Now look at the microtubules in a cell: unmistakably a truss. These are not analogies; I'm not saying that DNA transcription is like a paper-tape reader. I'm saying that ATP synthase *is* a pump.

(And by the way, macro-scale machines can get pretty kludgy too. Look at an old grain-windmill: Huge gears built in pieces with each tooth separately whittled. Hardwood shafts a foot thick and bound in iron. Bearings that require expert care on a frequent basis. And all to transmit a few horsepower a few yards! In this case, at least, the kludgyness is required by poor materials. There may be a lesson here. Could better materials reduce the kludgyness at the nanoscale as well?)

Machines are not alien to the nanoscale. Biology gravitates toward the machine paradigm as soon as it can, at all scales right down to the proton pumps. I'll advance a radical theory: There are several clear reasons why biology uses the chemicals it does, and none of those reasons applies to dry MNT, and they do not include the unsuitability of machines. And a radical conclusion: There is no reason why manufactured systems shouldn't use different and more efficient materials--at all scales right down to the atoms.

The reasons biology uses such sloppy chemicals:
A) The chemistry has to work underwater. This requires linear polymers. (Note that even dendrimers do not appear to have been invented in nature--or is starch an exception?) This also puts a limit on the reactivity of radicals.

B) The system has to be evolvable. Without deliberate engineering, evolution by incremental change is the only way to develop advanced systems like we see today. So you need lots of degrees of freedom and a separate simple specification code.

C) The system has to be able to metabolize itself. Whatever it builds, it'll probably want to break down again, both for conservation and to avoid waste buildup; another reason for linear polymers with simple backbones.

D) (A corrollary of C, and maybe A, and also of the need to compete for limited energy) The construction reactions have to be easily reversible.

So this is why something like diamondoid would be unsuited for biological systems. But that is not at all the same as being unsuited for nanoscale systems! An MNT fabrication system doesn't have to evolve or metabolize. It can use all the radicals it wants and leave them dangling while it builds 3D surfaces of single molecules. And it can afford to throw away multiple kT's per reaction because functionality is far more valuable than energy to humans. (If you doubt this, calculate how many kT's each transistor in your Pentium is throwing away, a billion times every second.)

I can summarize the above in two sentences. The only reason for the subtleties of biochemistry is to compensate for underwater chemistry, and to satisfy the conditions of life. Dry MNT systems can use machines just as cells do, but without water and without life, they can use better materials, and will be more efficient for it.

Chris

Since Kurt asked for a detailed technical argument, here's a five-minute summary of Nanosystems. Every chapter has lots of formulas and math that I won't mention again; but a counterargument without math is very likely to be wrong.

Chapter 1: Introduction and Overview
Molecular manufacturing should be able to build mechanical systems with amazing performance at the nanoscale. This book will explain how this works, building on basic physics and chemistry.

Part 1: Physical Principles
Chapter 2: Classical Magnitudes and Scaling Laws
A lot of nanoscale properties can be predicted directly from physics. Electromagnetism doesn't work. Things get floppy, but not too fast. Cooling small systems is easy. Small things move faster.

Chapter 3: Potential Energy Surfaces
Chemical reactions are more or less predictable. Mechanical properties can be derived from chemical bond properties. Surfaces are sticky and squishy.

Chapter 4: Molecular Dynamics
Atomic systems wiggle. Different configurations/positions have different energies. Large energies (relative to thermal noise) can form barriers between states; between barriers (in potential wells), the system can take any configuration and the probability of each configuration can be calculated.

Chapter 5: Positional Uncertainty
You can make engineering estimates of the positional uncertainty of things like nm-scale rods, springs, and gas-filled pistons exposed to thermal noise.

Chapter 6: Transitions, Errors, and Damage
If you know barrier heights, the probability of crossing between potential wells can be calculated as a function of temperature and time. Placement errors can be calculated. Covalent bonds don't usually break at room temperature in the dark. Radiation damage is a factor for >micron-scale systems.

Chapter 7: Energy Dissipation
There are lots of ways for energy to be thermalized. These are calculable.

Chapter 8: Mechanosynthesis
Mechanosynthesis has many advantages over solution-phase synthesis, and should have as broad a range of products. There are quite a few stiff reactive molecules suitable for vacuum-phase mechanochemistry. Several diamond-forming reactions are proposed.

Part II: Components and Systems
Chapter 9: Nanoscale Structural Components
Even small diamondoid rods and housings can implement useful stiffness and well-defined surface. Shape and size can be controlled with high precision by substituting atoms.

Chapter 10: Mobile Interfaces and Moving Parts
Atomic-scale moving parts are bumpy. But these bumps can be far less than thermal noise, implying zero static friction. Dynamic friction still an issue (Ch. 7). Atoms can make good gear teeth. Cute models including planetary gear. Ratchets.

Chapter 11: Intermediate Subsystems
Measurement devices, harmonic and toroidal drives, fluids, seals, pumps, fractal cooling, electrostatics (10^17 W/m^3 power density).

Chapter 12: Nanomechanical computational systems
Mechanical gates, registers, logic arrays, reversible logic, long-range data transmission. Sub-micron 10^6-interlock 1-GHz 60-nW CPUs. (Lower bound, can be improved.)

Chapter 13: Molecular Sorting, Processing, and Assembly
Sorting rotors to import molecules and purify the input stream; conveyors; binding sites; molecular mills for repeated mechanochemistry (and power generation); conditional encounter mechanisms; a robot arm stiff enough to do mechanochemistry at room temperature with a 100-nm range.

Chapter 14: Molecular Manufacturing systems
Joining intermediate-scale blocks; factory system layout; factory shells and product delivery; redundancy; productivity calculations (make its weight in an hour). Much better than conventional manufacturing on many counts. Design issues: shape-description languages and design compilers.

Part III: Implementation Strategies
Chapter 15: Macromolecular Engineering
Cells implement many mechanisms: struts, bearings, clamps, actuators/motors, etc. Biopolymer design (easier than the folding problem). Solution synthesis for bootstrapping dry-MNT systems. Using SPMs for fabrication and imaging.

Chapter 16: Paths to Molecular Manufacturing
Lots of paths; use backward chaining to find a likely one. Simple actuators and manipulators. Molecule handling. Solution-phase intermediate systems. Ways to reduce development time.

Appendix A: Methodological Issues in Theoretical Applied Science
Even with incomplete information, you can get reliable and useful numbers, and make reliable predictions.

Appendix B: Related Research
Many fields feed into MNT, but little targeted work has been done yet.

Whew!

This was written in 1992. In the last decade, Merkle has done architecture work on small self-contained diamondoid fabricators, and also on hydrocarbon chemistry, and a bit on nanofactory architecture. See Merkle's papers. I've written a very long paper on primitive nanofactory architecture (designable today, except for the fabricator) and rapid bootstrapping. Merkle and Freitas have recently done work (in press) on more detailed fabricator design, and also on mechanochemical simulation.

So the proposals are out there, waiting to be evaluated...

Chris

Chris, you summarise my argument more clearly and succinctly than I managed to do myself. “Mechanical systems work well at the macroscale, and biomaterials work well at the nanoscale, and they don't work well in each other's domain” is exactly what I meant. Whether this is deep, I don’t know, but it’s certainly both important and underappreciated.

Like you, when I compare machines at the macroscale and the nanoscale, I see differences. But I ascribe these differences not to the fact that biological nanomachines are just ineptly executed versions of macromachines, but to the fact that they operate on quite different principles, principles that are so foreign to us with our macro-scale experience that it takes some effort to see them for what they are.

I’m glad you mentioned ATP-synthase, because it’s a lovely example. As you describe it: “a protein that physically rotates in a socket to allow a proton to pass, bending another protein like a camshaft, to push two molecules together. It's a machine!” But it’s a very odd machine, by our standards. This oddness isn’t made clear in the beautifully rendered computer illustrations that we know it from, because they give a false impression of its rigidity. In reality everything in it is constantly jiggling round under the influence of Brownian motion. The shape of each protein molecule is fluctuating alarmingly, but it is these fluctuations that actually drive the operation of the machine. So it is a pump or a turbine in what it does, but the way it does it is very different to the way a macroscopic machine works. The key mechanism relies on Brownian motion and lack of stiffness – it is the sequence of macromolecular shape changes that occur when molecules bind and unbind to the component proteins. Is ATP-synthase an optimum solution to the problem it solves, or has evolution just stumbled on something that somehow miraculously gets by? This really is a deep question, but consider the fact that every living thing today – bacteria, fungi, jellyfish, ourselves – contains ATP-synthase that is almost identical. What that suggests to me is that evolution found a near-perfect solution rather early and several billion years of further evolution hasn’t been able to improve on it.

I do accept your point that evolution has to cope with a number of constraints. What isn’t clear is how many of these constraints are accidental, simply being the results of the random way history unfolded, and how many are necessary, coming from the nature of the physical world evolution operates in. You’ll be aware that this is a question much debated by evolutionary theorists, and I don’t think there’s a consensus. Some specific points, though – nature certainly does use highly branched molecular architectures. As you suggest, the waxy version of starch, amylopectin, is one of these. The glycoproteins that make mucus such a wonderful material supply some others; they look like bottle-brushes, with a central protein core surrounded by polysaccharide bristles. I think in general the importance and complexity of polysaccharide biochemistry is only starting to be appreciated now.

The issue of energy scales is an important one, though. It isn’t so much the fact that biology operates underwater that limits the energy scale, it’s the temperature. Self-assembly processes require energies to be neither too large nor too small compared to kT, so that when components come together they stick, but not so firmly that they can’t try out some other configurations to make sure they end up in the best one. Self-assembly is a fantastically powerful principle for parallel manufacture of nano-objects with atomic precision, and this is the price you pay for it. But there are ways of getting round this, too, by templating and precursor routes – this is how tendons and snail-shells get to be tough.

To move away from the debate about soft vs hard, and address Howard's questions:

Should there be an Apollo style crash program to develop MNT? No, because nanotechnology is at the stage of being science rather than engineering. It is far from clear what the first, let alone the subsequent steps, actually are. In these circumstances a centrally planned effort will inevitably end up expending a vast amount of money going up a blind alley.

Should there be a high-level study of whether MNT is possible? Again, I can’t really see the point. You’d get an opinion, but until you’ve got some concrete, experimental results, the opinion would be just that, an opinion of some very distinguished scientists. Unhappily, such opinions, in the past, have often turned out to be quite wrong.

Should “Nanosystems” be subjected to a careful, line-by-line, peer review? Again, what would this add? The book is in the public domain, lots of copies have been sold, it’s still in print, anyone who is interested can read it. The test isn’t whether you agree with the math or not, it’s whether you can think of a way of physically executing some of the ideas.

What is stopping people taking that first step? Nothing, and indeed people are right now trying out lots of different candidate first steps. Some will work, some won’t – that’s how science works. I think the supporters of the MNT project actually underestimate how much work is going on that could quite legitimately be considered to be part of that first step. Just to give two examples (UK centred because that’s where I live, you would find just as many examples in the USA or Japan or China or Germany), there’s the work of Malcolm Green’s group at Oxford, doing spatially localised chemistry with near-atomic precision by immobilising catalyst molecules on an AFM tip. In Sheffield there’s a group combining piezo-drives with nanometer control with a high resolution TEM so they can see what they are doing as they move groups of atoms from place to place. Other groups have radical nanotechnology as a goal, but are choosing the soft route – think of Andrew Turberfield at Oxford, who with his Bell labs collaborators is making freely cycling chemical motors from DNA.

Will the opposition of various high-profile scientists to the MNT project inhibit its development? Not materially. The way the scientific reward system works, if some hungry young scientist proves Smalley wrong, that will make his career rather than breaking it. Funding sources are diverse enough and decentralised enough that its very difficult for any one individual, no matter how eminent, to block any particular line of research. Look how long cold fusion sputtered on for, in the face of overwhelming negativity from the scientific community.

Is it a problem that so much work that is described and funded as nanotechnology actually is simply materials science under a new name? Yes, I do think this unfortunate, and I would like to see much more clarity about the different kinds of endeavour that are currently all classed as nanotechnology. I think this is beginning to happen.

Craig Venter's group announced that they made a bacterophage from scratch, and did it all in two weeks. I think it can be done in an hour. As you may know, this is the first step towards their goal of making a completely artificial single-celled organism, from scratch. I think that they will do this by the end of the decade, if not before.

Concurrent work in developmental biology and systems biology should yield how multi-cellular organisms develop over the same time period (end of the decade). The next logical step is to make completely artificial multi-cellular systems. This is real "wet" nanotechnology. This makes it possible to do the things that we really want, like "growing" the ocean-based city-state. Kind of like a floating Hong Kong.

The larger question is not if "dry" nanotech is possible or not. Rather, what is it that we want to create and what is the shortest technological route to creating it? It may or may not be "dry" nanotechnology. If course, if "dry" nanotech is possible, we want to get it first before anyone else does. So, yes the feasibility of dry nanotech should be answered as soon as possible.

BTW, "Nanosystems" did not address the issue of sensing and feedback loops on the nanoscale. This was the single biggest disappointment for me when I read it. I think this a fundamental breakthrough needed for "dry" nanotech to become a reality. The other problem was that there was no experiemental research underlying the book at the time. Of course this may have changed.

My point is that "wet" nanotechnology, which is really an extension of biotechnology, is developing at a very rapid rate. I know this for sure because I am going to sell a biochip scanner instrument and I have concerns about obsolesence. Biotechnology is very much a "tool-driven" field, and the tools are evolving very rapidly. Much faster than semiconductor ICs.

Chris:

Thanks for pushing the MNT debate forward. But back to some basic principles.

Your design for a nanofactory is fundamentally dependent on trillions of "fabricators" working in a phenomenally parallel fashion.

In my coverage of nanotech, I've seen very little serious work on such a fundamental component, the elusive transistor for MNT.

Doesn't the feasibility of MNT hinge on the whether such a critical cog can be fabricated itself in enormous arrays?

And even if such a foundation component were not just possible, but producible, what about the challenges of feeding it the detailed instructions to make different sorts of nanoblocks?

I find your proposed design fascinating and vexing -- the modularity of it seems elegant and common-sensical, yet the journalist in me wonders if some of the most basic assumptions have been glossed over.

I agree, MNT deserves a broader debate about its feasibility, but it has an uphill battle on almost every front.

To Richard Re: molecular machines

I'm glad I understood your argument well enough to restate it elegantly. It looks like you also understood mine, but rejected it perhaps too quickly, so here's some more evidence that floppiness and thermal noise aren't necessary for nanoscale machines.

I did know that ATP synthase wiggles and deforms as it works. The molecules kind of shimmy past each other, getting out of each other's way as needed, with no abrupt energy transitions or high energy barriers. And they deform in just the right way to drive the chemical reaction. Very elegant!

Now the question is, is this much deformation needed in any nanoscale pump? Consider the difference between a diaphragm pump and a piston pump. The former contains a deformable component that changes the volume of a cavity. The latter contains a rigid but moving piston that does the same. They each have their advantages, but they work by the same principle--despite the fact that a diaphragm pump cannot be built with only rigid materials, and a piston pump cannot be built with only floppy materials.

The reason I mentioned Drexler's neon pump (which is reversible to act as a turbine) is that it allows neon atoms to move from one side to the other, and it involves a rotor in a housing. In these ways it is similar to the proton-path half of ATP synthase. But it is completely rigid. Here's a picture and description I should have given last time.

Now, if you hooked it up to the reaction-driving half of ATP synthase, and put a (presumably osmotic) pressure difference of neon across it, it would turn a concentration gradient into rotational motion just like ATP synthase does, and might even be able to synthesize ATP if the torque was adequate.

So is floppy motion necessary for a device that converts concentration gradients into rotary motion? I think this example shows that it's not.

It's interesting that ATP synthase is highly conserved. But to me, that argues that working ATP synthases are pretty rare in the design space, so the design space is highly constrained, and this version is in some way near the limits of what you can do with protein. As such, it would be expected to have lots of kludges.

After I wrote that, I went looking for the efficiency of ATP synthase, expecting to find it was low. But it's very high. But I did find a page that says it can reverse itself and act as a proton pump; if this is biologically important, then thermodynamically, the thing must be reversible, implying very high efficiency. Perhaps this is what constrains the design.

But I also found something even more interesting: a mechanical analysis of its function that implies that thermal noise is not needed. Check this out:

"Very detailed simulations of molecular dynamics show that, while there's a lot of jitter, the important conformational changes of proteins can be decomposed into shearing motions and hinged bending motions. .... Oster was able to write down a very simple model-a "tinker-toy model"-for the mechanical and elastic properties of the synthase, incorporating only those hinges and shears. .... The solutions of these equations are in agreement with experiment, qualitatively and quantitatively. Moreover, when run in reverse-when provided with outside torque, and used to either synthesize ATP or build up a proton gradient-the equations again are in quantitative agreement with experiment."

In other words, a model without thermal noise appears to work just as well as the jittery version.

Chris

To Richard Re: policy and development

I agree that MNT is still at the stage of science and not engineering. And a focused program today would have to fund a lot of blind alleys. But I disagree that all the steps are unclear. For example, we could usefully fund quite a lot of research on diamondoid mechanochemistry. We could start designing a CAD program that can deal with large nanosystems. We could study the likely capabilities of nanosystems and integrated MNT-based products to allow better funding and policy decisions. The diamondoid mechanochemistry may be premature simply because computers are getting cheaper so fast, but the other two are definitely not premature.

Would a Nanosystems review or a study of whether MNT is possible do any good at this point? It's easy for those steeped in MNT to think that such a review would give a definite Yes answer: that the reason MNT is not widely accepted as workable is simply because no highly-credentialed person has looked at it closely. This is reinforced by the generally shallow character of criticisms of MNT. The emails we've been trading about nanoscale friction have shown me that the issue is not as cut-and-dried as I'd thought: in at least some cases, it's not MNT theory vs. nothing, it's MNT theory vs. other theory. Hard to get a definitive answer. Darnit.

This implies that Nanosystems may be a rich source of the kind of paradigm-shifting interpretations that lead to breakthroughs. Hungry young scientists would do well to mine it for ideas. But this process is slow. And when theories clash so blatantly, it shouldn't be hard to devise an experiment that can sort them out. I should note that most of Nanosystems is based on very simple theories conservatively applied, so even proving details of Nanosystems wrong will be a fruitful source of paradigm shifts. Thus a detailed "experimental review" of Nanosystems would constitute very useful basic research.

I'm glad to hear that there's lots of "first-step" work going on. I'd be gladder if it was acknowledged to be relevant to the MNT field--if an MNT field even existed! The existence of a recognized field augments funding and focuses research. And this is the problem with high-ranking scientists denying the possibility of MNT; they are preventing a field from nucleating. Which is more likely to happen in the next 15 years: dry MNT, useful quantum computers, or commercial fusion? Why is it that any trick that might lead to quantum computers is newsworthy, but a grant that includes anything "Drexlerian" is (so I'm told) guaranteed not to get funded by the NNI?

MNT may not be ready for an Apollo project, but I think it is ready for a Manhattan project. There were a lot of dead ends in that, but that was OK. Don't know which way of separating uranium will work? Fund them all! The question isn't whether an MNT project would waste money. The question is, how much money is it worth spending to develop MNT a year earlier, and would this money be enough to fund the necessary dead ends?

If my understanding of what MNT can do is correct, then it's economically worth spending well over $10 billion to get it just one year sooner, and militarily worth spending well over $100 billion to make sure no one else gets it first. These are high stakes--and they are only a pale reflection of the risk/benefit tradeoffs of various development scenarios. I also think that a decade from now, an entire development program will quite possibly cost under $1 billion and take less than three years. This strongly argues for planning ahead, which means spending significant effort to answer as many questions as possible, as soon as possible.

Chris

To Jack and Kurt, Re: technology

Kurt says that Nanosystems doesn't address nanoscale sensor and feedback loops. Not explicitly--there's only so much you can cram into a 500-page book. But it does cover sensors, a little bit (sec. 11.2--and notice that binding pockets, sec. 13.2, can make good chemical sensors). And actuators (sec. 11.6). And digital logic, even general-purpose CPUs (ch. 12). This is just about enough to make a loop. (Mechanical D/A and A/D converters are not hard to design.) And there's also a mechanism for retrying reactions repeatedly until success (sec. 13.3c).

And Nanomedicine I has a lot more info on sensors and actuators, and it's available online.

Jack complains of very little work on fabricators. The most relevant published work is several papers by Merkle in the mid-90's, putting Drexler's ideas into architecture. Start with "Casing an Assembler" and see the References for his other assembler-related papers. And here's his full list of papers.

Jack continues:Doesn't the feasibility of MNT hinge on the whether such a critical cog can be fabricated itself in enormous arrays?

It does hinge on whether a fabricator can itself be fabricated. But assuming one fabricator can be made that is capable of duplicating its own structure, it's pretty clear that it can make an array of two, and those can make an array of four, and so on. This is covered in my paper in section 4.5.

And even if such a foundation component were not just possible, but producible, what about the challenges of feeding it the detailed instructions to make different sorts of nanoblocks?

It depends how many different kinds of nanoblocks you need to make. It's not that hard to broadcast data at a high rate to lots of control computers in parallel, and have them distribute it to a few thousand fabricators apiece. I cover this in 6.3, 8.1, 8.2. Of course, if every fabricator is making a unique nanoblock, I don't know how to do that. But such a thing couldn't be designed either. Designing with repeating nanoblocks is covered in section 5.

I find your proposed design fascinating and vexing -- the
modularity of it seems elegant and common-sensical, yet the
journalist in me wonders if some of the most basic assumptions have been glossed over.

Not at all; the basic assumptions are made explicit, in the Abstract and Introduction, and e.g. at the beginning of section 3: "The fabricator used in the nanofactory is unspecified; the nanofactory design is sufficiently general that a wide variety of possible fabricator designs can be incorporated."

Chris

Edited by author 11-15-2003 12:07 PM
Chris, you found an excellent page on ATP-synthase that clearly states the important points: the motor is stunningly efficient, reversible, and highly conserved throughout evolution. But I’m still going to insist on my other point that flexibility and Brownian motion/thermal noise are essential for its operation. Clearly Oster has done what all good physicists do faced with such complexity – he’s tried to reduce the total number of effective degrees of freedom by introducing his hinged and jointed model. But the need for joints stresses the idea that internal flexibility in the molecule is what makes it work. As for the Brownian motion, the sentence you didn’t quote says: “The tinker-toy is only for illustration; the real model is a set of mechanical equations, fairly straight-forwardly deduced from Newton's laws, in which drag is opposed by stochastic, ultimately chemical driving forces”. Now, I need to look up the original paper to be sure what he’s done, but this sounds like the standard physicists’ treatment of a Brownian system, a Langevin equation, in which you add on to F=ma two extra terms to account for the random collisions a system endures from Brownian motion - a drag force and a stochastic, randomly fluctuating force.

Here’s a simple example to show why flexibility and Brownian motion are absolutely necessary to get these things to work. Imagine a toy model of a molecular motor, that consists of a protein that we can think of as two arms hinged at one end, like a pair of tongs. Halfway down each arm is a cup shaped depression, so that if the arms are brought together a fuel molecule snugly fits in the cavity, held firmly in place by sticky patches. The motor cycle starts with the arms apart. Then a fuel molecule binds to one cup. The arms close up, driven by the lowering of energy that occurs when the fuel molecule fits into both halves of its binding site. The binding of the fuel molecule catalyses its splitting into two halves. It does this, and now there’s nothing to hold the two arms together they come apart, releasing the products, to be ready for the next cycle. Now what actually moves the arms together in the first part of the cycle? The forces that hold the fuel molecule in place are all very short-ranged, so there can be no long-ranged attraction pulling the arms together. Instead, what moves them together is the random buffeting of the surrounding water molecules doing their Brownian motion, allowing the arms to try out many different configurations until they find the right one, with the lowest energy. The French theorist Jacques Prost has devised a beautiful abstract model to show how in general a cyclically varying potential can allow you to rectify Brownian motion to get useful work – the Brownian ratchet.

Now, of course you can design a pump like Drexler’s neon pump, which doesn’t exploit flexibility and Brownian motion in this way. But as we see how efficiently the soft machine can do the job, maybe the motivation for designing the hard/dry one becomes less compelling.

I think we're close to agreeing here.

I understand what you wrote about the need for flexibility, and about the use of Brownian motion to test the conformations and make the process happen. In fact, I'll go one step further, and say that thermal noise is useful to get the system past small energy barriers as it wanders through its configuration space.

Of course, thermal noise can be useful in systems with stiff components as well. As long as the motion of the system does not encounter energy barriers that are bigger than thermal noise, it can move without hindrance. And an arbitrarily small force can cause the motion to have a preferred direction.

You say flexibility is necessary to get things done. If we replace "flexibility" with the more general "degrees of freedom", (just as we can replace "Brownian motion" with the more general "thermal noise"), then I agree with you. But is there any reason why a system built of stiff parts can't have degrees of freedom built in? I'm not sure whether that's a scientific question, or an engineering one: is it possible to build stiff systems with hinges or bearings at the nanoscale?

In your toy motor, the arms wiggle around (binding molecules), clap together (releasing heat), and then separate (releasing products). But how do you get the energy out to do useful work? The arms have to be connected to something, whether in a bio or an MNT system (unless all you want is a catalyst). And that connection can be used to push it gently from one configuration to another, in order to extract work--or to push it firmly, and make it drive reactions "uphill". You may wonder why such explicit driving might be worth doing; I'll get into that below. And, as I said, thermal noise can also be used to drive the system (whether bio or MNT) in the absence of stronger designed-in forces.

Let me try to unify one more set of concepts. You say "the motor is stunningly efficient, reversible, ..." But this is redundant! A motor can only be reversible if nothing goes "sproing" at any point in its motion: if it never goes past an energetic, thermodynamic, or mechanical ratchet. But the "sproings" are what waste energy. To say that nothing goes "sproing" is to say that its path through potential energy space has no dropoffs. At no point in its motion does it undergo any high-energy transitions, so there's very little to thermalize.

So if reversibility and efficiency are the same concept at this scale, any system designer will want to make it as reversible as possible. Every motion and reaction should be a very small energetic step, or should be balanced by an (almost) equal and opposite reaction, as with the proton gradient and the ATP synthesis. But chemistry is not so accommodating. Many reactions don't correspond to whole numbers of ATPs. It's like a market in which you can only pay in $10 bills.

So why would you want a connection from a chemical reaction apparatus to a distant mechanical system? Among other reasons, because that system can be designed to absorb or supply energy by non-chemical means: stretching a spring, running an electrostatic motor/generator, and so on. Yes, ATP synthase is efficient, but there's not much "wiggle room" for it to evolve as shown by the fact that it's so conserved. Plenty of other biological reactions are less efficient despite billions of years of evolution. So I'm not so sure that soft machines are necessarily preferable in terms of efficiency. (Not to mention designability. And ability to fabricate diamondoid.)

And if efficiency is the same as reversibility, then the question "Can a stiff machine be efficient?" becomes the much more straightforward question, "Can a stiff machine be reversible?" That is, does a system made of stiff parts and constrained joints necessarily undergo abrupt potential energy changes as it moves, or can it be designed to move without any abrupt energy transitions?

Well, what can cause abrupt potential energy changes? Several things. A mechanical edge that a piece ratchets over. Forgetting a bit of information, or other forms of abrupt equilibration between two regions with different content. Crossing an energy barrier that is steeper than the stiffness of the system, so that it goes "sproing" as it passes. (This is the general case of the mechanical edge.) But all these things are identifiable physical phenomena, and AFAIK there's no law that says they have to be present in every stiff mechanical system. Certainly I can describe a system that appears not to have them. Such a system would be reversible. And therefore efficient, as long as you didn't drive it too fast.

But biological systems also get inefficient if you drive them too fast. The only reason ATP synthase is efficient is that it maintains the proton concentration gradient such that the energy of a proton exactly equals the energy of an ATP. If you confined the system so that the volume the protons are dissolved in is small compared to the distance between protons, you'd find that the efficiency suffers: the concentration gradient couldn't be maintained at exactly the right level as protons were added or consumed by metabolic processes. So in a very real sense, the volume of bio-space required for efficient operation is far larger than the motor itself.

Be careful talking about the Brownian ratchet; it could sound like perpetual motion. I assume the energy comes from the energy required to vary the potential. BTW, I went and looked up Prost and Brownian ratchets, and found a paper: "Reversible ratchets as Brownian particles in an adiabatically changing periodic potential", Juan M. R. Parrondo, http://seneca.fis.ucm.es/parr/PAPERS_PS/rev-rat.pdf . I couldn't follow all the math, but found some plain-English statements that appeared to be contradictory on p. 3:
"From this equation [7], it follows that no transport of particles occurs if one slowly modulates a potential..." vs. "From the above results, it is clear that, in order to have transport, the change of the potential must be driven, not only slowly..." He also claims to have found a flaw in Feynman's work, and elsewhere refers to potentials changing suddenly (a non-physical assumption--very suspicous) so I'm not sure whether his work is legitimate.

Chris

Yes, to the relief of any onlookers who have persisted this far, I think we can say we agree.

You could certainly design stiff systems with flexible hinges that would also be able to exploit Brownian motion/thermal noise. The key phrase is the one you use "As long as the motion of the system does not encounter energy barriers that are bigger than thermal noise". Actually, this is a very good definition of what I mean by a "soft" system.

Is reversible the same as efficient? The concepts are clearly closely linked, but what is in my mind in separating them is the important distinction between this kind of isothermal motor, which can, it seems, in principle achieve 100% efficiency, and a heat engine, which even if it is fully reversible can still only attain the maximum efficiency given by the Carnot limit.

I think you are quite right about minimising all the individual energy steps. If you look up the original paper about the model for ATP-synthase (Wang and Oster, Nature 396 p279 (1998)) this gives a lot of fascinating detail about how nature does this in this example.

As for Brownian ratchets, the original reference is Prost et al, Physical Review Letters 72 p2652 (1994). It's been a very influential paper, though not all biophysicists agree on its relevance to real protein motors. It is completely in compliance with both 1st and 2nd laws of thermodynamics; as you suggest you need to put energy in to change the potential.

Anyway, to summarise: my important point was that the MNT project isn't the only conceivable way of achieving a radical nanotechnology. The way that nature does it is surprisingly effective and surprisingly subtle, and we'd do well to learn its lessons.

If we're agreeing that one can build a "soft" system with "hard" parts, then I think we've settled the major point of the discussion.

Heat flow is never reversible, and heat engines depend on heat flow, so heat engines can never be reversible in operation even if their mechanisms allow them to move reversibly when there is no heat difference between the hot and cold side.

There are varying degrees of radical. I want to build macro-scale stuff with nano-featured diamondoid. I don't know how to do that other than with MNT. I also want to build intricate large-scale products that I can design directly, according to simple rules I can directly comprehend. I think that can be done with MNT much earlier than we can learn to do it with biochemistry.

So MNT has substantial benefits as a nanotech manufacturing system. (I say "has" not "may have" because I think we're past the point of arguing that it can't work at all.) If we can agree that it's a contender and may reward further investigation, I'll be content.

Chris