Followers of the Drexlerian flavour of radical nanotechnology often accuse nanoscientists of ignoring their approach for reasons of politics or prejudice, and take the lack of detailed critiques of books like Nanosystems as evidence that the whole Drexlerian program is feasible, and indeed imminent. Scientists, on the hand, find the Drexlerian proposals too futuristic and too lacking in practical implementation details to be even worth criticising. The result is an ever-widening gulf between the increasingly bitter Drexlerites and a dismissive and contemptuous mainstream nanoscience community, which does neither side any good. So it’s a very positive development that Robert Freitas has presented a detailed scheme for achieving the first steps towards the mechanosynthesis of diamondoid nanostructures, and even more positive that Philip Moriarty has made a detailed critique of these proposals, based on his deep practical knowledge of scanning tunneling microscopy and surface growth processes.
Philip’s critique is contained in an 8-page letter to The Center for Responsible Nanotechnology‘s Chris Phoenix. The letter was prompted by an approach from Chris, asking Philip to expand on the criticisms of the Drexlerian vision that I reported him making at our joint appearance at the Institute of Contemporary Arts. Chris has, in turn, replied to the letter, and will be publishing the whole correspondence on the CRN web-site in due course.
The letter covers a lot of ground; at its heart is an exploration of some fundamental problems with the Freitas scheme – just how will a diamondoid cluster grow, and is the assumption that the mechanosynthesis tool-tip will grow in the necessary pyramid shape at all realistic? The answer to this seems to be, in all probability, no.
Just as important as this critique of one specific proposal are the general comments Philip makes about the importance of proof-of-principle experiments and of the theory-experiment feedback loop. This gets to the heart of the gulf between conventional nanoscience and the followers of Drexler. To the latter, theoretical demonstrations of feasibility in principle are primary, and considerations of how one is going to achieve the goals are secondary engineering issues that don’t need detailed consideration now. But to nanoscientists like Philip, the devil is in the details. It’s these details that determine whether a theoretically possible outcome will in practise be achieved in 10 years, in 50 years, or never. The Drexlerites tend to say “if x doesn’t work, then we’ll just try y”. But the more and more specific systems we try out and have to discard, the further away we get from the MNT dream of a system that can make any combination of atoms consistent with chemistry.
Freitas and Merkle have taken a very positive step in addressing these issues of implementation and experimental detail. The fact that the proposals can be criticised is positive too; in science this type of criticism isn’t destructive. It’s at the heart of the process by which science moves forward.
Update – 26th January. The whole correspondence between Moriarty and Phoenix, including the original letter, is now available for download here.
[…] t route using diamondoid mechanosynthesis; see the ongoing discussion with Philip Moriarty here for the difficulties that this proposal may face. In conclusion, even if diamondoid-ba […]
[…] notechnology and MNT — Richard Jones @ 10:20 pm In my post of December 16th, Is mechanosynthesis feasible? The debate moves up a gear, I published a letter from Philip M […]
[…] by Robert Freitas for implementing diamondoid mechanosynthesis. The debate is introduced here. The critique received a riposte from Chris Phoenix, of the Center for Responsible Nan […]
It should be noted that the pyramid CVD growth is only one option for making manipulable tool tips. Freitas presented a second. An unterminated surface (e.g. a broken AFM tip) should bond to the tool molecule if lowered onto it. The bonds will be random, and the system will be less stiff and less well characterized, but this should allow the molecule to be plucked off the surface and used in experiments.
So the implication that Freitas’s proposal depends on CVD pyramid growth is incorrect.
Chris, your reply is a very good illustration of the point I make in my fourth paragraph. I’ll leave it to Philip to comment on the alternative proposal.
Chris,
As you know, I’ve raised a number of questions about the second option for Freitas’ tip tool handle strategy in both the original letter and in our more recent correspondence. Freitas himself has already identified a number of problems with this second option. I quote from Freitas’ presentation: “This method produces an inferior tool because the position, orientation, number, and stiffness of attached tooltip molecules is poorly controlled.” I entirely concur with this and have raised related issues in my most recent letter to you. [Remember, we’re trying to do atomic precision chemistry here: ‘poorly controlled’ atomic positions suggests to me errors of at least an atomic diameter (c.f. discussion of definitions in my most recent letter to you). Note that this is not a question of Heisenberg-related uncertainty: rather, it’s a question of controlling bonding sites with atomic precision. The ‘viable’ parameter space just got a little narrower…]
Another important issue related to the CVD growth of the pyramid is that you stated in a post on the CRN website related to Freitas’ straegy that “One way to attach a multi-micron “handle” to the molecule is simply to use a CVD diamond-growing process; this has already been experimentally demonstrated on other molecules”. This is at best misleading because although CVD growth of diamond crystals from nucleation sites was of course previously demonstrated by Giraud et al. (as cited in Freitas’ proposal) that group most definitely did not report the growth of ‘inverted pyramids’ as required for Freitas’ tip-tool handle. (That ‘seed’ molecules should nucleate crystal growth is just basic crystal growth surface physics, as I’ve pointed out in our – as yet unposted -most recent correspondence). You’ve asked me to be careful with distinctions and to ensure that I don’t conflate systems engineering-related issues (something of which I’ll admit I’ve been guilty). Might I ask that you also strive to distinguish between your perception of an experimental result and what has actually been achieved in the experiment?
I am enjoying our correspondence and look forward to your response to my most recent letter. (I also want to stress (once again) that Freitas et al. are to be applauded for putting forward a detailed strategy for implementing the ‘machine language’ of mechanosynthesis.
Philip
I’ve spent days and days studying the feasibility of Drexler’s work, and this is the first solid, technical criticism I’ve seen. Most other criticism–including the articles and letters by Dr. Smalley–is couched in metaphor. Smalley’s observation that, “You don‚Äôt make a girl and a boy fall in love by pushing them together,” doesn’t tell me where to look for specific flaws in Freitas and Merkle’s computational chemistry.
(Dr. Smalley may have perfectly understandable reasons for his rather condescending tone and his apparent avoidance of technical detail, but both these aspects of his style undermine his argument.)
So many thanks, Dr. Moriarty, for your detailed technical criticism. And thanks also to Chris Phoenix for participating in this discussion. I’d love to see some serious proof-of-concept experiments, and this kind of informal peer review is a step in the right direction.
And a couple of questions for Chris Phoenix: (1) Have you posted any more of this discussion online?, and (2) Is there an online bibliography of work relevant to assembler tips, with careful annotation of what each paper actually shows about mechanosynthesis? I’m interested in both experimental work and ab initio quantum chemistry.
I don’t actually see anything in Philip Moriarty’s letter which is a technical criticism of Drexler’s work. He complains that Drexler cannot show him a working prototype or experimental proof of principle, and Drexler basically says that’s still too far off. As an experimentalist, that leaves Moriarty nothing to grasp in Drexler’s work. Most of his effort is devoted to a critique of Frietas’ recent proposal for a near-term mechanosynthesis system. This is Freitas’ own conception and while it is no doubt inspired by Drexler, it is pretty much independent of Nanosystems.
It’s interesting to note that the preceding comments by Eric K and Hal Finney are completely at odds with each other. While Eric notes “this is the first solid, technical criticism I’ve seen”, Hal states that he doesn’t see anything in the letter which is a technical criticism of Drexler’s work. Let me address Eric and Hal’s comments in turn.
Eric:
First, thanks for your supportive comments – they’re much appreciated. It would be remiss of me not to point out that Chris Phoenix has responded to my letter with a carefully considered discussion of a number of the issues I’ve raised. This prompted a second letter from me to Chris to which I have yet to receive a reply (although it’s only relatively recently that I sent the second letter to Chris). There have also been rather a large number of e-mails back and forth regarding the fundamental physics/chemistry of nucleation and growth and the diamond CVD growth process outlined in Freitas’ proposal. As noted by Richard (Jones), Chris has promised that he will post this correspondence on the Centre for Responsible Nanotechnology website (www.crnano.org) in due course – I would be extremely interested in your comments when the discussion documents are uploaded. I’ll let Chris provide you with a comprehensive bibliography of mechanosynthesis papers but a good place to start is Freitas’ website. (Although, as noted in my more recent correspondence with Chris, one needs to be extremely careful with just how the term “mechanosynthesis” is defined).
Hal: As stated at the start of my letter, and as you correctly point out, a key focus of my first letter to Chris is Freitas et al.’s work. As you are no doubt aware, Frietas’ most recent proposal is a logical progression from Merkle et al.’s work on mechanosynthesis in the mid- to late nineties. This work in turn builds on Merkle’s diamondoid mechanosynthesis quantum chemistry research cited in Nanosystems (see, for example, Section 16.2.2). Importantly, however, and as I’ve pointed out in more recent correspondence with Chris, Drexler himself suggests that in his “backward chaining” strategy, Merkle’s tip-based mechanosynthesis could be used as the first, “nearest term” step (Section 16.2.3). Drexler’s argument, as you suggest, is that I’m asking for a “moon rock” before the appropriate rocket technology is developed. Why then is Merkle’s vacuum-based mechanosynthesis strategy (from which Freitas et al.’s proposal stems) cited as a possible Stage 1 for Drexler’s ‘back chaining’ process? I’ll not develop this argument any further here – it’s detailed at length in a second letter I’ve sent Chris Phoenix and which I’m more than happy to send to you if you’re interested. (Note also that I can’t agree with your suggestion that Freitas’ proposal is “independent” of Nanosystems: the proposal is based firmly on the diamondoid chemistry described in Section 8.6).
I obviously also do not agree that there is nothing for me “to grasp” in Drexler’s work. (Indeed, it’s intriguing that you state “**As an experimentalist***, that leaves Moriarty nothing to grasp…”. Do you mean that there is nothing in Drexler’s proposals for *any* experimentalist to grasp?). My key interest is in the “low level” or “machine language” steps in molecular manufacturing as these involve basic single molecule manipulation steps (something on which the Nottingham Nanoscience group has worked for the last decade or so). Note that I am not asking Drexler (or indeed any of the proponents of Drexler’s work) personally to provide an experimental “proof-of-principle”. If there were a coherent strategy anywhere in Drexler’s work for implementing the lowest level mechanosynthesis steps I would be more than willing to attempt the experiment myself! This is why I was so interested in Freitas’ proposal and I read it with an open mind, carefully considering just how I might try to implement the proposal experimentally. What is frustrating about “Nanosystems” and Drexler’s work in general is that although there are detailed blueprints of molecular machinery provided at one level of abstraction (e.g. Fig. 13.11, Fig. 13.14, Fig. 10.32, etc..etc..), the basic mechanosynthesis steps are, at best, ‘glossed over'(this is also covered in my second letter to Chris). Drexler has asked both Smalley and myself not to attack “straw men” we have put forward and to focus on specific proposals in Nanosystems. My responses to this are: (i) to the very best of my knowledge, I have not put forward a “straw man”, (ii) when asked directly about the detailed mechanism underlying a particular proposal in Nanosystems (e.g. Fig. 8.14 – see discussion in my first letter to Chris Phoenix), Drexler’s rather non-scientific response is to state “well, if that doesn’t work, we’ll just choose another reaction – there’s nothing fundamental about that one” – i.e., as Richard puts it, “if x doesn’t work, we’ll try y”. Fig. 8.14 is very important because, for me, it highlights key steric hindrance issues (whether those issues arise due to the presence of more than one manipulator/ finger or due to the mounts on which the manipulators are held).
So, contrary to your view, I feel that there is much for me “to grasp” in Drexler’s work. Chris Phoenix and I are currently discussing many of the issues you raise in your comment (and which I’ve briefly addressed here). Any comments that you might have on the discussion (when it’s posted on the CRN website) would be more than welcome.
Best wishes and happy holidays,
Philip
Drexler’s work in Nanosystems rests on the assumption that feasible mechanosynthesis operations exist for stiff hydrocarbons. If such pathways can’t be found, Nanosystems collapses like a house of cards.
Freitas and Merkle have done admirable work investigating one set of tools for adding carbon to a C(110) surface. This took them over 5 years of CPU time, and resulted in a tool that might place a carbon dimmer on a diamond surface 20% of the time. This suggests two things: (1) It’s not inconceivable that a plausible set of reactions exists, but (2) it’s going to be a long, ugly slog to find out whether or not Nanosystems is feasible.
In one sense, Drexler is right to say Nanosystems is still too far off. If Moore’s law holds up, we’ll have 200-300 times more computing power in 10 years, and we’ll be able to afford a lot more computational chemistry. So Drexler could choose to sit around for the next 10 years and wave his hands about critical technical details. But there’s no need to fund hand-waving. The only near-term work worth funding is (1) more investigation of mechanosynthesis reactions and (2) a certain amount of hypothetical policy work by people like CRN.
Obviously, if somebody finds a good set of tools for controlled diamond synthesis, then Drexler’s work becomes extremely relevant. But for now, Drexler is just a (fairly rigorous) futurist waiting for somebody to confirm a few enormous–but not completely implausible–assumptions. There’s nothing wrong with that, but it’s not something that will necessarily earn him respect from the establishment.
(And until the assumptions in Nanosystems are rigorously tested, any claims of “universal assemblers”–which appear to be a vastly more difficult problem–suggest the speaker is a bit of a crank.)
(Comment 7 was a response to Hal. I saw Dr. Moriarty’s comments after posting my own. More in a second.)
Eric,
I wholly concur with your comments. In addition, you might be interested to know that Drexler has recently written the following in response to my usage of the term “universal assembler”:
” ‘a universal assembler’…
— is a bad description, when used by an advocate of molecular manufacturing.
— is a straw man, when used as a basis for criticism.
— is a distortion of what I proposed in 1986, and has even less to do with ideas since 1992.
Broad mechanosynthetic abilities do not imply “an” assembler, much less a “universal”
assembler.’
I’m particularly confused by the idea that the term “universal assembler” is a “straw man” and “a distortion” of what Drexler has proposed when “Engines of Creation” features the following lines (under the heading “Universal Assembler”):
“Because assemblers will let us place atoms in almost any reasonable arrangement (as discussed in the Notes), they will let us build almost anything that the laws of nature allow to exist”.
…and there’s more (again from Engines of Creation): assemblers will “use as “tools” almost any of the reactive molecules used by chemists – but they will wield them with the precision of programmed machines. They will be able to bond atoms together in virtually any stable pattern, adding a few at a time to the surface of a workpiece until a complex structure is complete. Think of such nanomachines as assemblers.”
Philip
Dr. Moriarty: Many thanks for your kind response. But I hope I haven’t misled you as to my credentials. Like Chris Phoenix, I’m a sofware developer by training, and I’m not even remotely qualified to comment on the “machine language” of mechanosynthesis.
Nonetheless, I’m deeply interested in whether such a “machine language” exists. I have a personal interest in compilers, which translate high-level software designs into regular, old-fashioned machine language. There’s a lot of fascinating software problems[1] involved in a nanofactory, but there’s not (much) point in pursuing them unless the low-level details are credible.
Previously, I had found a series of papers on low-level mechanosynthesis by Merkle and Freitas, but no serious critical papers by anybody else. Your letter pointed me towards a good bit of literature, and allowed me to start learning about the critical issues. I still don’t know whether mechanosynthesis is plausible, but I’m developing a layman’s knowledge of the open problems. Obviously, I’m following this entire exchange with great fascination. Hence, my warm thanks to both you and Chris Phoenix.
[1] “Fascinating software problems involved in a nanofactory”:
A nanofactory is basically a giant, parallel hardware compiler. It has to be self-hosting (i.e., capable of compiling itself), and it has to be initially bootstrapped from something smaller and more primitive. In different forms, self-hosting and bootstrapping are day-to-day issues for compiler authors, although both processes always retain a certain brain-bending quality.
There’s also some pretty exciting parallel control algorithms involved in running a nanofactory. Obviously, it’s easiest to deal with the problem by sending identical instructions to thousands of assembly units at time, analogous to how a SIMD (“single instruction, multiple data”) processor works. The engineers at Thinking Machines studied these algorithms extensively back in the 80’s, with help from Richard Feynman. More recently, nVidia has been building pseudo-SIMD graphics cards, which are stunningly fast, fun to program, and involve lots of advanced (proprietary) compiler technology.
When I look at the aspects of a nanofactory that I am qualified to discuss, I see dozens of fascinating research projects. But speaking on a gut level, I don’t expect a lot of show-stoppers on the computer science end of things. Hence, my interest in the feasibility of a “machine language” of mechanosynthesis.
Once again, my thanks to you, Chris, Feitas, Merkle, and everybody else working on this problem.
(Hmm. WordPress silently ate my response. Perhaps it’s being held for moderation, or perhaps it’s been entirely lost. I’m going to try rewriting it.)
A clarification re. comment 9. I endorse all of Eric K’s comments save the final suggestion regarding Drexler’s ‘crank’ status. It’s worth noting that Drexler has carefully considered a number of physical principles underlying the ‘high level’ aspects of the nanosystems he proposes and, indeed, has thought in some detail re. power loads, operation bandwidth etc.. My overarching issue with “Nanosystems” (and all of Drexler’s work), however, is that the higher level components and processes are considered at the expense of a detailed study of the lowest level mechanosynthesis steps. Moreover, when challenged on these low level steps – and as noted above – Drexler stonewalls and states that it’s just a matter of finding the correct set of reactions. Nevertheless, I have found that reading “Nanosystems” is a useful exercise as it raises questions and issues related to the ultimate limits of nanoscience. If Drexler were simply a ‘crank’ then I don’t think that this would be the case.
(OK, WordPress absolutely refuses to post my longer comments. I’m going to have to bow out of this discussion for now. I’ll e-mail my comments directly to Dr. Moriarty.)
Eric, sorry your comment got swallowed. As you guessed I’ve got an anti-spam filter on that holds comments for moderation if they have more than a few external links. This is unfortunately all too necessary – I’ve just fished your comment out from a dozen spam comments advertising online poker.
Dr. Moriarty: I’m glad that Dr Drexler is not proposing universal assemblers. This speaks well for his intellectual integrity. (If my comments imply Drexler wrote an overheated paragraph or two in the 1980’s, so be it. Let the reader make his own decision as to whether Drexler has ever supported universal assemblers, and whether or not this invalidates his more sober work.)
Hi Eric,
Re. “universal assemblers”.
In our current discussion (and the accompanying debate with Chris Phoenix) I want to avoid, where possible, questions of intellectual integrity and the like. Nevertheless, regarding the question of the “universal assembler” and the statements reproduced in Comment 9 above, I unfortunately can’t share your opinion of Drexler’s stance. I’d like to spend some time on this point because the question of “universal” vs “limited parameter space” is extremely pertinent to not only my (and Richard Jones’) debates with Chris Phoenix but to the entire molecular manufacturing dispute.
First, for Drexler to state that a universal assembler ‚Äúis a straw man, when used as a basis for criticism‚Äù is, to me, absolutely remarkable and, moreover, completely groundless. To my understanding, by ‚Äústraw man‚Äù we mean an approach whereby an opponent’s position is effectively misrepresented to make it easier to attack. Moreover, a ‚Äústraw man‚Äù attack is based on one‚Äôs own misunderstanding of an opponent‚Äôs position and not the actual position. Hence, for Drexler‚Äôs ‚Äústraw man‚Äù assertion to be valid, a ‚Äúuniversal assembler‚Äù must be a construct which Drexler has not put forward and simply arises from critics‚Äô poor understanding/perception of his molecular manufacturing ‚Äòvision‚Äô.
Reiterating my remarks in Comment 9 above, Drexler has stated in “Engines of Creation” that assemblers will “use as “tools” almost any of the reactive molecules used by chemists” under a section entitled “Universal Assemblers”. If Drexler were to state in response to his critics that, yes, this particular section of “Engines of Creation” was, as you suggest, “overheated” then I would gain a great deal of respect for his intellectual integrity. Instead, Drexler argues that those who attack his position on the basis of a universal assembler do so because they misinterpreted his work. I am firmly of the opinion that it’s rather difficult to misinterpret the phrase “almost any of the reactive molecules used by chemists” (see also quotes and associated discussion below).
You asked (in our e-mail correspondence) whether there were further instances of material in “Unbounding the Future” or elsewhere which contradict Drexler’s suggestion that “universal assembler” is a distortion. The quotes below are taken directly from Drexler‚Äôs writings or from those of proponents of the Drexler molecular manufacturing vision. You are correct to state that it appears that the Foresight Institute don‚Äôt support the use of the term ‚Äúuniversal assembler‚Äù. However, not only do other statements on the http://www.foresight.org website directly contradict this stance (see quote 1 below), but Foresight‚Äôs lack of support for the term does not mean that the critics who attack the concept of a universal assembler have ‚Äúdistorted‚Äù Drexler‚Äôs views. [Note that I‚Äôm taking ‚Äúuniversal assembler‚Äù to mean a nanomachine which can assemble structures using any of the elements in the periodic table. (Chris, please correct me if I’m wrong).]
1. ‚ÄúMNT proposes to arrange elements drawn from the entire periodic table into useful structures‚Äù, RC Merkle and RA Freitas, http://www.foresight.org/stage2/project1A.html. [This is a particularly interesting quote as it apparently indicates that even key proponents of the Drexlarian vision have difficulties in interpreting Drexler‚Äôs work. Is this down to a ‚Äústraw man‚Äù attack or a ‚Äúdistortion‚Äù ‚Äì as Drexler imagines – or does it arise simply from Drexler’s ambiguous and unclear writing?]
2. “Assemblers will be able to make virtually anything from common materials without labour, replacing smoking factories with systems as clean as forests”. Engines of Creation, Chapter 4. [Now, what precisely does Drexler mean by “virtually anything” here? Does he mean that almost every single material and structure can be made from diamondoid technology? (Just what is meant by “common materials”?) Or does he mean that an assembler will be able to manipulate all the technologically important elements (perhaps “embedded” in organometallic molecules)? Or does he mean that, as Merkle and Freitas state in Quote 1 above, that assemblers will be able to arrange elements drawn from the entire periodic table? ]
3. “Assemblers and other machines in molecular manufacturing systems will be able to make almost anything, if given the right raw materials” from “Unbounding The Future”. [It’s “almost anything” rather than “virtually anything” in this case but I, at least, find it hard to draw a clear distinction between these descriptions. Again, if the ability to synthesise “virtually anything” does not suggest that Drexler had in mind a universal assembler then I’d very much appreciate an explanation of what, **precisely**, Drexler meant. (Chris, perhaps you could clarify (for both Eric K and myself) what’s meant here and how it relates to the concept of the “universal assembler”. Does he mean that we can make virtually anything via a limited set of mechanosynthesis reactions?)].
4. In his 2001 article in Scientific American, Drexler once again suggests that an assembler could in principle build “almost anything”.
Finally, I also find it remarkable that Drexler seems now to be stepping back from the use of the term “assembler” (see Comment 9) in mechanosynthesis when it appears ~ 119 times in “Unbounding The Future” and is both defined and used in “Nanosystems”. While Drexler is of course free to continually revise and refine his thinking on particular issues – this is how science and technology progress – to supersede the discussions in previous publications by accusing his critics of constructing “straw men” is not at all helpful.
Best wishes,
Philip
‘Note that I‚Äôm taking ‚Äúuniversal assembler‚Äù to mean a nanomachine which can assemble structures using any of the elements in the periodic table.’
I don’t believe that is the usual usage. Rather, a universal assembler is one which can construct every imaginable (stable) molecular structure. It is the difference between being able to construct “almost anything” or “virtually anything” and being able to construct literally “anything”. I’m not sure why it is an important difference but I assume that Drexler would not object if critics simply used those phrases. Perhaps an “almost universal assembler” would be an acceptable term, although as you note Drexler is now pushing the nanofactory concept rather than the assembler, I think for safety reasons.
Hal,
I’m now even more confused! To construct “every imaginable” stable molecular structure you need to have the ability to incorporate the majority of the elements in the periodic table!! Let’s think of a few simple materials systems:
1. Benzene, the acenes, and diamond – OK, in this case we need just carbon.
2. Let’s now think of a *slightly* more complex family of molecules: the phthalocyanines. These planar (or sometimes not-so-planar) molecules come in a plethora of ‘flavours’: Cu phthalocyanine (Pc), Pb Pc, Sn Pc, Co Pc…. So we need to choose elements quite far ‘up’ the periodic table to synthesise these molecules *and*, as compared to diamondoid chemistry, we need to use a relatively large subset of the table.
3. Phthalocyanines are, in fact, a subset of a very large family of “(stable) molecular structures” known as organometallics. I suggest that you try a Google search – there’s a vast amount of literature on these systems.
4. I work quite a bit with a rather exotic type of molecule called an endohedral fullerene. This comprises a fullerene cage with an atom on the **inside**. The encapsulated atoms can be rather large(e.g., a lanthanide).
There are countless numbers of other examples of “stable molecular structures”, as you put it, which, if to be constructed by an assembler, will require that the assembler can access a very, very large amount of the periodic table.
I think that the source of the confusion is, as I’ve stressed a number of times to Chris Phoenix, that a common vocabulary or set of definitions has not been agreed upon. You seem to be arguing that there’s a difference between the terms “virtually anything” and “anything” on the grounds that there’s only a subset of all **imagined** molecular structures that will actually be viable on the grounds that they obey the laws of physics. I have absolutely no problem with this. I’m not arguing that Drexler has ever suggested building a molecule that can’t physically exist.
To summarise, I’m of the opinion that we actually concur regarding the definition of a molecular assembler. That is, to build every “imaginable (stable) molecule structure” will require a “nanomachine that can assemble structures using any of the elements in the periodic table”.
Best wishes,
Philip
Erratum to Comment 18:
The second sentence should be: “To construct **”every imaginable”** stable molecular structure you need to have the ability to incorporate all the elements in the periodic table”.
(c.f. Freitas and Merkle’s statement: “MNT proposes to arrange elements drawn from the entire periodic table into useful structures.”)
Philip
Hal,
Another erratum to Comment 18 (apologies):
Regarding point 1: benzene and the acenes obviously also require hydrogen!! (As does diamond if we need to passivate the surfaces).
(Write in haste,…)
Once again, apologies.
Philip
Philip, yes, I agree, that the definition of “universal assembler” that I understand does include your definition of being able to assemble molecules using every element in the periodic table. But mine is actually a stronger definition, which includes yours as a subset. Your definition does not make it clear that not only must all elements be usable, but they also must be able to be put together in all possible configurations.
The point is that Drexler objects when people say that he claims it is possible to build a universal assembler. The stronger the definition which is being used, the more extravagant such a claim would become. When I suggest a stronger definition than you used, it therefore makes Drexler’s objections more reasonable, if in fact he never actually made that claim. (I don’t know whether he may have actually used the term in some of his earlier writings.)
You quoted him and his collegues as suggesting that nanotechnology will ultimately be able to build “virtually” or “almost” anything. Unfortunately as you note this fuzzy phrasing is open to a variety of interpretations. I don’t think you will be successful in pinning Drexler or any other advocate down as to what exactly the range of capabilities will be. It seemed to me that the main point of Nanosystems was to offer a lower bounds on what would be possible. What the upper bound is, I doubt that anyone knows.
I find these notions of mechano-synthetic access to any known structure to be somewhat naive in the face of the state of the art of synthetic organic chemistry. This is not to say that synthetic organic chemistry is in any way a limitation here, but rather it gives us a very large degree of experience in what to expect. Most notably, as any organic chemist can tell you, there are a variety of problems beyond those of simply introducing the right atoms into the right places:
1) The intermediate structures need to be stable — if they aren’t, the intermediate product might decompose / rearrange on your synthesis apparatus.
2) The reaction timing on an atomic scale is a quantum event and may be highly irreproducible. Getting reactions to occur on any dependable fashion is possibly going to be difficult. Sometimes the reaction may not happen before you go to the next step. Waiting long enough for the reaction to happen with any degree of certainty might be long enough to allow other competing reactions to happen or simply reduce the throughput of the assembly device to such a low level that the return on investment just isn’t worth it.
3) If something goes wrong, or even just doesn’t go at all, how would you know that it did? Nanoscale synthesis is going to require nanoscale detection, characterization and (likely) purification methods to remove corrupt intermediates from the assembly device. Can you build a device to purify arbitrary undesired reaction products from a universal synthesis machine?
4) Some positions are simply sterically inaccessible. You’d shatter the molecule or drive it off the synthesizer before you got a reaction to happen there. Endohedral fullerene are an obvious extreme example but there are really a great many.
I have no doubt that with extensive optimization, specialized structures can be created to do specific chemistry on highly specific substrates. Work ranging from classical enzymology to C.-H. Wong to Noyori/Sharpless tells us that it can be done. However, it is unreasonable to suggest that we will spend the effort to make an infinite number of highly optimized specialized catalysts. This approach might be good for the 10% of chemistry that covers 90% of industrially interesting process, but it isn’t the whole answer, and frankly I’m not sure that the field of nanotechnology has anything to add to this endeavor.
Likewise, it seems naive (to say the least!) that a single apparatus could do general purpose chemical synthesis. Such “chemical Turing machines” would no doubt simply reprove in the physical world what we already know about software. The more general purpose the software, the slower, more complex and less practical it is. In the cases where competing reactions are prevalent and therefore time or exquisite selectivity is of the essence, there would be a number of cases where a general purpose machine could not hope to yield product in any significant amount. One need only look at the complexity of the ribosome, which is hardly very general at all, to get an inkling of how complex (read: flaky, impractical, irreprodicible) a universal chemical synthesis machine is likely to be.
I would argue that the set of molecules comprising the difference between being able to synthesize “virtually anything” and “anything” on a universal synthesizer is in in practice going to be “nearly everything”.
In response to Comment 23:
Ian:
I completely concur with your persuasive arguments. As Richard points out in the commentary at the start of this thread, I’m currently embroiled in a debate with Chris Phoenix on the **fine detail** of the chemistry that might possibly be achieved with a computer-actuated nanomachine as envisaged by Drexler. (Richard Jones has also had debates with Chris Phoenix in the past and I believe that he will post them on this website soon. I believe that this will be of keen interest to the nano community because, in his discussions with Chris, Richard has raised some insightful and important points re. surface chemistry and friction). I hope that Chris will post our correspondence on the CRN website in the near future because I think that we have covered a lot of ground related (either tangentially or directly) to your posting.
In Comment 18, I chose a few very simple examples from the vast(!) set of possible stable molecules to point out to Hal that a universal assembler *necessitates* access to the entire periodic table. You strengthen this point eloquently in your posting. If you don’t mind, I’d like to follow up on the points 1-4 you raise:
1. Chris Phoenix’s or Ralph Merkle’s argument, when faced with the types of issue you raise re. intermediate structures, is simply to state “well, if we’re faced with a ‘difficult’ intermediate, we’ll make sure we don’t choose systems in which these difficult transition states appear.”. I paraphrase here. A direct quote from Merkle is as follows: “A simple strategy, therefore, is to ban intermediate structures with many dangling bonds” (http://www.foresight.org/SciAmDebate/SciAmResponse.html). Just how this is consistent with Merkle and Freitas’ assertion (http://www.foresight.org/stage2/project1A.html) that molecular nanotechnology “proposes to arrange elements drawn from the entire periodic table into useful structures” is beyond me.
2. I think that I can pre-empt the Drexlerite response to this. It will be that single molecule manipulation and bond formation have already been demonstrated in scanning probe experiments and that in covalently bound systems, such as the diamondoid structures detailed in Chapter 8 of “Nanosystems”, the chemistry is carried out molecule-by-molecule. To address throughput concerns, one uses a set of nanomachines working in parallel. You’d also probably be pointed to Chapter 6 of Nanosystems. [Chris, apologies if I misrepresent you here]. What’s again of key importance, however, is that to surmount the problems you raise, an extremely judicious choice of materials system is required. It is this issue that is key. For example, both Richard Jones and I agree that “Nanosystems” is a reasonably well-argued and systematic study which does not suggest anything that contravenes basic physical law (see also Richard’s post of 28/12/04: “Molecular nanotechnology, Drexler and Nanosystems – where I stand” on this website). However, Drexler simply glosses over the details of the mechanosynthetic chemistry and uses the “if ‘x’ doesn’t work, we’ll try ‘y'” argument when challenged on specific reactions. [My apologies for bringing this point up yet again but it is one of the issues at the core of my difficulties with molecular manufacturing].
3. The Drexlerite camp will argue [Chris, again please correct me if I’m wrong] that one works with a system whereby a high level of redundancy can be built in to address the problem of error correction. I’ll defer a discussion of this until later because it’s something I want to pursue at some length with Chris Phoenix (when I get a response to my second letter *and* we’ve addressed a number of other issues I’ve raised with regard to the ‘machine language’ of mechanosynthesis). If you’re interested, Chris’ recent arguments on the subject of error detection/correction are at the Nano Tsunami website.
4. Absolutely, wholeheartedly agree! But note that you say “some reactions”. To advocates of Drexler’s vision this simply means that we neglect those reactions! (For those of us who have extreme difficulties with the molecular manufacturing ‘vision’, the viable parameter space just shrunk once again…)
Finally: “I have no doubt that with extensive optimization, specialized structures can be created to do specific chemistry on highly specific substrates”. This is precisely the point that I have been making to Chris Phoenix and it’s helpful to see it expressed rather more concisely and eloquently than I ever could!
Hal:
Drexler of course objects when people state it’s possible to build a universal assembler. He objects for the very reasons outlined by Ian Ollmann and myself regarding the necessity for a nanomachine to access all the elements in the periodic table and mechanically position them anywhere consistent with physical law. That’s fine – he *should* object to this. Nevertheless, my point is that it is wrong for Drexler to state that the concept of a Universal Assembler is a
“straw man” and “a distortion” of what he has written in the past (for the reasons I list in my comment to Eric K , #16 above.) Drexler introduced the term “Universal Assembler” in “Engines of Creation” and has repeatedly suggested that molecular manufacturing will be able to build “almost anything”. Why doesn’t he simply retract those statements rather than argue that his critics have constructed a “straw man”?.
“I don‚Äôt think you will be successful in pinning Drexler or any other advocate down as to what exactly the range of capabilities will be”. I thoroughly agree. Nevertheless, admitting to not knowing what the range of capabilities will be is a very different assertion to that which suggests “we can build virtually anything”. I also can’t agree that “Nanosystems” shows what the lower bounds might be. Without a detailed consideration of the chemistry of each system (which Drexler does not carry out) we can’t confidently predict *any* type of bound. Moreover, it’s worth noting that Drexler has argued in Nanosystems that “it seems reasonable to assume that most reasonably stable structures – diamondoid or not – will prove susceptible to mechanosynthesis”. Just how can he make this assumption?! (And, once again, note the ‘fuzzy’ language – just what does “most reasonably stable structures” encompass?)
Best wishes,
Philip
Further to the preceding comment, while browsing the CRN website I came across the following statement from Chris Phoenix:
“So I’m not sure that redundancy is needed at the level of molecular machines. It’ll be needed at higher levels, say between 100 nm and 1 micron, but at those levels redundancy can be implemented engineering-wise.”.(http://crnano.typepad.com/crnblog/2004/11/mainstream_acce.html)
This is a very interesting statement bearing in mind the reliability of dimer placement observed in Freitas et al.’s quantum chemistry work (see Eric K’s comment (#7) above) and point 3 of Ian Ollman’s posting (Comment #23). Chris, it appears that I misrepresented your stance re. redundancy – my apologies. However, I’m now in the dark as to how error correction occurs at the lowest ‘machine language’ level. Are you assuming that the mechanosynthetic steps are effectively error free?
Philip
Still browsing the CRN website…
The following comment from Chris Phoenix encapsulates in a nutshell the “if ‘x’ doesn’t work then just choose ‘y'” rationale at the core of so much of the Drexlerite position:
“If a surface proposed by Drexler reconstructs, just choose a different surface.”
(http://crnano.typepad.com/crnblog/2004/07/conventional_wi.html)
Chris, do you realise just how small a subset of surfaces actually *don’t* reconstruct? (Let’s also assume that we’re considering surface reconstruction as distinct from surface relaxation. See “Physics at Surfaces”, Zangwill (particularly Chapter 3) and books with similar titles by Lueth, Prutton, Woodruff and Delchar, Venables for a discussion of the distinction between surface reconstruction and surface relaxation – it’s not covered in “Nanosystems”. See also http://www.nottingham.ac.uk/~ppzpjm/amshome.htm for further references). There’s a large free energy cost to the formation of dangling bonds and, as you are no doubt aware, surfaces will form very many fascinating and beautiful arrangements of atoms so as to reduce the dangling bond density.
Wow! – lots of comment here. It’s nice to see that people are discussing some of these issues. But, where are the experiments? I think Philip has commented on this several times. Many supporters of “mechanosynthesis” (i.e. Drexlerians or whatnot) lack the necessary background to suggest or perform useful experiments. After all, only a few dozen groups exist that can do UHV STM. This situation is unfortunate. Many of the concepts presented in Nanosystems are interesting; but, a lack of sufficient detail has probably discouraged many from pursuing experimental work.
Hi Matt,
“A lack of sufficient detail has probably discouraged many from pursuing experimental work.” I couldn’t agree more! As I say under Comment#6 above:
“If there were a coherent strategy anywhere in Drexler‚Äôs work for implementing the lowest level mechanosynthesis steps I would be more than willing to attempt the experiment myself! This is why I was so interested in Freitas‚Äô proposal and I read it with an open mind, carefully considering just how I might try to implement the proposal experimentally.”
You might also be interested in Richard’s most recent post, “Molecular nanotechnology, Drexler and Nanosystems – where I stand” (at http://www.softmachines.org/wordpress/index.php?p=60) and the follow-up comments. My stance is very closely aligned with Richard’s except that I don’t have his expertise in soft matter and ‘wet’ systems – my background is largely in UHV-based spectroscopy and UHV scanning probe microscopy. (Although over the last couple of years I have spent quite a lot of time on colloidal nanoparticle assemblies formed via spin-coating from simple volatile solvents).
Best wishes,
Philip
Happy New Year!
And a big ugh–I spent several days of my vacation studying Nanosystems in as much detail as my time and lack of expertise permitted. I think I understand Drexler better now, for what it’s worth.
In his technical work, Drexler frequently admits that “the work outlined here could readily absorb researcher-centuries of effort.” Drexler-the-scientist is clearly aware that he’s making high-level arguments for the existence of some workable solution to each given problem, even if he doesn’t know that solution himself. As such, he’s building castles in the air, and leaving the hard work up to everybody else. To my reading, this is a constant theme throughout Nanosystems, and Drexler rarely (if ever) pretends otherwise.
(In Drexler’s popular writing, though, he dispenses with the disclaimers. Intellectually speaking, let’s say this is a rather regretable habit. Making wild predictions and omitting your caveats understandably rubs many scientists the wrong way.)
Personally, the most intriguing part of Nanosystems was the concept of a molecular mill. Drexler’s designs don’t rely on “robot arms” for bulk synthesis. Instead, they bond each molecule to a stiff substrate and run that molecule past a fixed set of “tools”, which are essentially rather unusual reagents and catalysts, each prepared to modify the molecule under construction. The relative positions of molecules and catalysts are controlled by a stiff housing, and everything can be custom-built for a specific reaction pathway.
Obviously, this approach isn’t feasible for floppy molecules, because they won’t interact predictably with the tools. But to my completely untutored eye, none of this looks totally implausible–assuming you can attach the molecule in question to substrate and move it at constant speed past the tools, you’re looking at fairly ordinary chemistry under some very weird conditions. If your reactions were unreliable, you’d also need some sort of repeated reaction system similar to the one in Nanosystems 13.3.1.c. If you couldn’t do better than Freitas, et al.’s 20% deposition rate, you’d need a ridiculous number of repetitions.
Diamond, though, presents a special case for a number of reasons. First, it’s stiff enough to potentially allow mechanosynthesis of large molecules. Second, it’s a relatively tricky molecule to synthesize, given its nasty habit of surface reconstruction and the number of ways bonding can go wrong (as described, for example, in the C(110) paper by Freitas, et al). The average molecule of interest would be less stiff, but would also involve less-complicated frameworks of bonds to be made at each step (if section 8.6.1 of Nanosystems is to be believed).
It looks like a fully-operational version of Drexler’s nanofactory would essentially require the invention of a whole new branch of chemistry–many hundreds of reactions, reagents, catalysts–and a factory system for shipping everything around. On the other hand, it would only require one or two usable materials systems. And as Chris has frequently pointed out, there’s nothing special about any one reaction or design approach: all it takes is one viable solution for each problem. Of course, there’s hundreds or thousands of problems involved, and any one could prove extremely challenging to solve or work around.
Dr. Moriarty: In your earlier letter, you define mechanosynthesis as “the synthesis of molecular assemblies ¬? and ultimately bulk materials¬? from the mechanical positioning of reactive molecules with atomic precision,” and explicity assume it excludes electrical fields and tunneling effects. While this is a fairly accurate portrayal of how Drexler uses the term, I don’t see why any real-world design would be limited in this fashion. In Nanosystems 1.4.3.b, Drexler writes, “despite their likely utility, machine-phase electrochemical processes are mentioned only in passing.” He argued that 1992-era approximations of electrochemical effects were simply too unreliable to use such designs, hence the adorably baroque mechanical rod computers in chapter 12.
I agree the notion that “proof-of-principle” experiments for a few prototypical mechanosynthesis reactions would be highly desirable, as would further research into the behavior of the C(111) and C(110) surfaces under various conditions. For a reasonable investment, this research could say a lot about the feasibility of mechanosynthesis, and it could provide useful information on industrial diamond synthesis and perhaps a new AFM technique or two.
Dr. Jones: What’s the best way to order a copy of your book in the US? Studying Drexler’s designs, I’m pretty much convinced that any implementation pathway would involve fairly sophisticated use of self-assembling molecules in solution. Regardless of the relative merits of hard and soft systems, it’s clear that the soft systems can be built far more easily, and with less novel chemistry.
Eric, I think your comments on Drexler’s openness about the amount of donkey-work implied by Nanosystems are exactly right. Over the summer, in an email exchange with him, I stated my usual position, that I didn’t think that Drexlerian mechanochemistry wasn’t directly contrary to any law of physics, but that it, if possible, it would be much harder to implement than its proponents admitted. He came back with a riposte along the lines of “Which proponents? I never pretended it would be easy”. He was right, of course, but one can’t help feeling that he hasn’t been very active in correcting the opposite impression that is propagated by some of his supporters.
The US website of Oxford University Press says that Soft Machines is in stock – see http://www.oup.com/us/
Hi Eric,
Great to hear from you again and a Happy New Year to you too. Chris Phoenix and I have recently spent quite some considerable time discussing many of the issues you’ve raised. Chris has promised to put our entire discussion (which now comprises 6 letters and a considerable number of – at times, lengthy – e-mails) onto the CRN website. I would be very interested in your comments, particularly with regard to the discussions on feasibility/ success rates for the mechanochemistry. Chris and I have also discussed at length Freitas et al.’s proposal for diamondoid mechanosynthesis. (Note that there’s actually already quite a lot of published work on diamond surfaces (formed under different growth conditions) using a range of experimental characterisation techniques and theoretical tools).
“If you couldn‚Äôt do better than Freitas, et al.‚Äôs 20% deposition rate, you‚Äôd need a ridiculous number of repetitions.” Yes, you would! *And* Section 13.3.1c doesn’t discuss an “error correction” step. That is, if the molecule binds in the incorrect conformation or simply effectively irreversibly sticks to the tip/probe/finger/actuator (or whatever you want to call it!) after a failed attempt at deposition, repeated attempts to remove the molecule will have an excessively low success rate. Note that there’s an important proviso buried in Section 13.3.1.c: “..the formation of a bond between the probe and the site exposed by transfer of the group is among the acceptable outcomes, *** so long as that bond can be broken by the forcing retraction of the probe***” (emphasis is mine). There are quite a few comments pertinent to success rates and error correction in my most recent letter to Chris (…and I’m still awaiting a response to Comment 25 above!).
Best wishes,
Philip
Eric, to add to my earlier comments about the availability of Soft Machines – I’ve now had a report on the situation from the publisher. It’s out of stock in the UK, but the US warehouse does still have some copies, so you should be able to get it from the USA OUP website as I suggest above. A new printing is due to arrive in the UK warehouse any day now. The publisher blames unexpectedly large demand from the USA.
Folks – thanks for the wonderful brain chow, first off.
Second off – Dr Moriarty – Your arguements above seem to indicate that you’re targetting the concept of ‘general’ or ‘universal’ assemblers as being beyond reasonable, and make some very good arguements along those lines. However, does this doom mechanosynthetic nanotech, in your opinion?
Would it be ‘enough’ to be able to create carbon-carbon bonds in 3D, creating fullerines, diamondoid, and/or graphite type products? Would this or would this not be a reasonable initial goal, gaining a large amount of nanotechnological wet-dream capability in this way for a relatively minute number of chemical interactions that would need to be supported?
If this is ‘good enough’, what is your position on Merkle’s paper on mechanochemistry or ‘hydrocarbon metabolism’ (at http://www.zyvex.com/nanotech/hydroCarbonMetabolism.html)? (My apologies if I’ve missed reference(s) to your position previously, but I’ve not run across them in the reading I’ve done…)
Sincerely,
John B
Hi John,
Chris Phoenix and I have covered very many issues related to your post in our recent debate. This will be published in its entirety on “Soft Machines” in the near future. I’d very much like to hear your comments on the debate and if your questions haven;t been addressed by the material in the debate, I’d be more than happy to go into more detail.
Best wishes,
Philip
Fair enough, Dr Moriarty. Look forward to reading the material!
-John
I think something important should be pointed out here: When it comes to solid three dimensional consumer good items, from paper to swords, from chairs and cars, to clothing and houses, all we really need is a basic set of diamondoid mechanosynthesis systems, and then, using SURFACE CONTROL techniques, such as active color plates, or hinges and such in the right places, we can make almost any shape or texture, using diamond as the basic modular component. You can make “Diamondoid Wood” by giving the surface the right texture and color etc, and you can give it any color or pattern of colors on its surface. You do not *NEED* to be able to synthesize every actual molecule such as wood and cellulose and the rest.
Erin –
A point. However, diamond (and potentially other carbon materials – graphenes, buckytube, buckyball) does not do everything that you’d want objects to do. For instance, diamond is slightly unstable in ‘normal’ PVT conditions, buckytubes are vulnerable to atomic oxygen degredation, etc.
While diamondoid mechanochemistry, if it comes to be, would be wonderful in many ways, it won’t solve the ‘universal assembler’ problem. That is, there’ll still be lots of problems that need solutions using non-carbon materials.
However, it *may* be possible to use a carbon nanoassembler to make the tools to make the tools to do the job you want. IE – use diamondoid reaction vessels to handle chemical synthesis of other elements to make ‘food’ or other materials that aren’t just carbon. This is a non-trivial R&D project, however, on top of the already non-trivial R&D needed to get diamondoid mechanochemistry – or any other nanofactory/molecular mill capability – off the ground.
-John
On “universal assemblers”: The relevant section of EoC does not suggest that a single machine will be able to do almost any reaction. It suggests that for almost any reaction, it will be possible to develop a machine that can do it. The text of the section simply does not invoke the idea of a single “universal” assembler machine. That idea grew up afterward.
On material systems and reactions: Drexler has done some very interesting work over the last few months, that will be published over the next few months. Basically, he’s found a tool that does not leave dangling bonds on the tool when the deposited moiety is removed. Thus the transfer is far more energetically favorable than the Freitas/Merkle tool, and the dimer will reliably leave the tool when given a chance to move to diamond, graphite, or buckytube.
Chris
On error handling: I think all MM architectures assume that a molecular construction tool will be able to form several times its own mass of product before it either breaks or makes an uncorrected deposition error. And implicitly assume that reliably correcting a correctable error should not take too much time. I agree that Freitas’s tool is not suitable for this. But Drexler’s may be.
Chris
Chris,
First, I very much look forward to reading Drexler’s most recent work when it’s published. It’ll be interesting to see whether the error rate (in the calculations) is significantly lower than that in Freitas et al.’s work and, moreover, whether there is a useful strategy for “porting” the theoretical study to experiment.
I’m not certain that I agree with you on the “universal assemblers” point. You seemingly argue that Drexler suggests that we’d need a family of assemblers (one for each reaction). That’s not at all how I interpret the section in Engines of Creation which you cite. Drexler states that the assemblers “will be able to use as “tools” almost any of the reactive molecules used by chemists – but they will wield them with the precision of programmed machines”. This strongly suggests to me that an assembler will be able to pick up various tools to carry out different reactions. What you’re suggesting is much closer to the ‘molecular mill’ concept…
Nevertheless, even if a family of assemblers (one for each reaction) is to be used, the problem is that the parameter space is severely limited by the choice of material/ surface/ reactant. For example, high dangling bond densities are ‘verboten’ (so we need to use passivated surfaces), high diffusion barriers are required (ruling out a number of metal surfaces), directional covalent bonds are required..etc..etc.. Leaving aside our discussion of diamondoid mechanosynthesis for a while, the idea that one can construct a family of assemblers with each assembler dedicated to a particular reaction **for practically every element in the periodic table** is deeply flawed.
Philip
I completely agree with Philip that mechanosynthesis will not be able to make everything that can be made by more conventional types of chemical synthesis. And I also think that the design of a fully programable nano-factory becomes more difficult with every additional mechanosynthesis tool that is needed. The only way around this problem (that I see) is to make the molecule / nano-crystal by other methods then put it into a container that is compatible with the material inside and the diamond / graphite structure that you are building, then add the container to what you are making. This still is not a Universal Assembler but it does expand the possibilities.
A really good way to increase the range of products (e.g. reduce diffusion and spontaneous reconstruction) is to lower the temperature.
Also, you don’t have to restrict yourself to adding atoms; there’s nothing wrong with taking them off. For example, you could passivate a surface, removing H atoms just where you want to deposit C, and then passivating what you just deposited and moving to the next site where you remove H, deposit C, hydrogenate…
But I don’t think “universal” is important. Certainly the word and the concept have not been a part of nanofactory proposals. A nanofactory can be very flexible, even general-purpose, with just a few reactions and atom types. You can do a lot without ever getting into atoms heavier than chlorine (except perhaps as catalysts for mechanosynthesis in a narrow set of special-purpose molecules). Build a “toolbox” of nanoscale components, and simulate anything else. The toolbox might not need much more than actuators, wires, digital logic, bearings, structures, mechanosynthetic tools, and maybe photon stuff.
Chris