From: Robert J. Bradbury (bradbury@aeiveos.com)
Date: Wed Nov 27 2002 - 00:14:05 MST
I'd love to have some more time right now to devote to this thread
but unfortunately that isn't the case.
But here are some minimal comments on things said by a number of
people. I'm sorry in advance for any possible misattributions.
To discuss this topic from a moderately informed perspective
you have to start with Chapter 3 of Nanosystems. Chapters 4-8
wouldn't hurt as well. (These are most of the chapters that
make my eyes glaze over but I've gotten through most of them
at one time or another.
For a simpler background on Molecular Mechanics, the NIH
guide is good:
http://cmm.info.nih.gov/modeling/guide_documents/molecular_mechanics_document.html
For those seeking something more detailed the Colby College
Molecular Mechanics Tutorial is fairly detailed (at 78 pages...):
http://www.colby.edu/chemistry/CompChem/QUANTAtutor99.pdf
Now, as Eugene has pointed out, if one is doing mechanosynthesis
in a vacuum or inert atmosphere the complexity added by molecules
in a solution (the solvation complexity) is removed.
Mez made a passing comparison with protein and/or drug design modeling.
It isn't good for two reasons -- the first is the complexity of the
solution (cells are not simply some proteins and a few water molecules).
Drug design modeling is generally grossly simplified with respect to
what is *really* in the cytosol. Has anyone ever run a simulation with
what is *actually* in the cytosol?
The second and more important oversight (IMO) is the degrees of freedom issue.
Proteins are *particularly* flexible (most of the atoms are not bonded to 3 or 4
other atoms in such a way that their rotational freedom is severely
constrained) -- that isn't a property of most diamondoid -- this is one
reason the protein folding problem is so hard computationally (too many
degrees of freedom). The fact is that you don't generally want those
degrees of freedom in diamondoid (that's what makes it strong(!) *lots* of
covalent bonds) is one of the reasons modeling it is so much less
computationally taxing. [Its what lets Goddard's group spin Eric's
molecular bearings in a computer at a GHz (or some similar jaw-dropping
number) until they fly apart.] [Now whether you can assemble a nanopart
with all of the covalent bonds that you desire is likely to be another
very deep computational and design question. {more below}]
I think Eugene may have raised the point about GROMACs and/or a paper
involving simulations of metals or semiconductors -- [I'm not completely
positive here]. I don't think this is a good comparison -- you have to
consider things like band gaps, the motion of electrons and holes, much
more significant temperature and crystal lattice distortion effects, etc.
It is precisely because you eliminate some of these "problems" that
semiconductor manufacturers are moving to things like SOI. Now there
may be future technologies -- diamondoid-electronic or diamondoid-metallic?
where these things will be important but for most classical diamondoid
nanoparts these can be ignored from a simulation perspective.
Mind you the fact that nanopart designs may be a smaller computational
problem *eventually* may not help us initially since Chapter 16 (esp.
Table 16.1) of Nanosystems specifies a path through biotech to nanotech.
Whether Zyvex (or maybe IBM) can pull a rabbit out of the hat and bypass
the intermediate stages remains to be seen. [If not we will have to
wrestle with the Folding Orc and several future offspring.]
There are two critical papers (other than Eric's work) that have to be
read to have an informed discussion re: nanoassemblers (IMO). They are:
Hall, J. S., "Architectural considerations for self-replicating manufacturing systems",
Nanotechnology 10(3):323-330 (September, 1999).
http://www.foresight.org/Conferences/MNT6/Papers/Hall/index.html
and to a lesser extent
Merkle, R. M., "Casing an assembler", Nanotechnology 10:315-322 (1999).
http://www.zyvex.com/nanotech/casing.html
It is worth noting that Josh's paper does make some rather eye-popping
claims about what nano-assembly lines may be capable of (in contrast
to Eric's general-purpose assembler approach). [Hint: Special purpose
tools can be much more efficient than general purpose tools.] It also
blows a big hole in the idea that general purpose self-replication is
an essential component of a nanotech "vision".
Now, I think I've advanced the path Eric has outlined towards a nanoassembler
a little bit with my "Protein Based Assembly of Nanoscale Parts" essay.
http://www.aeiveos.com/~bradbury/Papers/PBAoNP.html
I've at least taken a whack at the complexity and cost issues (which *are*
not trivial) which I don't believe anyone else has done.
My conclusions from the process of writing that paper are I don't think
the *real* problems will lie in the atomic or sub-atomic simulations
(what Mez may be asserting) or in the "swarm of assembler" complexity
(Anders?). The simulation problem I think will be minimized
by the limits on the degrees of freedom (for atomic motion) in "quality"
nanoscale parts and the swarm problem can be dealt with if you build
some minimal intelligence/cooperative ability into the assemblers and
let them communicate with a higher authority for anything they can't handle.
(Swarms of ribosomes manage to do quite well with very little "intelligence"
or communication ability).
I think the really tough nut to crack is going to be design-for-assembly.
It took us 20+ years to crack the problem of how to teach the computers
to wire the chips themselves -- [you don't think humans could wire 100 million
transistors together do you?]. Its going to be the same problem all over
again with nanopart designs -- you may have a nanopart design, you can simulate
it until the cows come home and it works just great -- but you can't find a
way to put it together using the existing assembler technology. Now the
assembler technology will improve over time -- so something you couldn't
assemble in 2040 is a piece of cake for a 2050 assembler. But I strongly
suspect there will be a class of nanoparts that are exceedingly beautiful,
run on the simulations just fine that can never be assembled nohow noway.
We will henceforth label these Escher Class Nanoparts (ECNs).
The problem with this discussion is that we don't, at this time, have a
clue as to what fraction of the nanopart phase space consists of ECNs.
I suspect that there will be ample time in the future for Lord of the Rings
class epic adventures by scientists determined to manufacture ECNs.
(I can see Anders designing another role-playing adventure here...
much to my dismay.)
Perhaps this is what the SIs in the universe devote their attention to.
It certainly (to me) seems an appropriate balance of capabilities and
problems of rather large difficulty.
Robert
This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:58:26 MST