[Date Prev][Date Next][Thread Prev][Thread Next]   [Date Index] [Thread Index] [Author Index

Re: Maximally constrained



In the 1986 paper, Rosen drew a clear distinction between holonomic and non-holonomic constraints. Imposing holonomic constraints to a system reduces the number of configuration variables (and their corresponding velocity variables) necessary to describe a system. Imposing a non-holonomic constraint does not allow a reduction in the number of configuration variables, but it does allow a reduction in the number of velocity variables. With each added non-holonomic constraint, the number of independent velocity variables become fewer - the other velocity variables have now become determined by the configuration variables via the constraints. With a sufficient number of such non-holonomic constraints imposed, there are no more independent velocity variables - they are now all a function of configuration.

As Howard quoted from Rosen:
> Rosen (p. 112): "If we impose the maximal number of nonholonomic
> constraints, then the velocity vector is uniquely determined by
> the configuration. The result is an autonomous dynamical system,
> or vector field, on the configuration space. At this point, the
> impressed forces of conventional analytical dynamics disappear
> completely; their only role is to get the system moving."
A "maximally constrained system" is thus one in which there is a sufficient number of non-holonomic constraints imposed to create this condition. So, it is "maximal" in the sense that this certain number of constraints allows a system to eliminate the maximum number (namely, all) of the independent velocity variables. On the other hand, this certain number of constraints is "minimal" in the sense that it only takes this number of constraints to achieve this condition - adding more such constraints are not required to eliminate the already-eliminated independent velocity variables.

So, whether it is "maximal" or "minimal" seems to me to be a "glass half-full or half-empty" question. Rosen appear to me to have chosen the term "maximal" to indicate the condition of maximally eliminating the independent velocity varables. In my view, such a  number of constraints could also be regarded as an optimal number insofar as it achieves this condition without adding excess constraints.

As for "totally constrained system", Howard quotes:
> Rosen  (p. 114): "As an example of a totally constrained system,
> of the type we have discussed in the previous section [the
> section from which I quoted above], is a neural network, or more
> generally, a special purpose digital computer, considered as hardware."
The quote (in section IV), refers to "previous section", which would be section III "Constraints and Controls" rather than (as Howard indicates) section II "Constraints in Mechanical Systems". Section III discusses the notion of how a maximally constrained system could be controlled by parameters, restoring degrees-of-freedom from inside the system. The summary on of section III on p. 113 [ital. orig.]:
    "We can think of the specification of the "forcing function" Beta(x,t) [i.e., parameters that are a function of configuration and time - TG] as a program, superimposed on what is a maximally constrained mechanical system. Thus, these systems are programmable, even though maximally constrained. It is this factor which renders such systems "plastic", and which will be of crucial import to us in our subsequent discussion of machines, cells, and brains.
    We should note explicitly that these "forcings" have little to do with the mechanical impressed forces of physics, which reflect themselves in accelerations. These "forcings" instead manifest themselves in velocities, with impressed forces held fixed. Thus, the "plasticity" of maximally constrained systems, which we have described, has a quite different epistemological basis than the behaviors of conventional mechanical systems."
So, it could be that by "totally constrained system", Rosen meant: a maximally constrained system which also has degrees-of-freedom restored to it by virtue of such parameters. Or, it could simply be that "totally constrained system" is a synonym for "maximally constrained system". In either case, it was indeed a term he used, and it involves, and depends centrally on, the definition of a "maximally constrained system".
 

 
Howards main objection to Rosen's paper:
HP: Both Morowitz and Rosen have drawn a mistaken conclusion from this example. They both miss the absolutely essential distinction between rate-independent (non-dynamic, holonomic) memory constraints and rate-dependent (dynamic, nonholonomic) control constraints. This is the conceptually equivalent distinction that von Neumann makes between descriptive (quiescent) constraints and constructive (nonholonomic, dynamic) constraints, that ironically, Rosen also misinterpreted.
 
As I see it, Rosen's paper does not miss that distinction. As noted at the top of this post, Rosen precedes his discussion of maximally constrained systems by laying out the differences between holonomic and non-holonomic constraints in some detail, including their respective mathematical ramifications for a system's analytical description.
 
I don't think Rosen dwells on the holonomic constraints because they are not central to what is required in order to achieve a maximally constrained system. That is, holonomic constraints do not contribute to the property of making all the independent velocity variables become dependent solely upon configuration variables.
 
However, I see nothing in his argument which precludes that such a system could also possess holonomic constraints in addition to the required non-holonomic ones. It seems to me that non-holonomic constraints could be imposed on holonomically constrained portions (e.g., rigid 'chunks of particles') of the system just as readily as they could be imposed on individual 'particles', and could still result in a maximally constrained system. The resulting system would indeed inherit properties due to those holonomic constraints along with properties inherited from its non-holonomic constraints. I just don't see Rosen's math or verbal arguments as precluding this possibility; I just think that it is not central to his focus in that paper.
 
 
Howard also states:
Unlike program-controlled computers, neither folding nor enzyme catalysis is a "maximally constrained" process as Rosen states. Innumerable rate-dependent, parallel impressed natural physical forces play an essential role. In fact, quite the contrary, the one-dimensional sequence constraints can be viewed as the MINIMUM number of constraints necessary to achieve the enzyme?s specific rate-controlling function.
This is very unclear to me. As I read this, if folding depended upon "impressed physical natural forces", then for a protein to fold reliably into the same shape every time would mean that the outside world must always be imposing the exact same forces each and every time, which is not reasonable at all to expect. Additionally, such a situation would negate the proposal that "one-dimensional sequence constraints can be viewed as the MINIMUM number of constraints...". Instead, the outside world (via impressed forces) would determine the resultant conformation. Conversely, if the "one-dimensional sequence constraints" were "the MINIMUM number of constraints..." then impressed forces could not "play an essential role" in determing the final folded shape.
 
Even if we view the primary sequence as a constraint, it is insufficient by itself to describe how the folding will occur. It only says to me that: the primary sequence is the minimum number of constraints given the forcing (and other) properties that inhere in and among the constituents of this sequence. To me, omitting this last italicized portion simply hides the necessary non-holonomic constraint relationships between aspects of the sequence which do determine how a given sequence will fold. Informationally, having to specify only a primary sequence does seem evolutionarily efficient, but such information implicitly relies upon the aforementioned non-holonomic constraints for it to be meaningful in any way. Rosen on p. 123 [ital orig]:
    "These questions are part of a larger one; namely, that most system descriptions do not contain sufficient "information" to allow realization at all. For instance, from a specific knowledge of kinetic parameters (e.g., rate constants, turnovers, numbers, specificities, and the like) we cannot pass to the primary (or even tertiary) structure of an enzyme which manifests these parameters. It appears that only very special kinds of descriptions, which we may call blueprints, allow themselves to be realized. The primary structures of an enzyme, or a sequence of DNA or RNA, exemplify such blueprints, but these in turn are very special descriptions; they do not generally allow us to pass to other modes of description. Thus, if I am given the primary sequence of a protein, I can in general deduce nothing about its functional or kinetic properties."
Regards,
Tim


> -----Original Message-----
> From: ROSEN Forum [
mailto:***]On Behalf Of Howard
> Pattee
> Sent: Sunday, October 31, 2004 8:30 AM
> To: ***
> Subject: Re: Maximally constrained
>
>
> Judith wrote:
> I believe Howard Pattee has misinterpreted what Robert Rosen
> wrote about maximally constrained systems. In order to understand
> why this is so, we have to discuss how my father defined and used
> the terms "maximally" and "constrained".
>
> HP: I was commenting only how Rosen actually uses the phrase
> "maximal constraint" in the 1986 paper in question. (I am aware
> of his later usages, as in Life Itself, where he lumps local
> constraints with natural laws) This is a well-written paper so
> there is not much room for misinterpretation. In the 1986 paper
> he explicitly uses Hertz's definitions of constraints, the ones
> accepted by all physicists. Also, by "maximally constrained" I
> see no way he could mean "optimally constrained" as Judith wants
> to interpret him. I simply quote the relevant paragraph:
>
> Rosen (p. 112): "If we impose the maximal number of nonholonomic
> constraints, then the velocity vector is uniquely determined by
> the configuration. The result is an autonomous dynamical system,
> or vector field, on the configuration space. At this point, the
> impressed forces of conventional analytical dynamics disappear
> completely; their only role is to get the system moving."
>
> HP: This is a fairly technical statement for which very little
> ambiguity is possible. If the forces of analytic dynamics (i.e.,
> forces resulting from the laws of nature) disappear except to
> provide the energy to get the system moving then the motion must
> be entirely the result of the constraints. It is in this sense
> that the system is maximally constrained. This completely defines
> the system. There is no room for optimality to play a role here.
>
> Judith: A "maximally constrained system" is not a "totally
> constrained system" or a system with the maximum number of
> constraints. Instead, it is a system with such constraints as
> maximize its potential.
>
> HP: Again, I simply quote Rosen. He equates "maximally
> constrained system" with "totally constrained system." He argues
> that any "potential" must arise from programmability that is a
> special type of constraint. In any case, no amount of word
> processing by Judith can interpret Rosen's concept of  "maximal
> constraint" to mean the same as my interpretation of "minimal constraint."
>
> Rosen  (p. 114): "As an example of a totally constrained system,
> of the type we have discussed in the previous section [the
> section from which I quoted above], is a neural network, or more
> generally, a special purpose digital computer, considered as hardware."
>
> You will also see that this 1986 paper is entirely consistent
> with Rosen's discussion of nonholonomic constraints in the
> Appendix of Anticipatory Systems (p. 414 and note 1, p. 427). In
> fact, he wrote the paper as an elaboration of this too brief Appendix.
>
> If you actually read the entire paper you will see that Rosen is
> using the same concept of nonholonomic constraint that I use to
> distinguish the laws (forces of analytical dynamics) from the
> informational constraints of genetic memory. We had discussed the
> concept of nonholonomic constraint at great length, so I know
> there was no intent to disagree with me. In fact, he was using my
> views to introduce his discussion. That is why I said he only
> misinterpreted what I had in mind. I fully understand how I could
> easily be misinterpreted, because in 1986 things were not as
> clear to me as they are today. I had not yet clearly stated the
> evolutionary principle of "minimum constraint" in writing. I had
> proposed only a vague "principle of optimal constraint." With
> more thought about natural selection it only made sense that life
> uses the forces of nature as much as possible and the
> informational constraints only where necessary for survival. The
> universal life processes of folding and self-assembly are prime
> examples of the creative potential of using the fewest
> constraints to control the largest universe of lawful dynamics.
>
> Howard