Layering and physics• Rethink “everything” emphasizing layering as
the key concept (admittedly procrustean)• Connecting layered architectures with “layering”
(called coarse graining) in multiscale physics• Look for persistent sources of confusion• Highlight needs for clearer explanation of what
we already know• New theory is also needed for multiscale
physics, and progress is encouraging
1
PassiveLossless
• We’ve also been focusing on this theory. • Note that logically, the Venn diagram on the right holds • Reconciling this apparent contradiction is the challenge • Fluctuation-dissipation is first essential theorem
ActivePassive
PassiveLossless
• Classical statistical physics “explains” only this (badly).
Active
Passive
Lossless
• It would appear logically that the diagram on the left is equivalent to the Venn diagram below• So there is actually a nontrivial result here• As opposed to “what is SW” which is just pedagogical
ActivePassive
PassiveLossless
Active
Passive
Lossless
Lossless
Passive
Active
Finite time horizon
Infinite time
horizon
• Note that without active control, there is nothing that corresponds to what we call “cause”• As in, the “algorithm caused the robot to turn right”• So explaining to scientists that “algorithm caused” is what we mean by “cause”• While at the same time, SW only existing embodied in HW
Passive
PassiveLossless
Passive
Lossless
Lossless
Passive
Caution
• This is “deep” background• As is, not accessible or useful• Need deep experts to rethink how we
explain things we already know• There are edges of this that are research,
but the immediate need is pedagogical • Elements should go in immediate papers• Longer term issues are mixed in here
5
Big big picture• I want to ultimately argue that there are
essentially two flavors of “complexity” (and many subflavors, but deferring that for now…)
• The origins are physics vs engineering (or disorganized vs organized)
• Both have been successes in some respects and failures in other
• A key distinction is the role of “architecture”• Expanding on themes started in Alderson and
Doyle 2010
Systematic error/confusion in “new sciences”• The main idea is “emergent complexity from minimal tuned
random ensembles”• Architecture = graph topology• Dominates science and misapplication is main source of
errors• Big success story is the “modern synthesis” (not normally
thought of this way) in evolutionary biology• In physics, a standard recipe, vetted, refined, honed
– widely adopted in PhysRev, NatPhys, etc– allows great rhetorical scope– applicable everywhere (wrongly, and nowhere
correctly)• Ancillary errors from
– bad statistics, – logical errors (e.g. flipping if and only if), – emphasis on patterns (particularly superficial)
Systematic error/confusion in biology• The primary error is the same
– “emergent complexity, minimal tuned, random”– has dominated in the “modern synthesis”– evolution = small, random mutation plus selection– essential in Davrolis EvoArch
• New alternatives are radically different (better) – “Natural genetic engineering”– Savageau, Shapiro, Gerhard & Kirschner, Mattick… – Claim: Needs architecture/layering to make coherent sense
of collection of facts – Contrast with attempts to just tweak the old version
• No detail here, big a topic on its own, more elsewhere
Systematic error/confusion elsewhere
• What systems engineers know is poorly explained*• Available statistical tools are inadequate and don’t
reflect state of the art (from 50 years ago)• “Correct” theories are fragmented and incoherent• Even what constitutes “correct theory” is poorly
explained, conventional philosophy is weak• Notions of explanation, causality, mechanism,
emergence, etc etc are murky and incoherent• Multiscale and layered systems not explained
* engineers apparently have a long tradition of secrecy
SoftwareHardware
Apps
OS
Libs, IPC
kernel
DigitalAnalog
ActivePassive
ClassicalQuantum
LumpedDistributePassive
Lossless
• Start with this cartoon• Probably badly done as is• Believe this is important, but• Needs clear explanation• But of things • We thoroughly understand now• Except at the very bottom
SoftwareHardware
Apps
OS
Libs, IPC
kernel
DigitalAnalog
ActivePassive
ClassicalQuantum
LumpedDistributePassive
Lossless
• Need coherent view of layering• Turing focus on analog and up.• Physics has a coherent, consistent view that varies from confused to wildly wrong • Must ultimately redo physics all the way down• For now, understand it’s limitations• Clearly explain what we already know
Issues
SoftwareHardware
Apps
OS
Libs, IPC
kernel
DigitalAnalog
ActivePassive
ClassicalQuantum
LumpedDistributePassive
Lossless
Of course, a consequence of good
layering is that you can only indirectly know what is going
on below the layer in question. (This does
recurse…) Makes reverse engineering
challenging.
SoftwareHardware
AppsOS Libs, IPC
kernel
DigitalAnalog Active
Passive
ClassicalQuantum
LumpedDistributeWhat are
the right cartoons?
Software
Hardware
AppsOS
Libs, IPCkernel
DigitalAnalog
ActivePassive
?
?
Modularity of digital hardware
What are the right cartoons?
Software
Hardware
AppsOS
Libs, IPCkernel
DigitalAnalog
ActivePassive
Layers up here
are very different
from layers down here
This needs clearer exposition
Software
Hardware
AppsOS
Libs, IPCkernel
DigitalAnalog
Layers here are “stacked” and nonintersecting, a more familiar kind of modularity
Whereas• SW is X of HW• Digital is X of Analog
What is “X”?State, organization, large/thin…???
Need better nomenclature
Software
Hardware
AppsOS
Libs, IPCkernel
DigitalAnalog
ActivePassive
Laye
rs h
ere
from layers here
are very different
Drawn a different way
I’d be thrilled with a coherent explanation of this. (Sloman and VMs is a start.)
Software
Hardware
AppsOS
DigitalAnalog
New idea: Turing style?
Maybe start from here with Turing’s 3 step research:1. hard limits, (un)decidability
using standard model (TM)2. Universal architecture
achieving hard limits (UTM)3. Practical implementation in
digital electronics
Maybe start from here with Turing’s 3 step research:1. hard limits, (un)decidability
using standard model (TM)2. Universal architecture
achieving hard limits (UTM)3. Practical implementation in
digital electronics
Essentials:0. Model1. Universal laws2. Universal architecture3. Practical implementation
Software
Hardware
DigitalAnalog
Software
Hardware
Apps
OSLibs, IPC
kernel
Digital
Analog
Active
Passive
Laye
rs h
ere
from layers hereare very different
• Can this be explained by differences in the nature of scope?
• In applications, scope is named, logical, functional, semantic, …
• In hardware/resources, scope is addressed, physical,
• OS kernel is the “waist” between the two
Important questions
ActivePassive
ClassicalQuantum
LumpedDistributePassive
Lossless
The essence of multiscale
physics
PassiveLossless
• We’ve also been focusing on this theory. • Note that logically, the Venn diagram on the right holds • Reconciling this apparent contradiction is the challenge • Fluctuation-dissipation is first essential theorem
ActivePassive
PassiveLossless
• Classical statistical physics “explains” only this (badly).
Active
Passive
Lossless
Repeat for emphasis:• These two diagrams express logical relations that are superficially contradictory• Theory is needed to reconcile this• Standard StatPhys story is at best murky, at worst wrong• Our approach is working and should fix this, but is just a baby step (so far)
ActivePassive
PassiveLossless
Active
Passive
Lossless
• These two pictures illustrate the essential challenge• Not sure how to draw them to highlight this…
ActivePassive
PassiveLossless
Active
Passive
Lossless
PassiveLossless
… and underscore the difference with the physics view
PassiveLossless
Note:In our theory, “highly organized” and extreme nonlinearity play an essential role in active devices, and hence in life and technology.
ActivePassive
PassiveLossless
In physics, even mild nonlinearity is synonymous with chaos, while “highly organized” and active devices are not treated at all.
ActivePassive
PassiveLossless
In physics, even mild nonlinearity is synonymous with chaos, while “highly organized” and active devices are not treated at all.
“emergent, far from equilibrium, Prigogine, etc”
ActivePassive
PassiveLossless
These are extremely
different, and need to make
this clear.
Note:In our theory, “highly organized” and extreme nonlinearity play an essential role in active devices, and hence in life and technology.
Us: Stochastic models are a convenience, the result of natural and unavoidable approximations, and are explained mechanistically
PassiveLossless
Passive
Lossless
Our theory is also different at this level, while there are not obvious experimental consequences, the differences show up later in other layers.
Them: Stochastic models are assumed a priori and never “explained” except with vague notions of “chaos”(This is perhaps a minor flaw here but will make things much worse higher up.)
Our theory:Idea is that lossless are dense in passive
PassiveLossless
Passive
LosslessApproximation arbitrarily good on finite (but arbitrarily long) time horizons.
High dimensional lossless circuit passive
Looks
Really lossless
power supply
ActivePassive
Active
Passive
active
Looks
Really passive
Our theory: Active requires “hidden” power supply and nonlinear circuitry
Approximation arbitrarily good on finite (but arbitrarily long) time horizons.
power supplyactive
LooksReally passive
High dimensional lossless circuit passive
LooksReally lossless
• Both approximations arbitrarily good on finite (but arbitrarily long) time horizons.• Both require finely tuned (highly organized) circuits
• Biology and technology= active/passive circuits• Condensed matter physics = passive/lossless gases, …
• Note: fine tuning for (not vs.) robustness • Completely unlike standard physics• Many unresolved issues (e.g. fine tuning here?)
High dimensional lossless circuit passive
LooksReally lossless
Standard physics • Takes infinite time and complexity limits a priori• Takes random ensembles a priori• No other “tuning” required!
• Extensions: phase transitions, criticality, chaos everywhere, scale-free, SOC, edge of chaos, …
• Big (wrong) idea: All complexity is emergent from random ensembles with minimal tuning
We have been using lumped analog systems here, but there are two opposite directions to head in:1. Digital2. Distributed
ActivePassive
PassiveLossless
1. Digital: I think we can do much of this story using CAs to boolean nets to TMs. Easier to understand and math is almost trivial
2. Distributed: Natural direction to connect with physics and QM
ActivePassive
PassiveLossless
“emergent, far from equilibrium,
Prigogine, etc”
ActivePassive
PassiveLossless
“highly organized” with extreme nonlinearity
Huge gap
Can we illustrate this with both automata and lumped circuits (ODEs)?
(Later do distributed/PDE/QM)
power supplyCactive
LooksReally passive
New idea inspired by Deacon
Aim to connect with “dissipative” systems (Prigogine) ideas.• How to distinguish tornadoes from airplanes from birds?• Random circuits from designed circuits from digital?• Deacon’s “morphodynamic” but too much is grouped here• What does this look like if we can “look inside”?• Play with this in the next few slides.
CactiveLooks
Really passive
power supply
Look inside Passive too
PassiveLossless
ActivePassive
PassiveLossless
Thermo-dynamic
RandomMorpho-dynamic
?
?Analog
Active
BiologicalTeleo-
dynamicDeacon has these 3 kinds of systems
“emergent, far from equilibrium, Prigogine, etc”
PassiveLossless
ActivePassive
PassiveLossless
Thermo-dynamic
RandomMorpho-dynamic
DesignedMorpho-dynamic
ActivePassive
PassiveLossless
SoftwareHardware
Apps
Libs, IPC
kernel
DigitalAnalog
Active
EngineeredTeleo-
dynamic
?
?Analog
Active
BiologicalTeleo-
dynamicNeed to
distinguish these
BiologicalTeleo-
dynamic
Probably need to distinguish these
bacteria
eukaryotes
animals
mammals
primates
humans
PassiveLossless
ActivePassive
PassiveLossless
Thermo-dynamic
RandomMorpho-dynamic
DesignedMorpho-dynamic
ActivePassive
PassiveLossless
Need to distinguish these
Statistic physics
“non-equilibrium”
Engineered
Huge gap
Passive
Lossless
Active
Passive
Passive
Lossless
Thermo-dynamic
RandomMorpho-dynamic
DesignedMorpho-dynamic
Active
Passive
Passive
Lossless
SoftwareHardware
Apps
Libs, IPC
kernel
DigitalAnalog
Active
EngineeredTeleo-
dynamic
?
?Analog
Active
BiologicalTeleo-
dynamic
Need to distinguish these
Huge gap