[HN Gopher] The Gambler, Maxwell's New Demon
___________________________________________________________________
The Gambler, Maxwell's New Demon
Author : dsr12
Score : 37 points
Date : 2021-02-27 12:42 UTC (2 days ago)
(HTM) web link (physics.aps.org)
(TXT) w3m dump (physics.aps.org)
| mrfredward wrote:
| There's an awesome introductory book on information theory which
| in one chapter resolves Maxwell's demon. The argument is that the
| energy needed to reliably send a single bit of information (which
| is needed for making the detector talk to the trap door) depends
| on the temperature of the fluid (because you have to overcome
| background noise), and it turns out the minimum energy needed for
| this communication is exactly equal to the amount that can be
| recovered from separating out a molecule.
|
| If my understanding is correct (admittedly a big if), it seems
| the same problem applies to gambling. The energy it takes to
| communicate whether a gamble has paid off is at least as big as
| what can be recovered from the gamble.
|
| The book is "An Introduction to Information Theory" by John R
| Pierce, and it's the most I've ever learned from a $10 paperback.
| alkonaut wrote:
| The resolution of the original "Paradox" of Maxwell's demon is
| that the demon is unable to perform the measurements required.
| The fallacy was that the measurements could be "free".
|
| Isn't this the same thing? If you measure your succcess and
| choose to stop, you expend energy in order to measure whether you
| have succeeded. If you measure after each particle, you must do
| lots of measurements for very little gain. If you measure rarely,
| you can gain a lot of energy but your expected gain is low
| because the particles will average out?
| ec109685 wrote:
| Isn't this the difference between N=1 and N=1000? If you run an
| experiment once, you could get lucky and by constraining the
| timeframe, you prevent the law of averages from catching up to
| you.
|
| However, try sufficient number of times or for sufficiently long,
| and you will face gambler's ruin.
| PeterisP wrote:
| The difference from the gambler's fallacy is that standard
| gambling systems do not have memory and keep giving the same
| random results, but the physical system of Maxwell's demon
| _does_ have an innnate trend towards equilibrium. In the game
| Maxwell 's New Demon is playing, the dice have memory - the
| more it has won, the less likely it is to win some more, and
| the more it has lost, the more likely it is to win in the
| future.
|
| So if the demon wants side A to be hotter than side B but
| currently there's a random fluctuation that's not in its favor,
| then unlike a random walk or a gambler in a coin-flip game the
| long-term expected value is not the current status but the
| equilibrium with equal temperature on both sides, so if
| measurements would be free, then this now demon should work.
| However, they aren't, and apparently the cost of knowing if
| it's winning (and thus should quit now) or losing (and thus
| should continue) is larger than the energy value of "the
| winnnings".
| bollorior wrote:
| Yes, this invention appears to resolve a fallacy simply by
| replacing with another fallacy, perhaps less well known to
| physicists.
| alex_young wrote:
| https://en.m.wikipedia.org/wiki/Martingale_(betting_system)
| lubesGordi wrote:
| I'm not sure this adds much to the original thought experiment.
| Correct me if I'm wrong, but the idea behind Maxwell's Demon was
| to illustrate the notion of adding information to the system to
| prevent/slow entropy (information being the decision to swing the
| trap door one way or the other). This article just seems to
| describe a physical device that sort of approaches that.
| thebigspacefuck wrote:
| Kenny Rogers's Demon
| pjmorris wrote:
| Because it knows when to hold 'em (molecules) and when to fold
| 'em.
| bluesign wrote:
| ' In general, the demon sets some threshold of performance (wins
| or losses) over a given time period that will tell it whether or
| not to stop. There's no unique best gambling strategy, but some
| are better than others. '
|
| Can someone explain this in a simple way?
| kikokikokiko wrote:
| I believe the device keeps looking at the sensors and checks to
| see how many times the electron jumped in the "correct"
| direction, if the device "won" an ammount of times that,
| according to their historical performance, is above a certain
| threshold, the proccess is stopped and the device will use the
| energy it gathered to do work. When all the energy was used,
| the proccess will be repeated. I have serious doubts about how
| many real "work" can such a device really do, when you take
| into consideration the energy necessary to keep the sensors,
| data keeping, and the "trapdoor" all working. It's like using a
| tab of how many green candles have appeared before on a stock
| price chart to decide when to sell, it's a temptative idea, but
| in the end your results always return to the mean. The 2nd law
| is an unavoidable beast.
| bluesign wrote:
| Yeah but it says there are some worse algorithms, is it
| because of the cost of taking an action or reset?
| bollorior wrote:
| As a probability theorist (who knows very little about
| physics) I understand this machine as relying on the well-
| studied and widely debunked martingale fallacy - that you
| can play until you are slightly ahead, and then stop. Then
| repeat from the start. Perhaps I am missing something.
|
| If I'm right, there are lots of ways of tweaking this
| system to make it appear more effective on different time
| scales and with different constraints. This is exactly what
| the folklore of 'gambling systems' looks like - although
| none can violate the laws of probability, they can be
| optimized to give a better appearance of doing so.
|
| Suppose I told you to go to Vegas, bet $1 on black at
| roulette. If you lose, bet $1 million on red. If you lose a
| second time, bet $1 trillion on black. You would probably
| recognize this as an unworkable strategy quite quickly. Now
| try another strategy: write the numbers 1, 2, 4, 4, 3, 7 on
| a piece of paper. Bet the sum of the first amount and the
| last amount in the list (on an even chance such as red or
| black). If you win, cross out both numbers. If you lose,
| write down sum of the two numbers (the amount you lost) at
| the end of the list. Repeat until all the numbers are
| crossed out. You're $21 up. Repeat as many times as the
| number of dollars you want, divided by 21. The second
| strategy has been highly optimized, not to beat casinos,
| but to give gamblers the strong sense that they are either
| guaranteed to win, or that they are coming very close to
| winning, except for some outlying bad luck.
|
| However, the two strategies are equally constrained by
| mathematical laws from making you infinitely rich with
| certainty.
| bob1029 wrote:
| This seems like a good intersection between information &
| physics. What is the fundamental limit of efficiency in
| measuring/mutating the state of the system, assuming nearly-ideal
| transducers and computation?
| piokoch wrote:
| The Second Law of Thermodynamics is a beast. As a PhD student I
| was given from time to time letters from people who claimed that
| they invented perpetual motion machine (University was obliged to
| provide answers for letters and obviously all the effort was
| thrown on poor PhD candidates).
|
| Some of those invention were breaking energy conservation
| principle - those were easy. But there were some really ingenious
| ideas how to construct perpetual motion machine that was breaking
| II Law of Thermodynamics in a rather tricky way. Proving
| incorrectness was a nightmare sometimes.
|
| Funny times!
| siver_john wrote:
| As an undergraduate in physics I was spared from any absurdity
| like that, but I knew professors got emails and the like for
| those types of things. However, the most amusing thing that I
| heard about was that our department (Physics and Astronomy)
| would get calls inquiring about their horoscope. Which caused
| us every year to change our name to Department of Psychics and
| Astrology on Halloween.
| currymj wrote:
| what university was this where there was an obligation to
| respond to letters from cranks??
|
| how did that policy get put into place, that's fascinating.
| jpmattia wrote:
| > _in a rather tricky way._
|
| Any particularly memorable ones?
|
| From my bag: A friend's husband collared me at an all-weekend
| outing with some crazy contraption idea of lawnmower engines &
| generators & other junk. It was a fairly long-winded
| description followed by a question of "why wouldn't this work?"
| My reply was simply: Because the gods are out to fuck you over.
| He just looked into the air and said: I can accept that. The
| wife (MIT educated): How come when I explain the 2nd law
| forbids it from working, you keep asking, but when he tells you
| the gods are out to fuck you over, you accept it?
|
| PS. What university requires answers for such letters?
| SilasX wrote:
| If I'm understanding this right, this version of Maxwell's Demon
| has the demon wait for random fluctuations between the chambers,
| and then close the gate when there's a gradient.
|
| So dumb question then, doesn't this have the same problem with
| the original, that you have to spend negentropy to measure one
| side and know if it's higher potential than the other?
| siver_john wrote:
| Yes, it still falls to the same fate. It just has less
| intervention so it goes to that fate slower which means you can
| recover more energy from the system. Is the basic idea.
| api wrote:
| I've wondered for ages if it might be possible to transfer energy
| as information.
|
| Consider a hypothetical setup in which two identical (possibly
| entangled) systems are run. On the sending side the system is
| observed and additional energy is spent in computation to compute
| interventions that would have (but did not) result in free
| energy. These interventions are then sent to the receiving side
| where the energy spent in computation on the sending side is
| partially recovered by actually performing those interventions
| and recovering free energy.
|
| If something like this or at least loosely analogous to it were
| possible then it would be possible to do something like put up a
| Dyson swarm in a close solar orbit around the sun and transmit
| back the energy not as dangerous and difficult to receive
| microwave beams but as data feeds containing an endless stream of
| "cheat codes" to obtain apparent (but not really) free energy at
| the receiving end.
|
| Energy is conserved because a given cheat code can only be used
| once to intervene where the sender determined an intervention
| _would_ have worked. More energy must be spent at the sending
| side than is obtained at the receiving side.
|
| Given that there are environments like close solar orbit where
| energy is stupidly abundant, it could work very well even if the
| efficiency of transfer were terrible. Even if you only got say 1%
| of sent energy, a small-ish Dyson swarm in an orbit between the
| sun and Mercury could easily power the Earth, a Mars settlement,
| and a few hundred spacecraft with ion thrusters.
| DennisP wrote:
| It's amazing that we're at a point where the next step after "I
| figured out an interesting variation of Maxwell's Demon" is to
| actually build the demon and try it out.
___________________________________________________________________
(page generated 2021-03-01 23:03 UTC)