[frevd]

No, if I understood right, this thing is a quantum computer solving NP-complete problems in linear time. And the article claims to have a simulation doing the same, but they are not sure whether this is actually only a special case scenario.

[gus_massa]

It's not a "quantum computer". It's an analog computer instead of a binary computer. It can be simulated with a normal (binary) computer using floating point numbers (remember that the amoeba has a finite precision).

(Form time to time there are articles that try to explain why analog computers are better than binary computers. Usually they assume infinite precision, that is impossible in a real system.)

The article claims that the time is linear, but it use quadratic space. So if you campare thy to simulate the algorithm with a big system in a sequential computer you will get probably cubic run time.

A better way to understand this is that the amoeba uses a good heuristic to solve the TSP in a small case. There are many heuristics out there, so it would be nice to implement this heuristic and compare with all the other one is some kind of standardized example set. It's more easy to find a good solution in a small system.

(Under the hood, the amoeba is a quantum system. But if you consider this a quantum computer then your phone is also a quantum computer.)

[frevd]

Yeah I guess in order to make this into a simulation, one would effectively add all the exponential time back, since it requires all that to "run" an amoeba on a classical computer. In that article they mention the organism communicating across its body, i.e. reacting to stimuli in all parts in a correlated fashion, that is the bit referring to a quantum system. Of course all physical systems are quantum systems (if I actually had a phone that one too yes, although the operating system does not make use of its nature). So there is no question whether we can simulate that algorithm, it cannot be possible without calculating all the parts, and that is no longer linear time.

[gus_massa]

> In that article they mention the organism communicating across its body, i.e. reacting to stimuli in all parts in a correlated fashion, that is the bit referring to a quantum system.

The internal communication inside the cell is classical, not quantic (classic, like your phone). I guess it use some kind of internal hormone, but it may be some signal that propagates in the cell membrane. IANAB.

You can't have a good quantum correlation in something that is as big as a cell (unless you freeze it at ridiculous low temperature that are not posible for now and would kill the cell anyway, or you have a more ordered system like a extremely pure crystal, or you only want some correlation for ridiculous small amount of time).

You can have big entangled systems, but they look more like pair of very clear optic fiber, not like a cell with a lot of water and crap moving randomly inside.

There are also some interesting "big" quantum effects in molecules like chlorophyll. (I'm not sure if they are 100% confirmad yet.) But a molecule of chlorophyll is much much much smaller than the cell and the effect is very short lived.

[Filligree]

It found a solution. Not an optimal solution.

[sano1]

Similar to the way my NN (brain) can solve the TS problem by just guessing the best route. Sure that's <i>a</i> solution. edit:typo

[whatshisface]

On Hackernews you can use asterisks to produce italics, like in markdown.

[frevd]

Right, it is not guaranteed to find _the_ best solution possible, but at least it does find _some_ solution to NP-complete problems in linear time, which no classical algorithm is expected to be able to do. Now they claim in that article to have made a simulation of this running on a classical computer exhibiting the same characteristics, which frankly should not be possible. so where is the catch?

[gus_massa]

You can always find a solution to the TSP. All cities are connected with all cities, so you can travel the cities in the alphabetical order (or the order in list). It's (usually) a very bad solution that is much longer than the best solution.

There are good heuristic, specially if the distances are not chosen at random but are the distance between points in the plane. So it's posible in some cases to find quite good solutions.

The problem is NP-complete only is you ask for the best solution.

[adrianN]

You can't say anything about the runtime of amoeba because they only solve a finite number of instances and you have no accurate model that would allow you to say something about the asymptotic behavior.