That seems to be an argument for the uselessness of the distinction between P-with-large-exponent and not-P, not for the uselessness of having a reduction of NP to P.
Also, it's worth noting that, if one is a theoretical computer scientist, then there is more of interest in the study of computing than what can be done in the universe. (I am a mathematician, and am, as I suspect most of my colleagues, serenely unconcerned with the applicability or non- of my work.)
And the parent is pointing out that if P=NP, we already have a construction.
So (if his opinion was presented accurately), Knuth is wrong about a positive proof being non-constructive.
---
I believe there are already known problems whose best solution is O(n^k) for arbitrarily high k. So yeah, naturally not everything will be practically tractable.
A good argument can be provided by an analysis of statistical thermodynamics gas laws. It highly disturbs some people to think that atoms being in motion implies that all the oxygen atoms in the room air might coincidentally cluster into a small corner of the room leaving the victim to suffocate. There is no reason to think that impossible under most thermodynamics theories. However, it is nearly infinitely unlikely and you'd merely have to wait a ten to the hundred times the lifespan of the universe for it to coincidentally happen. I suppose it scales quite strongly with room volume and linearly with air pressure, you have one atom in the ISS airlock and being in one half the air lock at any given moment is a very predictable 50:50 odds...
Likewise if the smallest polynomial solution to a NP problem takes a google's worth of universes to calculate, then even if it is poly you need not worry too much about mere human inconvenience.
I'm not saying its a wise way to go thru life, but there exists a large set of people who get hung up on whats convenient or beautiful must be right, even if whats right turns out to be very ugly or inconvenient, so for them "P=NP but your bank account is still secure" is a very big deal.
That said, this algorithm does not solve the decision problem because it can never return a no. Only a yes if there is one to be found.
Donald Knuth: (Retired from Stanford) It will be solved by either 2048 or 4096. I am currently somewhat pessimistic > by either 2048 or 4096
It's an odd statement for a mathematician/computer scientist. It's equivalent to "by 4096", surely.Because of some hand wavy string theory calculation completed in 2030, either the universe doesn't exist as it is measured to exist (or whatever McGuffin you'd like as a bullet proof excuse) or there exists a Turing machine that outputs a proof of P=NP then halts and the fastest way to find out if it halts is to run it and we try and it never stops. Maybe a century later someone figures out a lower bound on the number of steps the machine will make before it halts and its larger than the computational power of the lifetime of the universe, even very optimistically. Or a ruleset is proven to exist but the ruleset is at minimum a number larger than the number of particles in the universe so its kinda difficult to construct or emulate. Or the infinitely long Turing tape is merely proven to contain a number of states equal to a trillion times the number of atoms in the universe.