I am not sure that there were any legal fees. According to her blog post, Ms. Seidel was represented by the First Amendment team at Public Citizen. Perhaps Public Citizen should be the ones recovering some of the expenses? In any case they should be congratulated for the win!
1. "High T_c" is a technical term. Indeed, 55 kelvin is "high" (though not as high as the record for cuprates). You have to compare it with the typical T_c for metals (a few kelvin). The difference is between liquid helium temperatures and liquid nitrogen temperatures (which cuprates have reached already and perhaps the new compounds also will).
2. More improtantly, this will ignite a "firestorm of research". You see, we don't have a good model of high T_c superconductivity (unlike the BCS model for metals). Having several different superconducting systems will help theorists isolate the significant features of the system from the less significant ones.
3. Seeing superconductivity in a totally new material is exciting. This is interesting basic research even if today we dont' have a practical application. If we don't do the research we'll never get to the practical stage.
The black hole itself is, indeed, black for all intents and purposes. However, matter falling into the black hole (but still outside the horizon) heats up as it accelerates, emitting thermal radiation, typically in the X-ray spectrum. Thus one talks about "brightness", the brightness of the region right around the black hole.
An illustrative example: for an outside observer, the "temperature of the sun" can mean the temperature of the part one sees, that is the surface temperature (roughly 6000 kelvin). This is not the same as the core temperature of the sun (roughly 1.5x10^7 kelvin).
Warning: that expression does not quite meant what it seems to. The "timeline" under discussion here is from our point of view as light from that area arrives here, after about 26,000 years. On the other hand, that doesn't quite mean that the events actually happened "26,300 years ago" -- there's no good global notion of time that is applicable here.
Well, you put the jury in the courtroom, you live with the vote of the jury. Me, I'm not sure what to think of verdicts which don't include a written explanation of the evidence and the reasoning. I know this sounds like heresey to common-law natives, but in my line of work, if you can't produce a coherent written (or at least, oral) account of your reasoning, then it can be presumed that your opinions aren't reasoned.
While it's true that high-tech patents last too long, technology is also a field where many inventions are easy to implement once you know about them. Without patents, the R&D you do is automatically shared by the whole industry -- which reduces your motivation to do R&D.
Regarding extensions, a major point of the patent system (as opposed to making everything a trade secret) is to encourage everyone to publish their inventions so others can build up on them. Patents (in theory) only cover actual inventions, not ideas, and indeed you can patent a modified version of an existing invention.
A bigger problem today is with idea patents like "business methods" and software patents, but the real problem is that "obviousness" has ceased to be a meaningful limitation on patentability. As courts rule that the idea of "I'll do things the way I always did them, but the calculations will be done by computer and not by hand" is patentable, or that it is an original idea that a course-management system should accomodate users which are students and TAs at the same time, patents become a problem. I'd like to ban software patents in general, but I don't think it makes good sense. Most software patents are bogus because they are obvious, but let's say you invent a new faster algorithm for multiplying matrices. This costs money to develop, and the basic rationale behind patents leads me to think that you should be able to patent this algorithm. Of course, if someone can improve the algorithm -- they would gain.
This calculation illustrates a good point about the difference between asymptotic analysis of algorithms and real-world implementation of the same algorithsm. Computer science defines "efficient" as "bounded polynomially in terms of the input size". In practice, even if polynomial has a small degree (like a cubic) it already means that the resource rquirements are very large. Theory and practice are only the same in theory.
It seems there is one: TeX the world. Unfortunately, you'd need to replace the equation delimiter: they use [;;] where I used $ $. The actual equations can remain the same.
The "membranes", "modes" and "frequencies" here are already a physical analogy. Number theorists study objects (``automorphic forms'' -- no matter why they are called this way) that live on some ``manifolds'' (no matter what that means, either). But to get some intuition you can replace ''manifold'' with ''taut membrane'' (like a drum) and ''automorphic form'' with ''normal mode'' a.k.a. basic ''standing wave'', as you call it. An important problem in mathematical physics is to find what are the possible frequencies of standing waves on a particular surface. The problem here is analogous.
To see a picture of the 2-dim membrane I was talking about, see here. Start by taking a half-infinite strip of width 1, and cut off a semi-circular bit at the bottom like in the picture (the strip extends infinitely far at the top. Next, glue the two infinite sides together so the strip becomes a cylinder. Finally (that's not in the picture) imagine that as you go further and further up the cylinder, its radius becomes smaller and smaller, so the real thing is a kind of infinite funnel.
To see what a standing wave on this membrane looks like, see here (this was computed numerically by Dennis Hejhal).
The "lift" that takes a standing wave on this space to a standing wave on the 5-dim space is really complicated (and is a very indirect construction). There just isn't a non-technical way to describe it.
However, we know what the "lift" does to the frequencies: if you start with a standing wave you found numerically, and approximately know its frequency, then you know there will be a lifted guy of a calculatable frequency on the 5-dim space. So the interesting problem is to find standing waves with frequencies which are different from the ones we already know about (because we have calculated a lot of standing waves on the 2-dim surface).
One symmetry this infinite funnel has is left-right reflection (it is apparent both in the picture of the strip and in the picture of the vibrational mode). The other symmetries are difficult to describe in a blog post. What's important is that the modes of vibration must respect the symmetries.
It is true that to each such ''standing wave'' (on the 2-dim surface, on the 5-dim space, and on others) there is an associated L-function. The Riemann Hypothesis for these L-function (the same formulation: all zeros are on the critical line) is called the "Generalized (or Grand) Riemann Hypothesis" or GRH.
It was possible to calculate a few zeros of the newly-found modes, and see that indeed they are where they are supposed to be. This gives some evidence for the GRH. Calculations like this can always falsify the GRH (by finding a zero off the line). However, these calculations don't represent any progress toward proving the GRH -- that was confusion on part of the person who submitted the story to slashdot.
Now what the heck are degrees of those L-functions?
This is where things get technical. The Riemann Zeta-function $\zeta(s) = \sum_n n^{-s}$ has the Euler product representation $\zeta(s) = \prod_p \left( 1 - p^{-s}\right)^{-1}$. Similarly, the Dirichlet L-functions $L(s;\chi) = \sum_n \chi(n)/(n^s)$ have the Euler product $\prod_p L_p(s;\chi)$ with $L_p(s;\chi) = 1/( 1 - \chi(p)/(p^s))$. In both cases, the factor at each prime $p$ takes the form $1 / ( 1 - a(p)/p^s )$, for some number $a(p)$ depending on $p$. We think of this factor as a the inverse of a polynomial of degree 1 in the variable $p^(-s)$ (the polynomial is $P(T) = 1 - aT$).
Similarly, to GL(3) Hecke-Maass forms such as the ones computed by Booker and Ce Bian, there is an attached L-function $L(s;f)$ which can be represented as an Euler product, $\prod_p L_p(s;f)$. This time, however, the local factors $L_p(s;f)$ are the inverses of cubic polynomials, that is $1/L_p(s;f)$ takes the form $P(p^-s)$ where $P(T) = 1 - aT - bT^2 - cT^3$ for some coefficients $a,b,c$ depending on $p$ (and on $f$, of course). This is why we call it an L-function (or Euler product) of degree 3.
Using the Fundamental Theorem of Algebra, it is common to factor the polynomial $P(T)$, and write it in the form $\prod_{j=1}^{3} ( 1 - \alpha_j(p) T)$. Thus an Euler product of degree $d$ takes the form:
For "modes of vibration of the membrane" read "joint eigenfunctions of the ring of invariant differential operators on the locally symmetric space". I hope you find that clearer.
I don't know if somebody asked this before: "writing Hamlet contributed to find|build|formulate what?" ?
Seriously, numerically computing "automorphic forms" (whatever they are) contributes to understanding these objects, for example for formulating conjectures about them and for testing known conjectures. This is good for analytic number theory, but at the moment has few applications outside mathematics. Doing mathematics is important because it is a triumph of human intellect, not because it has any practical application (or perhaps you also judge Shakespeare by the practical application of his writings?). In some cases, mathematics has applications to everyday life (modern cryptography is one). In some cases we have yet to find applications.
Actually, we can tell quite easily -- but sometime we get so excited we rudely keep trying to explain anyway. In any case, you are right that because it is so technical, this story doesn't belong on this forum.
Actually, what the RH tells us about the distribution of prime numbers is be pretty useless regarding RSA. To get anywhere you need the Extended Riemann Hypothesis (covering Dirichlet L-functions) and even the full force of the "Generalized Riemann Hypothesis" (covering all automorphic L-functions) is not known to help with the really important problem here -- factoring.
Booker and Ce Bian constructed certain degree 3 L-functions, but it is best to think of their discovery as follows: there is a complicated 5-dimensional membrane known to mathematicians as "SL(3,Z)\SL(3,R)/SO(3)". This membrane has subtle number-theoretic symmetries, so that its modes of vibration encode number-theoretic information. These modes (and their vibrational frequencies) are being extensively studied, but they are very transcendental objects so they cannot be written down explicitely and must be computed numerically. While certain modes (and frequencies) were already known numerically (they can be constructed from vibrational modes of the 2-dimensional membrane "SL(2,Z)\SL(2,R)/SO(2)" via something known as the Gelbart-Jacuqet lift) we now have the the first numerical computation of "native" modes of the 5-dimensional membrane -- those that aren't related to lower-dimensional cases. To each such mode of vibration there is an associated "L-function (similar to the Riemann zeta function), and it is the L-functions that were constructed. In fact, verifying that the approximate L-functions that were found correspond to actual modes of vibration is not easy (in the 2-dimensional case there is important work of Booker with others about this).
In short, this is an important advance in automorphic forms, but it is so technical that it doesn't belong on SlashDot.
It is important to realize that while indeed there is a ("Generalized") Riemann Hypothesis associated to these L-functions, numerically computing them represents zero progress toward proving the Riemann hypothesis for these L-functions or the original Hypothesis for the Riemann zeta function. At most this will allow very approximately computing some of their zeros and thus a weak check on the GRH for these L-functions.
"All possible execution paths" would be too many to enumerate. You simply can't. Especially with software that has a non-trivial UI allowing many options, has support for disabled users etc. Instead, the software should incorporate intenral consistency checks so that it knows at all times that it is in a good state, or can immediately flag an error. For example, if a "Republican" ballot is being displayed, then the voting machine should not have a green light next to the "Democrat" ballot selection button. That this is possible tell you that they did something wrong.
The ability to resolve the spectral lines is a great advance. The hype is getting old. We know that small "organic" molecules are not hard to come by; we expect to find them everywhere we look, yet the press goes ape whenever we find them. Worse, since we have no idea what "life" is, it seems a bit odd to go crazy over molecules that are somewhat related to our kind of "life".
There exists individual computer programs such that humans cannot prove whether they terminates or not. For example, the program that enumerates all proofs in Peano arithmetic looking for a proof of a contradiction. This is an aspect of Gödel incompleteness.
Mathematically speaking, proving a program correct from the source code is in generaly impossible (if you could do that you could, in particular, solve the halting problem). From the software engineering perspective it's true that examining the source code gives you greater confidence in the software than just black-box testing.
Actually, threatening Felten with a lawsuit is quite effective. Now actual lawsuit needs to be filed. Since this would be consulting work, not research, he would not be covered by Princeton's Lawyers and will have to fend for himself, which is quite expensive -- so the threat of litigation might be enough to deter him. New Jersey, being a state, cannot be sued without its consent.
Here's one idea: store the low-order bits of the fingerprint on the card, in plain text, and get the high-order bits from the scanner. More technically, use error-correction ideas: store some information on the card such that, given a fuzzy fingerprint allows you to correct errors in it, but that by itself doesn't give the fingerprint. Your question should be phrased as follows: "how many bits on information can be reliably recovered from a fingerprint?". I'm sure the answer is positive; but I agree it's not obvious that the the answer is large enough for a strong encryption key.
It sounds to me like more people trying to claim intellectual property of something that they did not come up with themselves.
Wrong. The journal is not "claiming" any "intellectual property". The journal is saying that, if you want them to publish your work (which no-one is forcing you to do) then you must assign them the copyright. If you don't like it, publish in a different journal. Since the journal makes money from subscription, they don't want you to benefit from their prestige by getting the paper accepted, and then turning around and posting the content somewhere else so no-one has to subscribe to the journal. Also note that in any case we're only talking about copyright, and hence the text of the paper, not the scientific content.
That said, I think the policy is silly. First of all, APS journals will already accept material that's already been posted on the arXiv (compare with Science and Nature which only take stuff that's never been presented before, even in a seminar talk). All the journal needs is a license from the authors. There's nothing wrong with the authors giving the journal an exclusive license to publish the article journal-style, as long as the authors retain the ability to post works derived from the article in other fora.
Slashdot Mirror
I am not sure that there were any legal fees. According to her blog post, Ms. Seidel was represented by the First Amendment team at Public Citizen. Perhaps Public Citizen should be the ones recovering some of the expenses? In any case they should be congratulated for the win!
1. "High T_c" is a technical term. Indeed, 55 kelvin is "high" (though not as high as the record for cuprates). You have to compare it with the typical T_c for metals (a few kelvin). The difference is between liquid helium temperatures and liquid nitrogen temperatures (which cuprates have reached already and perhaps the new compounds also will).
2. More improtantly, this will ignite a "firestorm of research". You see, we don't have a good model of high T_c superconductivity (unlike the BCS model for metals). Having several different superconducting systems will help theorists isolate the significant features of the system from the less significant ones.
3. Seeing superconductivity in a totally new material is exciting. This is interesting basic research even if today we dont' have a practical application. If we don't do the research we'll never get to the practical stage.
The black hole itself is, indeed, black for all intents and purposes. However, matter falling into the black hole (but still outside the horizon) heats up as it accelerates, emitting thermal radiation, typically in the X-ray spectrum. Thus one talks about "brightness", the brightness of the region right around the black hole.
An illustrative example: for an outside observer, the "temperature of the sun" can mean the temperature of the part one sees, that is the surface temperature (roughly 6000 kelvin). This is not the same as the core temperature of the sun (roughly 1.5x10^7 kelvin).
Warning: that expression does not quite meant what it seems to. The "timeline" under discussion here is from our point of view as light from that area arrives here, after about 26,000 years. On the other hand, that doesn't quite mean that the events actually happened "26,300 years ago" -- there's no good global notion of time that is applicable here.
Well, you put the jury in the courtroom, you live with the vote of the jury. Me, I'm not sure what to think of verdicts which don't include a written explanation of the evidence and the reasoning. I know this sounds like heresey to common-law natives, but in my line of work, if you can't produce a coherent written (or at least, oral) account of your reasoning, then it can be presumed that your opinions aren't reasoned.
While it's true that high-tech patents last too long, technology is also a field where many inventions are easy to implement once you know about them. Without patents, the R&D you do is automatically shared by the whole industry -- which reduces your motivation to do R&D.
Regarding extensions, a major point of the patent system (as opposed to making everything a trade secret) is to encourage everyone to publish their inventions so others can build up on them. Patents (in theory) only cover actual inventions, not ideas, and indeed you can patent a modified version of an existing invention.
A bigger problem today is with idea patents like "business methods" and software patents, but the real problem is that "obviousness" has ceased to be a meaningful limitation on patentability. As courts rule that the idea of "I'll do things the way I always did them, but the calculations will be done by computer and not by hand" is patentable, or that it is an original idea that a course-management system should accomodate users which are students and TAs at the same time, patents become a problem. I'd like to ban software patents in general, but I don't think it makes good sense. Most software patents are bogus because they are obvious, but let's say you invent a new faster algorithm for multiplying matrices. This costs money to develop, and the basic rationale behind patents leads me to think that you should be able to patent this algorithm. Of course, if someone can improve the algorithm -- they would gain.
This calculation illustrates a good point about the difference between asymptotic analysis of algorithms and real-world implementation of the same algorithsm. Computer science defines "efficient" as "bounded polynomially in terms of the input size". In practice, even if polynomial has a small degree (like a cubic) it already means that the resource rquirements are very large. Theory and practice are only the same in theory.
It seems there is one: TeX the world. Unfortunately, you'd need to replace the equation delimiter: they use [; ;] where I used $ $. The actual equations can remain the same.
- The "membranes", "modes" and "frequencies" here are already a physical analogy. Number theorists study objects (``automorphic forms'' -- no matter why they are called this way) that live on some ``manifolds'' (no matter what that means, either). But to get some intuition you can replace ''manifold'' with ''taut membrane'' (like a drum) and ''automorphic form'' with ''normal mode'' a.k.a. basic ''standing wave'', as you call it. An important problem in mathematical physics is to find what are the possible frequencies of standing waves on a particular surface. The problem here is analogous.
- To see a picture of the 2-dim membrane I was talking about, see here. Start by taking a half-infinite strip of width 1, and cut off a semi-circular bit at the bottom like in the picture (the strip extends infinitely far at the top. Next, glue the two infinite sides together so the strip becomes a cylinder. Finally (that's not in the picture) imagine that as you go further and further up the cylinder, its radius becomes smaller and smaller, so the real thing is a kind of infinite funnel.
- To see what a standing wave on this membrane looks like, see here (this was computed numerically by Dennis Hejhal).
- The "lift" that takes a standing wave on this space to a standing wave on the 5-dim space is really complicated (and is a very indirect construction). There just isn't a non-technical way to describe it.
- However, we know what the "lift" does to the frequencies: if you start with a standing wave you found numerically, and approximately know its frequency, then you know there will be a lifted guy of a calculatable frequency on the 5-dim space. So the interesting problem is to find standing waves with frequencies which are different from the ones we already know about (because we have calculated a lot of standing waves on the 2-dim surface).
- One symmetry this infinite funnel has is left-right reflection (it is apparent both in the picture of the strip and in the picture of the vibrational mode). The other symmetries are difficult to describe in a blog post. What's important is that the modes of vibration must respect the symmetries.
- It is true that to each such ''standing wave'' (on the 2-dim surface, on the 5-dim space, and on others) there is an associated L-function. The Riemann Hypothesis for these L-function (the same formulation: all zeros are on the critical line) is called the "Generalized (or Grand) Riemann Hypothesis" or GRH.
- It was possible to calculate a few zeros of the newly-found modes, and see that indeed they are where they are supposed to be. This gives some evidence for the GRH. Calculations like this can always falsify the GRH (by finding a zero off the line). However, these calculations don't represent any progress toward proving the GRH -- that was confusion on part of the person who submitted the story to slashdot.
I hope this helps.This is where things get technical. The Riemann Zeta-function $\zeta(s) = \sum_n n^{-s}$ has the Euler product representation $\zeta(s) = \prod_p \left( 1 - p^{-s}\right)^{-1}$. Similarly, the Dirichlet L-functions $L(s;\chi) = \sum_n \chi(n)/(n^s)$ have the Euler product $\prod_p L_p(s;\chi)$ with $L_p(s;\chi) = 1/( 1 - \chi(p)/(p^s))$. In both cases, the factor at each prime $p$ takes the form $1 / ( 1 - a(p)/p^s )$, for some number $a(p)$ depending on $p$. We think of this factor as a the inverse of a polynomial of degree 1 in the variable $p^(-s)$ (the polynomial is $P(T) = 1 - aT$).
Similarly, to GL(3) Hecke-Maass forms such as the ones computed by Booker and Ce Bian, there is an attached L-function $L(s;f)$ which can be represented as an Euler product, $\prod_p L_p(s;f)$. This time, however, the local factors $L_p(s;f)$ are the inverses of cubic polynomials, that is $1/L_p(s;f)$ takes the form $P(p^-s)$ where $P(T) = 1 - aT - bT^2 - cT^3$ for some coefficients $a,b,c$ depending on $p$ (and on $f$, of course). This is why we call it an L-function (or Euler product) of degree 3.
Using the Fundamental Theorem of Algebra, it is common to factor the polynomial $P(T)$, and write it in the form $\prod_{j=1}^{3} ( 1 - \alpha_j(p) T)$. Thus an Euler product of degree $d$ takes the form:
For "modes of vibration of the membrane" read "joint eigenfunctions of the ring of invariant differential operators on the locally symmetric space". I hope you find that clearer.
I don't know if somebody asked this before: "writing Hamlet contributed to find|build|formulate what?" ?
Seriously, numerically computing "automorphic forms" (whatever they are) contributes to understanding these objects, for example for formulating conjectures about them and for testing known conjectures. This is good for analytic number theory, but at the moment has few applications outside mathematics. Doing mathematics is important because it is a triumph of human intellect, not because it has any practical application (or perhaps you also judge Shakespeare by the practical application of his writings?). In some cases, mathematics has applications to everyday life (modern cryptography is one). In some cases we have yet to find applications.
Actually, we can tell quite easily -- but sometime we get so excited we rudely keep trying to explain anyway. In any case, you are right that because it is so technical, this story doesn't belong on this forum.
Actually, what the RH tells us about the distribution of prime numbers is be pretty useless regarding RSA. To get anywhere you need the Extended Riemann Hypothesis (covering Dirichlet L-functions) and even the full force of the "Generalized Riemann Hypothesis" (covering all automorphic L-functions) is not known to help with the really important problem here -- factoring.
Booker and Ce Bian constructed certain degree 3 L-functions, but it is best to think of their discovery as follows: there is a complicated 5-dimensional membrane known to mathematicians as "SL(3,Z)\SL(3,R)/SO(3)". This membrane has subtle number-theoretic symmetries, so that its modes of vibration encode number-theoretic information. These modes (and their vibrational frequencies) are being extensively studied, but they are very transcendental objects so they cannot be written down explicitely and must be computed numerically. While certain modes (and frequencies) were already known numerically (they can be constructed from vibrational modes of the 2-dimensional membrane "SL(2,Z)\SL(2,R)/SO(2)" via something known as the Gelbart-Jacuqet lift) we now have the the first numerical computation of "native" modes of the 5-dimensional membrane -- those that aren't related to lower-dimensional cases. To each such mode of vibration there is an associated "L-function (similar to the Riemann zeta function), and it is the L-functions that were constructed. In fact, verifying that the approximate L-functions that were found correspond to actual modes of vibration is not easy (in the 2-dimensional case there is important work of Booker with others about this).
In short, this is an important advance in automorphic forms, but it is so technical that it doesn't belong on SlashDot.
It is important to realize that while indeed there is a ("Generalized") Riemann Hypothesis associated to these L-functions, numerically computing them represents zero progress toward proving the Riemann hypothesis for these L-functions or the original Hypothesis for the Riemann zeta function. At most this will allow very approximately computing some of their zeros and thus a weak check on the GRH for these L-functions.
"All possible execution paths" would be too many to enumerate. You simply can't. Especially with software that has a non-trivial UI allowing many options, has support for disabled users etc. Instead, the software should incorporate intenral consistency checks so that it knows at all times that it is in a good state, or can immediately flag an error. For example, if a "Republican" ballot is being displayed, then the voting machine should not have a green light next to the "Democrat" ballot selection button. That this is possible tell you that they did something wrong.
The ability to resolve the spectral lines is a great advance. The hype is getting old. We know that small "organic" molecules are not hard to come by; we expect to find them everywhere we look, yet the press goes ape whenever we find them. Worse, since we have no idea what "life" is, it seems a bit odd to go crazy over molecules that are somewhat related to our kind of "life".
There exists individual computer programs such that humans cannot prove whether they terminates or not. For example, the program that enumerates all proofs in Peano arithmetic looking for a proof of a contradiction. This is an aspect of Gödel incompleteness.
Mathematically speaking, proving a program correct from the source code is in generaly impossible (if you could do that you could, in particular, solve the halting problem). From the software engineering perspective it's true that examining the source code gives you greater confidence in the software than just black-box testing.
Actually, threatening Felten with a lawsuit is quite effective. Now actual lawsuit needs to be filed. Since this would be consulting work, not research, he would not be covered by Princeton's Lawyers and will have to fend for himself, which is quite expensive -- so the threat of litigation might be enough to deter him. New Jersey, being a state, cannot be sued without its consent.
Flash and Silverlight are fully documented, and there exists free implemenetations: Gnash and Moonlight, respectively.
Not implementing the industry standard while putting in their own competing product would have serious anti-trust implications.
My reply was conjecturing that such things should exist ... but you know they do ...
Here's one idea: store the low-order bits of the fingerprint on the card, in plain text, and get the high-order bits from the scanner. More technically, use error-correction ideas: store some information on the card such that, given a fuzzy fingerprint allows you to correct errors in it, but that by itself doesn't give the fingerprint. Your question should be phrased as follows: "how many bits on information can be reliably recovered from a fingerprint?". I'm sure the answer is positive; but I agree it's not obvious that the the answer is large enough for a strong encryption key.
Wrong. The journal is not "claiming" any "intellectual property". The journal is saying that, if you want them to publish your work (which no-one is forcing you to do) then you must assign them the copyright. If you don't like it, publish in a different journal. Since the journal makes money from subscription, they don't want you to benefit from their prestige by getting the paper accepted, and then turning around and posting the content somewhere else so no-one has to subscribe to the journal. Also note that in any case we're only talking about copyright, and hence the text of the paper, not the scientific content.
That said, I think the policy is silly. First of all, APS journals will already accept material that's already been posted on the arXiv (compare with Science and Nature which only take stuff that's never been presented before, even in a seminar talk). All the journal needs is a license from the authors. There's nothing wrong with the authors giving the journal an exclusive license to publish the article journal-style, as long as the authors retain the ability to post works derived from the article in other fora.