DrChinese said:
There is no violation of the uncertainty principle when the momentum of 2 entangled particles are measured precisely. The uncertainty principle applies primarily when an attempt is made to deduce conjugate observables, such as momentum and position, on a SINGLE particle with high precision (i.e. low standard deviation). Further:
a) You can know the total momentum of 2 particles precisely, and that does NOT mean they are entangled.
b) They ARE entangled if you know the total momentum of 2 particles precisely, but do not know their individual momenta. Usually you might start with situation a), and then allow the 2 particles to interact so that their individual momenta change.
Indeed, in the original EPR example of a decaying particle the two resulting asymptotic free particles are represented by a wave function, where both the total momentum is quite well defined (in the center-of-mass frame it's 0) and the relative position is quite well determined to be at a distance ##L## large compared to the range of the interaction between the particles. The wave function is thus described by something like
$$\psi(\vec{x}_1,\vec{x}_2)=\phi_{\epsilon}(\vec{x}_1-\vec{x}_2-\vec{L}) \int_{\mathbb{R}^3} \mathrm{d}^3 P \tilde{\psi}_{\epsilon'} (\vec{P}) \exp[\mathrm{i} \vec{P} \cdot \vec{R}/hbar]$$
with ##\vec{R}=(m_1 \vec{x}_1+m_2 \vec{x}_2)/(m_1+m_2)##.
Here ##\phi_{\epsilon}## is a position wave function sharply peaked with some small width ##\epsilon## around 0, i.e., ##\vec{r}=\vec{x}_1-\vec{x}_2## is very likely around ##\vec{L}##, and ##\tilde{\psi}_{\epsilon'}## is a momentum-space wave function sharply peaked around 0 too, such that the total momentum is very likely around 0. Of course the center-center-of-mass position ##\vec{R}## is very indetermined (uncertainty relation between ##\vec{P}## and ##\vec{R}## components: ##\Delta R_j \Delta P_j \geq \hbar/2 ##) and the momentum of the relative motion ##\mu \dot{\vec{r}}## is also very indetermined. That ##\vec{P}## and ##\vec{r}## are both very well determined is no contradiction to the uncertainty relation of course, because ##\hat{\vec{P}}=\hat{\vec{p}}_1 + \hat{\vec{p}}_2## and ##\hat{\vec{r}}=\hat{\vec{x}_1}-\hat{\vec{x}}_2## compute, i.e., are compatible observables.
The entanglement is such that although each ##\vec{x}_1## and ##\vec{x}_2## (the positions of each of the particles) is also pretty indetermined, but if you measure the position of one particle pretty precisely, you also know the position of the other particle pretty precisely due to the fact that ##\vec{r}## is known pretty precisely in the EPR state. Then of course the momenta of either particle are still pretty indetermined.
The same holds for the individual particle momenta, which are also both pretty indetermined, but when measuring one of the particles momenta very accurately you also know the other particle's momentum pretty accurately to be just the opposite momentum, because the total momentum is prepared to be well determined in this EPR state. Then of course also the position of either particle is still pretty indetermined.
There is no paradox, as soon as you accept the statistical meaning of the quantum state. The choice of which single-particle observable you measure accurately decides which observable of the other particle is then also well determined (if you measure momentum of particle 1 accurately, you know the momentum of particle 2 also accurately, but neither particle's positions and vice versa). The correlations in case of the one or the other measurement are due to the preparation in this entangled state and there's no need of an action at a distance influencing one particle by measuring the other at a far distant place.