Is the Uncertainty Principle valid for information about the past? My layman understanding of the Uncertainty Principle is that you can't determine the both the position and momentum of a particle at the same point in time, because measuring one variable changes the other, and both cannot be measured at once.
But what happens if I measure a charged particle with any number of detectors over a period of time? Can I use a multitude of measurements to infer these properties for some point in the past? If not, how close can we get? That is, how precise can our estimate be?
 A: The uncertainty principle should be understood as follows: The position and momentum of a particle are not well-defined at the same time. Quantum mechanically, this is expressed through the fact that the position and momentum operators don't commute: $[x,p]=i\hbar$. 
The most intuitive explanation, for me, is to think about it in terms of wave-particle duality. De Broglie introduced the idea that every particle also exhibits the properties of a wave. The wavelength then determines the momentum through $$p=\frac{h}{\lambda}$$ where $\lambda$ is the De Broglie wavelength associated with the particle. However, when one thinks about a wave, it is clear that the object described by it will not be easy to ascribe a position to. In fact, one needs a specific superposition of waves to create a wave that is essentially zero everywhere except at some position $x$. However, if one creates such a wave packet, one loses information about the exact wavelength (since a wave with a single, well-defined wavelength will simply extend throughout space). So, there is an inherent limitation to knowing the wavelength (i.e. momentum) and position of a particle. On a more technical level, one could say that the uncertainty principle is simply a consequence of wave-particle duality combined with properties of the Fourier transform. The uncertainty is made precise by the famous Heisenberg uncertainty principle,
$$\sigma_x\sigma_p\geq \frac{\hbar}{2}$$
More generally, for two non-comuting observables $A$ and $B$ (represented by hermitian operators), the generalized uncertainty principle reads
$$\sigma_A^2\sigma_B^2\geq \left(\frac{1}{2i}\langle [A,B] \rangle\right)^2\ \implies \ \sigma_A\sigma_B \geq \frac{|\langle [A,B]\rangle| }{2}$$
Here, $\sigma$ denotes the standard deviation and $\langle\dots\rangle$ the expectation value. This holds at any time. Therefore, the measurement occurring right now, having occurred in the past or occurring in the future has nothing to do with it: The uncertainty principle always holds. 
A: Found this great article (https://arxiv.org/abs/0906.1605) researching this topic myself. The short answer is no, the uncertainty principle doesn't apply to the past in the same way it applies to present measurements. It is possible to "update" our best guess for a particle's position and momentum made at time $t_0$ at some later time $t_1>t_0$ as we make new measurements. The new measurements inform the likelihood that a specific set of conditions at $t_0$ were present beyond the normal limit of the uncertainty principle, but it's important to recognize that some uncertainty always remains (makes me think of the Truman show; no matter how real and consistent your version of events appears, can you really be sure it's not all a massive conspiracy or coincidence?). It seems probable that there is some formulation or extension of the uncertainty principle that bounds this historical knowledge, but I've been as of yet unable to find it. Might be worth a paper if there isn't one already.
