I completely agree with the claims in the OP (
Edit: See my comments in post #15), and I think it's a very good question. I've been thinking about these things off and on for several years now, and I still haven't found a really satisfactory answer. As far as I know, there are no answers to be found in the literature either. These things are completely ignored in physics books. I even looked for answers in a couple of "philosophy of science" books, but I didn't find anything.
Mathematical structures like smooth manifolds are obviously "pure mathematics". (To me this means that they have mathematical definitions, nothing more, nothing less). Physics on the other hand, is science, so a theory of physics must be falsifiable. To be falsifiable, it has to make predictions about results of experiments. So if we're going to use a piece of mathematics (like a smooth manifold with a metric and stress-energy tensor that satisfy Einstein's equation) in a theory of physics, we need a set of rules that tells us how to interpret the mathematics as predictions about results of experiments.
These rules are definitely
not pure mathematics. They are statements about real-world objects. They can't be derived from the mathematics, and must be postulated. The fact that a theory isn't falsifiable without an adequate set of rules means that the rules should be considered part of the definition of the theory.
It might seem that it should be straightforward and easy to write down a set of rules that tells us how to interpret the mathematics of GR as predictions about the results of experiments, but it's not easy for
any theory. In GR, I think the most obvious statement to include in the rules is this one:
A clock measures the proper time of the curve in spacetime that represents its motion.
I suppose that this should be preceded by a few statements that tell us that motion is represented by timelike curves in spacetime, and explains the idealizations we make (e.g. when we pretend that a clock is a point particle).
There's a major problem with statements like the one above. What's a "clock"? We can't define it mathematically, because that would completely mess up what we're trying to do, which is to specify how to interpret the mathematics as statements about the real world. This can't be done by eliminating references to the real world.
So the best we can do is to define the term by writing a set of instructions that tells you how to build the type of device that we would like to call a "clock", and say that what you get when you follow these instructions is called a "clock". This is where we run into another major problem: The best set of instructions that we can write today can't be understood by someone who doesn't already understand something about relativity
and quantum mechanics!
This looks circular, but there's a way to avoid circularity. I'll quote myself from another thread.
Fredrik said:
I think the only way to describe this process is in terms of a hierarchy of theories. [...] You start with the definitions of the purely mathematical parts of a collection of theories (say pre-relativistic classical mechanics, SR and QM). Postulate a correspondence between mathematical observables and measuring devices in any way you can. You can e.g. define the term "clock" by a describing an hourglass or something, and define a "second" by saying that it's the time it takes a certain amount of sand to run through. A few such definitions is enough to define "version 1" of pre-relativistic classical mechanics and start using it to make predictions.
Experiments will show you that you're on the right track. So now you have a reason to believe that the theory says something useful. One of the things it tells you is that the swings of a pendulum take roughly the same time. So you redefine a "second" to be the time it takes a specific pendulum to swing away and back, and you define "version 2" of the theory with the term "clock" defined by a description of how to build a pendulum clock. This way you can continue to define new versions of the theory, each one more accurate than the previous version.
You do the same to your other theories, including QM. At some point, you will see that to go from version n to version n+1 of pre-relativistic classical mechanics, you will have to use a version of QM(!), because it's the predictions of (some version of) QM that justify the new definition of a second that we're going to use in version n+1 (a statement about radiation emitted from a cesium-137 atom). At this point we define the term "clock" by a description of how to build an cesium clock, and we won't be able to do that without using earlier versions of several theories, including pre-relativistic classical mechanics and QM.
So the process of refining the correspondence between mathematical observables and measuring devices involves a large number of steps, and it's also clear that theories aren't refined in isolation from each other.
Fredrik said:
To go from the nth level in the hierarchy to the (n+1)th, you just write down a set of instructions on how to built a (n+1)th level measuring device that can be understood and followed by someone who understands the nth level theories and has access to nth level measuring devices. There's nothing circular about this.
The rule stated above tells us how to use clocks (real-world objects) to measure proper time (a mathematical property of a curve). At the very least, we need a second rule that tells us how to use some kind of length measuring device to measure the proper length of spacelike curves. If we e.g. decide to use rulers, we have to deal with the practical problem that rulers are deformed by acceleration and gravity. In principle, we can attach accelerometers to many different parts of the ruler, and make our length measurement axiom a statement about how to use a ruler with a bunch of accelerometers glued to it to measure proper lengths in situations where the accelerometers all read zero.
But what's an accelerometer? You could e.g. measure acceleration by attaching clocks to opposite ends of a hollow cylinder made of a solid material and having them exchange information using light signals inside the cylinder. If they stay synchronized, we're not accelerating (in that direction at least). But this raises a number of new issues, including how to determine if your hollow cylinders
would be straight if they were in free fall in a region with negligible tidal forces, or if they just look straight right now because they're under the influence of acceleration and/or gravity.
As you can see, this is quite complicated, and one must be very careful to avoid circularity.