Sie sind auf Seite 1von 23

Thoroughly Muddled McTaggart

Or How to Abuse Gauge Freedom to Generate Metaphysical Monstrosities

Tim Maudlin
With a Response by John Earman

Philosophers' Imprint <www.philosophersimprint.org/002004/> Volume 2, No. 4 August 2002 2002 Tim Maudlin and John Earman

Just when the hubbub over the infamous hole argument seemed to have died down, and you thought it was safe go back into the placid pool of classical GTR (away from the white-water rapids of quantum gravity), John Earman has conjured up yet another monster to trouble poor old Einstein.1 This time, a sprinkling of the magic powder of the constrained Hamiltonian formalism has been employed to resurrect the decomposing flesh of McTaggart, who intones that the GTR (yes, the plain old vanilla GTR) implies that "no genuine physical magnitude countenanced in GTR changes over time" (p. 6), i.e., that if the GTR is complete, we not only live in a block universe, but the block is "frozen", with no real physical quantities changing. I hope to drive a stake through the heart of the undead McTaggart and end this new rampage before it has begun. I choose the image of driving a stake through the heart with some care. There is much in this paper, especially in the latter sections, which will be of considerable use to philosophers of physics. Professor Earman has, as usual, mastered quite a lot of mathematical physics that would be beyond the grasp of most of us, and has done us the great favor of reviewing many technically demanding programs that are of especial interest to philosophers of physics. But the motor of this project is supposed to be a very, very surprising feature of the "deep structure" (p. 6) of GTR: namely that according to the deep structure, nothing physically real changes. This, and only this, is the claim I seek to demolish. Even if I succeed, much of interest may be found in the disjecta membra of Professor Earman's paper.
"Thoroughly Modern McTaggart: Or What McTaggart Would Have Said If He Had Learned the General Theory of Relativity," Philosophers' Imprint Vol. 2, No. 3 (August 2002): http://www.philosophersimprint.org/002003/.
1

Tim Maudlin is Professor of Philosophy at Rutgers University.

Tim Maudlin Let's begin with what we agree on. Earman rejects the original McTaggart's A-series/B-series argument. He agrees that if one means by a "block universe" only a universe which can, in its entirety, be modeled by a single 4dimensional manifold with Lorentz metric, then a block universe can contain real physical change. This is, I think, the common sense view. The sort of models of the GTR we are most familiar with, the solutions of the Einstein Field Equations, represent worlds in which things change: stars collapse, perihelions precess, binary star systems radiate gravitational waves and increase their rate of spin. The representation, as a mathematical object, does not change, but that's just because no mathematical object changes. As Earman approvingly paraphrases Savitt: to have a picture of animation, one doesn't have to provide an animated picture. But more than that is here conceded. Earman characterizes the common sense argument for physical change in the GTR as
based on a naively realistic reading of the surface structure of the theory-tensor, vector, and scalar fields on manifolds. But this naive reading must be radically modified if GTR is to count as a deterministic theory, and the modification undercuts the common sense picture of change by freezing the dynamics. [p. 7]

Thoroughly Muddled McTaggart definition of an observable in GTR, which I dub the "Observables Argument". I will treat these two arguments separately, since they rely on different principles. The Hamiltonian Argument The Hamiltonian Argument derives from the work of Dirac on the "constrained Hamiltonian" formalism for presenting a physical theory. Earman presents this formalism and its interpretation very compactly. I would like to explicate the leading ideas in a somewhat different and, I think, more intuitive way. I will not touch on many of the technical details which, I think, play no essential role in understanding the final result. Let us then approach the Hamiltonian argument in a series of steps. First, we are to cast the GTR in Hamiltonian form. We then notice that, so cast, the dynamics of the theory appears to be indeterministic. Next we consider similar cases in which apparent indeterminism arises because of a "gauge freedom" of the theory. We review a standard method for removing this indeterminism by "quotienting out" the gauge freedom in the phase space of the theory. Applying this standard method to the GTR does indeed restore the determinism of the theorybut at a price. The price is that the dynamics of the theory becomes "pure gauge"; that is, states of the mathematical model which we had originally taken to represent physically different conditions occurring at different times are now deemed equivalent since they are related by a "gauge transformation". We find that what we took to be an "earlier" state of the universe is "gauge equivalent" to what we took to be a "later" state. If gauge equivalent states are taken to be physically equivalent, it follows that there is no physical difference between the "earlier" and the "later" states: there is no real physical change. The key step to this argument lies in the technique for 2

So whatever the new threat to change is supposed to be, it does not appear in the "naive reading". The new problem is not to be founded, for example, on the absence of a preferred foliation of space-time into instantaneous spaces, since no such foliation exists in the "naive reading". Rather, the problem only appears when the GTR is recast in a way somewhat different from its usual presentation. There are two wholly distinct arguments for this new "problem of change" for GTR. One argument involves rewriting the GTR Hamiltonian form, which I will call the "Hamiltonian Argument". The other turns on the proper

Tim Maudlin removing indeterminism by quotienting the phase space. In order to understand the nature and prima facie justification of this apparatus, it is best to begin on familiar territory: gauge freedom in classical electromagnetic theory. If we take the ontology of Maxwellian electrodynamics at face value, we get a picture like this. The complete electromagnetic state of the universe at some moment is specified by the values of the electric and magnetic fields at every point of space at that moment, together with their first derivatives. (This gives the free field. We may want to add the distribution of electric charge and its flux to this state, in which case there will be constraints between the field values and the charge densities, but these complications are not of the moment here.) Maxwell's equations then provide the dynamics of this system: they specify how the electric and magnetic fields evolve through time. Generically, the state of the universe changes deterministically under this dynamics. We can reformulate this theory in Hamiltonian form as follows. We begin with a Hamiltonian function that, roughly, specifies the total energy of the system. We then construct a phase space of the system. Each point in the phase space specifies the complete global distribution of the electric and magnetic fields, as well as their conjugate momenta. The time evolution of the system is specified by Hamilton's equations: the rate of change of any of the canonical variables is given by a partial differential of the Hamiltonian. The net result of all this is that the history of the entire system is now represented by a trajectory through phase space, parameterized by time, which solves Hamilton's equations. And since the development of the electromagnetic field in the original (Maxwell) theory was deterministic, we expect the dynamics here to be deterministic: each point in phase space will belong to a unique trajectory which 3

Thoroughly Muddled McTaggart satisfies the equations of motion. Now, it is well known that if a classical electromagnetic field satisfies Maxwell's equations, then the field can be represented by vector and scalar electromagnetic potentials, A and , such that E = -grad - A/c t and B = curl A. It is equally well known that the relation between the potentials and the fields is many-one: different scalar and vector potentials yield the very same electric and magnetic fields. A mathematical operation changing one pair of potentials into another that yields the same fields is a gauge transformation, and the potentials themselves are said to be gauge equivalent. The justification for this terminology is clear: since the fields are taken to be the fundamental ontology, potentials that are gauge equivalent are taken to represent the very same physical state of affairs. The freedom to choose among gauge-equivalent potentials is not a physical degree of freedom: it rather results from the fact that we have many distinct mathematical objects all of which represent the same physical state. Now suppose we wish to formulate the dynamics of the theory in terms of the potentials rather than the fields. Without any further ado, we should automatically expect that (unless something is done), the dynamics in terms of the potentials ought to be indeterministic. For if the original dynamics implies that a state of the electromagnetic field E0B0 will evolve, after a period of time, into E1-B1, then we should expect the new dynamics only to demand that a pair

Tim Maudlin of potentials which yields (by the equations given above) E0B0 ought to evolve into some pair of potentials which yields E1-B1. But since many different pairs of potentials yield E1B1, we have no reason to expect the dynamics to pick out one of these pairs over any gauge-equivalent pair. That is, the dynamics should now manifest itself as a constraint on the evolution of the potentials, but a constraint that can be met by many different trajectories which originate at the same state. In this case, it is obvious that the apparent indeterminism of the dynamics is merely a consequence of the gauge freedom and does not represent any real physical indeterminism. In practice, the apparent indeterminism is removed by gauge-fixing: specifying an additional condition on the scalar and vector potentials such that exactly one member of each set of gauge-equivalent potentials meets the condition. Once the gauge is fixed, there is a one-one correspondence between states of the fields and states of the potentials, and so one expects the determinism of the field dynamics to imply determinism of the dynamics of the potentials. There are many such possible conditions for the gauge, which go by names like Lorentz gauge and Coulomb gauge. One typically picks a gauge so as to make the particular problem at hand more mathematically tractable. All of this is quite uncontroversial and is presented with little comment in the standard texts.2 Formulating the dyAn example from a text I pulled off my shelf (Classical Electromagnetic Radiation by J. M. Marion [New York: Academic Press, 1965], p. 113): Now, it is the field quantities and not the potentials that possess physical meaningfulness. We therefore say that the field vectors are invariant to gauge transformations; that is they are gauge invariant. Because of the arbitrariness in the choice of gauge, we are free to impose an additional constraint on A. We may state this in other terms: a vector is not completely specified by giving only its curl, but if both the curl and the divergence of a vector are specified, then the vector is
2

Thoroughly Muddled McTaggart namics of electromagnetic theory in terms of potentials rather than fields will yield an indeterministic dynamics, which indeterminism arises solely from the gauge freedom and may be eliminated by fixing gauge. And all of this holds mutatis mutandis if one were to try construct a Hamiltonian formulation of classical electromagnetic theory in terms of the scalar and vector potentials rather then in terms of the electric and magnetic fields. That is: given a Hamiltonian, one would begin by constructing a phase space such that each point represents a complete global specification of the scalar and vector potentials and their conjugate momenta. The development of the state of the universe would be represented by a trajectory through this A- phase space. And what one should expect, before writing down or solving a single equation, is that the resulting dynamics will be indeterministic: many trajectories through a given point should be solutions to Hamilton's equations. A given initial state, specified in terms of the potentials, should be able to evolve into any of a set of gauge-equivalent final states, all of which represent the same disposition of the electric and magnetic fields. And so long as one continues to regard the basic ontology of the theory as the electric and magnetic fields, one will regard this indeterminism as completely unphysical: it arises solely from the freedom to choose different gauges (i.e. to choose among different gauge-equivalent potentials) at any time. What is one to do about this unphysical indeterminism? There are at least three options:

uniquely determined. Clearly, it is to our advantage to make a choice for div A that will provide a simplification for the particular problem under consideration.

Tim Maudlin Option 1: Ignore It The simplest option of all is just to disregard the indeterminism once one has recognized its source. That is, one could frame the dynamics in terms of the potentials and admit any trajectory that satisfies Hamilton's equations as a solution, recognizing the existence of multiple solutions from the same initial state as due to gauge freedom. After all, if one regards the fields as the real ontology, then one knows that the dynamics of that ontology is deterministic, and one sees the gauge freedom that comes along with switching to the potentials. One can use the gauge freedom to make particular problems more tractable, but the prima facie indeterminism can just be ignored. Option 2: Fix Gauge As discussed above, one could cut down the space of potentials by adding an additional gauge condition: one can pick a gauge. The phase space will thereby be reduced andif there is a unique pair of potentials that meets the gauge condition for each state of the electromagnetic fieldone would expect the dynamics on this reduced phase space to again be deterministic. In the case of electromagnetism, these conditions will hold if, for example, the gauge conditions specify the divergence of A and the value of at spatial infinity. One might wonder why one would go to the trouble of inflating the phase space by working in terms of the potentials and then commensurately shrinking the phase space by fixing gauge, but the liberty of choosing the gauge could make the problem at hand more mathematically tractable. Option 3: Quotienting Suppose one begins with an "inflated" phase space, in which multiple points correspond to the same physical state. And suppose one has a clear formulation of the conditions in 5

Thoroughly Muddled McTaggart which a pair of points in the phase space are gauge equivalent. Then one can form equivalence classes of points in the phase space, all of which are gauge equivalent and so represent the same physical state. Call these equivalence classes gauge orbits. Finally, one can construct a new phase space, each point of which corresponds to a gauge orbit in the original space. The new phase space is the quotient of the old one by the equivalence classes. And again, intuitively, one expects the dynamics on the new phase space to be deterministic if the theory one started with was deterministic. Quotienting has certain formal advantages over gauge fixing as a way to recover a deterministic dynamics. As we have seen, when one fixes gauge one requires that each distinct physical state have exactly one representation in the "inflated" variables which meets the gauge condition. For if two gauge-equivalent points in the inflated phase space meet the gauge condition, then the dynamics may not determine which of these points the trajectory of the system will pass through. And if some physical state has no representation that meets the gauge condition, then when one fixes gauge one will lose the power to represent some physical possibilities. In contrast, one is automatically guaranteed that each physical state will correspond to exactly one gauge orbit, since the orbits by definition contain all the points in the phase space that represent the state. So if one has a deterministic theory to begin with, but gets an indeterministic dynamics when casting it into Hamiltonian form, there is good reason to believe that quotienting will restore the determinism. Of course, one should not overlook that fact that quotienting may be a more difficult mathematical matter than, say, fixing gauge. Nor should one overlook the fact that none of these formal tricks are really necessary to maintain one's belief in the fundamental determinism of the theory,

Tim Maudlin once one has seen that the phase space is inflated so that different trajectories can correspond to the same course of physical events. Option 1, ignoring the merely apparent indeterminism which inflation creates, is always available. Notice that our whole discussion up to this point is predicated on the assumption that one has an antecedent understanding of gauge equivalence and gauge freedom. This is true in classical electrodynamics, where one accepts the basic field ontology and regards the scalar and vector potentials as mere mathematical conveniences. If one were to become convinced that the potentials were physically real, so that classically "gauge equivalent" potentials represent distinct physical states, then all bets are off. In that case, one might take the indeterminism in the dynamics of the potentials to reflect real physical indeterminism. Or more likely, one might conclude that the dynamics needs to be supplemented to render it deterministic again: after all, the original dynamics is couched in term of the fields rather than the potentials. So one is confident about how to interpret the "indeterminism" in the dynamics that arises for the potentials in electromagnetic theory (whether that dynamics be in Hamiltonian form or in standard differential equations) because of what one accepts from the outset about the basic ontology of the theory. But once the technique of quotienting becomes familiar, there is a temptation to turn this whole process on its head. That is, suppose one is given a phase space and a Hamiltonian, and suppose that the resulting dynamics is not deterministic. (Typically, this manifests itself in certain freely specifiable functions that appear when one solves the dynamical equations.) And suppose that, for some reason or other, one thinks that the true physical dynamics ought to be deterministic. Then one might well be tempted to render the dynamics deterministic by finding equivalence classes of 6

Thoroughly Muddled McTaggart phase points such that, when one quotients by them, the resulting dynamics is deterministic. That is, one does not begin with a clear notion of gauge equivalence, but rather postulates the "gauge orbits" in such a way as to render the dynamics deterministic. Having so determined the gauge orbits, one then concludes that all the states in an given orbit represent the same physical state: the "apparent" differences among the states arise only from different choices of gauge. This topsy-turvy use of quotienting contains several dangers. One danger arises because, in principle, it is always possible to make it work and so render a theory deterministic. One could, for example, assign the whole phase space to a single gauge orbit, so that the quotiented phase space has but a single point. The resulting dynamics is certainly deterministic, if boring: the universe has only one physical state available to it and so always remains in that state. Any amount of seeming indeterminism in a dynamics can be removed by this expedient, but at a heavy price: one would have to abandon both one's belief that the physical state of the universe changes, and one's belief that it might have been different from what it is. It is hard to imagine what could recommend this course of interpretation. Or take an only slightly less extreme example. Begin with a stochastic dynamics for, say, Democritean atoms: the atoms can, from time to time, swerve. Intuitively, a single initial state can then evolve in many different ways. If we cast this dynamics on a phase space, we would expect the laws to admit of solutions that agree for some time and then diverge. But by clever quotienting, we can remove this indeterminism. Begin with some initial state, then form the class of all the states that this state could evolve into or from which it could have evolved. Repeat the process for all the states in this class, and repeat until no more states are added. What we will end up with is the set of all states that have a

Tim Maudlin certain number of the various types of atoms, irrespective of how those atoms are disposed. Now quotient out with respect to these sets of states. The dynamics will again be deterministic but "frozen": no system ever leaves its "gauge orbit". There will be alternative possible physical states, inhabited by different numbers or different sorts of particle, but no such state will ever change in time. All of this is, of course, both formally correct and completely crazy. If we lived in a such a Democritean world, we would reject the "deterministic" dynamics for the simple reason that we see particles moving around and changing position: the world is not frozen. What possible grounds for believing that the world is deterministic could make rational the wholesale rejection of all sense experience? We must, then, be very judicious in our use of the topsyturvy method. If we are sure that the dynamics of some theory ought to come out deterministic, then we had best keep careful track of our grounds for that belief and of the point where some faux indeterminism has entered our mathematics. For if we blindly demand determinism from quotienting, it can certainly meet our demand, but perhaps in a rather nonsensical way. Let's apply all this to the GTR. We begin with the idea that the GTR is, indeed, deterministic, at least in typical applications. Solutions to the Einstein Field Equations are four-dimensional manifolds. Conditions for determinism are most easily stated for globally hyperbolic space-times. Such space-times admit of Cauchy surfaces, i.e. surfaces that every inextendible timelike curve intersects exactly once. If we specify the intrinsic curvature of such a surface, and the physical state on such a surface, and the way that the surface is embedded in the space-time, then there is typically a unique maximal globally hyperbolic 7

Thoroughly Muddled McTaggart space-time consistent with that data. Now suppose we want to cast the GTR into a Hamiltonian form. (Why would we want to do this? There is some reason to hope that it might help when searching for a quantized version of the theory, but otherwise there is no very compelling reason.) We are immediately faced with a difficulty. Classical mechanics and classical electrodynamics are formulated in space-times with absolute simultaneity, and so there is a clear-cut notion of the instantaneous state of the universe. Points in phase space represent these global instantaneous states, and a trajectory though phase space represents the history of the universe as a succession of such states. Furthermore, the problem of gauge freedom only infects the instantaneous states: the states in a gauge orbit are all representations of the same instantaneous state. Thus a deterministic dynamics over the gauge orbits yields a deterministic succession of instantaneous states over time, which is the history of the universe. Now, the fundamental problem when dealing with the GTR is that the four-dimensional solutions to the field equations do not come equipped with anything like absolute simultaneity, and so there is no unproblematic notion of the instantaneous state of the universe. We may foliate a globally hyperbolic space-time by families of Cauchy surfaces, which can in many ways serve the role of instantaneous states, but such foliations are by no means unique. And in the freedom to foliate lies the key to the Hamiltonian Argument. Consider a solution to the EFE's that contains two clocks (figure 1). This solution can be "split up" into a stack of instantaneous states in various ways. One obvious way is by the foliation depicted in figure 2. We can now depict the solution as a succession of global states, in each of which the clocks indicate the same time.

Tim Maudlin .

Thoroughly Muddled McTaggart But the very same solution can also be foliated as in figure 3:

t'3

t'2

t'1

figure 1
t'0

t4
figure 3

t3

t2

t1

figure 2 8

Now in each "instantaneous" state the left-hand clock is ahead of the right-hand clock. This is just as legitimate a way to carve up the model as figure 2. Once we have a foliation, we can begin to apply the Hamiltonian formalism. The points in phase space will represent instantaneous states, i.e., states of Cauchy surfaces. If we use the foliation of figure 2, then the complete fourdimensional solution will be represented by a trajectory through the phase space, and each point on the trajectory will contain clocks that indicate the same time. If we use the foliation of figure 3, then the very same solution will be represented by a completely different trajectory, such that each

Tim Maudlin point on the trajectory contains clocks that indicate different times. Obviously, these two trajectories in phase space will have no points in common. But now comes the critical observation. We can also foliate the solution as in figure 4, with Cauchy surfaces that agree with the figure 2 foliation early but morph into the figure 3 slices later on:

Thoroughly Muddled McTaggart from the figure 2 slicing at t0 (and, if one likes, at all times prior to t0) but later diverges, wandering over to the region of phase space occupied by the figure 3 slicing. And in general, the complete trajectory through phase space up to some time does not determine the future trajectory for the simple reason that the foliation of the space-time up to that point does not determine the foliation later on. And this, in turn, is a consequence of the fact that we have arbitrarily chosen the foliation from among the infinitude of ways of splitting the solution into Cauchy slices. Different slicings yield different trajectories through phase space, and slicings that agree to a point and then diverge yield trajectories which agree to a point and then diverge, i.e. , yield dynamical indeterminism. So before we have written down a single equation, we can make a prediction: casting the GTR into Hamiltonian form will yield a theory with an indeterministic dynamics. But we also understand the source of the indeterminism: it comes from forcing the GTR into the Procrustean bed of the Hamiltonian formalism. In order to do so, we have to import a foliation into our solutions to the EFE's, a foliation that has no basis in the GTR itself. It is the arbitrary nature of the foliation that makes the resulting trajectory through phase space somewhat arbitrary. But we equally see that this indeterminism is completely phony: it has nothing to do with any real physical indeterminism. Given the initial state on a Cauchy surface like t0, the GTR admits of a unique maximal global solution. Carving up that single solution by different foliations yields different trajectories through phase space, but all of these trajectories, in their entirety, represent the same four-dimensional solution. So it should come as no surprise at all that when we put the GTR into Hamiltonian form we get an apparently indeterministic dynamics. How should we deal with this? 9

t''3

t''2

t''1

t''0
figure 4 Again, this slicing will yield a trajectory through the phase space, and the complete trajectory will correspond to the complete four-dimensional solution. But we can immediately see a problem for determinism: the trajectory one gets from the figure 4 slicing agrees precisely with the trajectory

Tim Maudlin Reviewing the options above, we could, first of all, simply ignore it. The theory in Hamiltonian form represents no more physical indeterminism than there is in the EFE's, namely (given the restriction to maximal globally hyperbolic space-times) none. All of the different trajectories through phase space that are solutions to the dynamical equations represent the very same complete four-dimensional spacetime, and the physical magnitudes in the space-time evolve deterministically. Furthermore, the magnitudes in the space-time evolve deterministically: the universe may expand, perihelions may precess, binary star systems may speed up their rotations, just as we always thought all along. Changing to the Hamiltonian formalism gives us no new insight at all into the basic ontology or dynamics of the theory. If one does not want to just ignore the faux indeterminism of the Hamiltonian form of the theory (if, for example, it makes quantization more difficult), then one could try to eliminate the apparent indeterminism by fixing gauge. In practice, this means formulating some constraint on the way the space-time is foliated, so that diverging foliations that yield diverging trajectories no longer exist. This could be done in two ways. One way would be to find some method for canonically foliating a space-time. This would require discovery of some condition that exactly one foliation of any space-time can fulfill. In certain cases, such a condition might exist (e.g., there might be a unique foliation in which a background radiation field is homogeneous and isotropic on every slice), but clearly no such generic condition exists. Minkowski spacetime, for example, is a vacuum solution to the EFE's, but there can be no condition that picks out a unique foliation of Minkowski space-time on account of its symmetries (e.g., symmetries under Lorentz boosts, rotations, and translations). Any canonical slices would also have to be invariant 10

Thoroughly Muddled McTaggart under those symmetries, but no slices exists that are invariant under them all. So no generic condition can pick out a unique foliation of all of the models of the GTR. A more promising approach is this: find a condition that given a single Cauchy surface as data then induces a unique foliation of the space-time. The idea is pretty straightforward: one would expect, for example, that given the slice t0 in figure 2, the method would generate the complete slicing of figure 2, and given the slice t'0 of figure 3 it would generate the slicing of figure 3, and no initial Cauchy slice would generate the slicing of figure 4. Such a project will be mathematically quite non-trivial, but in certain cases one could imagine how it might go: given one Cauchy surface (call it t0), let the slice ti (i a positive real) be the locus of all points p in the space-time such that the maximal future-directed time-like curve from t0 to p is of proper relativistic length i, and similarly for i negative, using a past-directed curve.3 Or there are other ways one could go about this, using the so-called lapse function and shift vectors. In any case, this sort of "gauge fixing" would help solve the indeterminism problem: since the initial data would be consistent with a unique global foliation that satisfies the constraint, one would not get diverging trajectories through phase space that correspond to
The suggestion here, although it would work in some cases, would not solve the generic problem of fixing a foliation into Cauchy surfaces. It is easy to see that the suggested method would indeed foliate the space, and that no timelike curve would intersect any of the hypersurfaces of the foliation more than once, but one is not guaranteed that each inextendible timelike curve would intersect every hypersurface, so the hypersurfaces need not all be Cauchy surfaces. In addition, the method is impractical as a means to solving the EFEs, since one is using the metrical structure of the space-time to determine the slicing, but one does know the metrical structure until one has solved the EFEs. (Gauge fixing in electrodynamics can make solving the equations easier since the gauge constrainte.g., fixing the divergence of Acan be specified before the solution is known.) So we are here making an abstract point about gauge-fixing, not a practical suggestion.
3

Tim Maudlin the same solution merely because the foliations diverge. Of course, there would still be distinct trajectories through phase space that correspond to the same global solution because they are generated from different initial Cauchy surfaces (like t0 and t'0 above), and they might cause problems for, say, quantization. But that has nothing to do with indeterminism. So both ignoring the apparent indeterminism and fixing gauge appear to be viable solutions to the indeterminism "problem". But if one has fallen in love with the "constrained Hamiltonian formalism" and one has solved other faux indeterminism problems (as in electromagnetism) by quotienting, then one might be tempted to try this route. But there lies disaster. Recall the basic strategy of quotienting. We begin with apparent dynamical indeterminism: the dynamical equations permit different solutions with the same initial data so that, for example, according to one allowable trajectory initial state S0 evolves into state S1, but according to another allowable trajectory S0 evolves instead into S'1 and never enters S1.4 This apparent indeterminism could be removed by declaring that S1 and S'1 are really physically identical states: they are gauge equivalent. One then quotients by the complete gauge equivalence classes (the gauge orbits), reducing the phase space and restoring determinism. But if the source of the apparent indeterminism is the
The problem is most easily stated for a case like electromagnetism done in terms of potentials, since there is a common universal time in all models, so one can say that according to one trajectory S0 evolves into S1 five minutes later, while according to the other trajectory it evolves into S1' five minutes later. The solution to the apparent indeterminism is then to make S1 and S1' gauge equivalent. Since in the GTR there is no such common universal time (introducing a universal time function is equivalent to picking a foliation) things are not so simple, as we will see.
4

Thoroughly Muddled McTaggart freedom to foliate the space-time, this solution will be a complete disaster. As we have seen, if we are free to foliate, then the state on t0 could evolve into t3 (as in figure 2), but it could equally well evolve into t''3 (as in figure 4) depending on how we foliate. So the quotienting solution would have to declare that the state on t3 and the state on t''3 are gauge equivalent, i.e. that they are merely mathematically distinct ways of expressing the very same physical state. But this is crazy: t3 has two clocks that indicate the same time, while t''3 has clocks that indicate different times: these are physically distinct states. Even more damning is this: the left-hand clock on state t3 indicates a different time than the left-hand clock on t''3, so if t3 and t''3 are really the same physical state, then the physical state of a clock does not really change as it comes to indicate different times. By similar argumentation, we would come to the conclusion that, no matter how the clock seems to have its hands oriented, it is always in precisely the same physical state: the "change" is merely apparent, not real. These claims are, or course, rather sillybut they are precisely the claims that, according to Earman, the constrained Hamiltonian formalism reveals about the deep structure of the GTR. It is, in fact, not hard to see that if one is going to restore determinism by quotienting, then the gauge orbits have to contain every state on every Cauchy surface in a solution to the EFE's. Only then is one assured that the states along a trajectory will belong to the same gauge orbit no matter what foliation is used to generate the trajectory. And so what holds for the clock would have to hold for the universe as a whole: its physical state never changes, from a millisecond after the Big Bang to a minute before the Big Crunch. In the technical terminology, the dynamics of the theory is pure gauge, since all of the states along every trajectory have to belong to the same gauge orbit. McTaggartor more prop11

Tim Maudlin erly Parmenidesis vindicated: according to the way of Truth, the universe is ever One and Unchanging, only according to the way of Seeming is there change. Now there is an attenuated sense in which the state on every Cauchy surface in a solution of the EFE's is the same: each such state implicitly represents everything that happens at all times, since each such state is compatible with exactly the same maximal globally hyperbolic solution to the EFE's. But that does not make them all physically equivalent: otherwise we get an immediate argument from determinism to No Real Change. There is real physical change because the physical states on the different Cauchy surfaces are different (i.e., non-isomorphic), even if each surface (together with the EFE's) implies the same global solution. After all, how could it be otherwise? We know that the GTR is a theory which predictsand explainsmany changes: the precession of planetary orbits, the collapse of stars, the rate of expansion of the universe, the red shift of light coming out of a gravitational well. It is, indeed, the observation of precisely such changes that provides our evidence for the theory. Any interpretation which claims that the deep structure of the theory says that there is no change at alland that leaves completely mysterious why there seems to be change and why the merely apparent changes are correctly predicted by the theoryso separates our experience from physical reality as to render meaningless the evidence that constitutes our grounds for believing the theory. So the only real question is not that the constrained Hamiltonian formalism (interpreted as Earman suggests) is yielding nonsense in this case, but why it is yielding nonsense. And the freedom to foliate provides the perfectly comprehensible answer. It is only proper to note that Earman did not construct these arguments for the unreality of change: Dirac devel12

Thoroughly Muddled McTaggart oped the abstract form of the constrained Hamiltonian formalism, and more importantly, Dirac suggested that "the gauge transformations be identified as the transformations generated by the first class constraints, where the intended interpretation is that two points of the phase space which are connected by a gauge transformation are to be regarded as representing the same physical state" (p. 8). It is here that the method is turned topsy-turvy: instead of starting with an understanding of which points in phase space represent the same state, one rather does the dynamics first and then concludes from some formal feature of the dynamics that two points represent the same physical state. And we can see why the topsy-turvy method would work for, say, classical electromagnetic theory: the "gauge transformations" so identified really would be the intuitively correct gauge transformations. But generalizing from this sort of example that we should always identify transformations generated by constraints in the Hamiltonian as gauge transformations between physically equivalent states is a dangerous business, as we have seen. It is also only proper to note that criticisms of this use of Dirac's method are not original: as Earman notes, Karel Kuchar, for example, makes exactly the same point. I only hope to have made clear why this particular result is to be expected if the GTR is put in Hamiltonian form. Earman seems to feel the force of these objections, since he is at pains to argue that the "frozen time" results are not "merely formal tricks or artifacts of the constrained Hamiltonian formalism" (p. 9). To allay these doubts, he proposes to derive "similar, if not identical, results in the spacetime setting rather than the (3+1) Hamiltonian formulations" (p. 9). This would be significant, since my argument so far has been that the problems arise from using the (3+1) formulations: it is precisely a foliation that one needs to turn a four-

Tim Maudlin dimensional relativistic space-time into a (3+1) dimensional object. And indeed, the arguments that Earman rehearses next do not hinge on the same mistakes that sink the Hamiltonian argumentthey hinge rather on a completely different set of mistakes. To these we now turn. The Observables Argument Earman next takes up the approach suggested by Bergmann for guaranteeing determinism in the GTR. Like the quotienting technique, Bergmann's approach is certain to yield the result that the GTR is deterministic, at least with respect to all observables, because it secures this result by definition: in order to be an observable, a quantity must be "unequivocally predictable from initial data" (p. 9). The only question left is what the observables of the theory are. We would expect the observables to include: the position of the perihelion of Mercury after some number of orbits, the amount by which light from the Sun is redshifted when it reaches the Earth, the angle at which light from a distant star will reach the Earth during an eclipse, and so on. And we should expect that these observables will change: the perihelion of Mercury will advance at a predictable rate. So one might expect that we would start with these sorts of observables, the ones that provided evidence for the theory in the first place, the ones that were predicted by the theory, and try to discover some generic characterization of observables that includes these sorts of quantities. But instead we are given an abstract characterization of observables that has as a consequence that none of the things that were actually observed and measured and brought forward as evidence for the theory were observable at all! This isn't just topsy-turvy, it's through-the-looking-glass. As Earman puts it: "What may not be familiar to most readers is that Bergmann's proposal implies that there is no 13

Thoroughly Muddled McTaggart physical change, i.e. no change in his observable quantities, at least not for those quantities that are constructable in the most straightforward way from the materials at hand" (p. 10). One might think that, if true, the moral is to find observableslike the position of the perihelion of Mercury relative to the Sunthat are constructable in a less-thanstraightforward-way rather than concluding that there is no real physical change in the world, but having gone through the looking glass, we are apparently to accept some Alice-inWonderland logic from this point on. To fix ideas, let's start with the sort of prediction that we can make using the GTR. It will also help a bit to reflect on the way that the GTR can be used in conjunction with other principles to make predictions, since our focus from here on will be the sorts of predictions that can be made from the GTR neat. So let's start with a "mixed" prediction and then try to work back to a more "pure" one. The GTR has been used successfully to predict the outcome of the following experiment: take two synchronized atomic clocks, put one on a plane and fly it around the Earth on some specified route, then bring the clocks back together and compare them. When brought back together, the clocks will no longer be synchronized, and the amount of disagreement can be accurately predicted using the GTR. (Nota bene: the GTR can be used to make a deterministic prediction about how the synchronization of the clocks will change from the beginning of the experiment to the end, so a fortiori the GTR can be used to predict that things will change. So the only way to secure McTaggart's conclusion is to argue that the relative synchronization of the clocks is not an observable!) How is this prediction made? One starts with facts about the size and mass of the Earth. These are the sorts of data that can be put on a Cauchy surface. One then solves the EFE's to get a full four-

Tim Maudlin dimensional space-time. All of this is pure GTR. But now things get a bit tricky. Knowing the flight plan of the airplane, one can then pick out a trajectory through the spacetime (by reference to the position of the Earth), representing the trajectory of the flying clock, and one can similarly pick out the trajectory of the stay-at-home clock. One can calculate the proper time along these paths, and assuming that the clocks measure the proper time, one can predict what they will show when brought back together. Now it is evident that this whole procedure is not simply a matter of solving the EFE's. In addition to the solution to the EFE's, we have to specify the trajectories of the clocks and have to deploy a principle about what clocks measure. This second principle could be reduced to a purely relativistic one if we used a light-clock (rather than an atomic clock) which can be shown to measure proper time, but even so the first additional bit of information has to be added from the outside. The only way to avoid this would be to include in the model a complete physical description of the airplane and of all physical objects that influence the flight of the airplane and then solve for its trajectory, but this is clearly a practical impossibility. So we need to be a bit careful. What we want to say is that some quantities, such as the proper time along a time-like trajectory, are on the one hand predictable from the GTR and on the other hand observable (by means of clocks), even though the directly observable instruments (the clocks) are not themselves represented in the models of the GTR we actually use. This shortcoming cannot be overcome until we have the resources to represent the physics of the clocks in the models and would probably be mathematically intractable even then. So when young McTaggart speaks of a "genuine physical magnitude countenanced by the GTR" (p. 6) and when those magnitudes are characterized as "observables", one ought to pause: in many cases, the observable 14

Thoroughly Muddled McTaggart physical magnitudes (like clock readings) which are most directly relevant to laboratory operations are not represented in the purely relativistic mathematics used to make predictions. There are two ways to mitigate this problem. One is simply to declare that relativistic quantities like the proper time along a world line, or the gravitational tidal forces at a point, are observables since we have instruments (clocks or water drops) that allow us to observe them, even if we don't include those instruments in our models. Furthermore, we can just grant that we can know, e.g., how to represent the trajectory of the flying clock, even though we don't solve our equations for it. Note that the relevant trajectory is given relative to the Earth: the plane flies at a certain altitude above the Earth for a certain distance before coming back. The second way to mitigate the problem is to investigate experimental conditions where gravity is the only important factor. If we could send our two clocks free-falling along different paths through a gravitational lens, then (supposing their initial trajectories are part of the Cauchy data) we could solve for their trajectories using the GTR: they will follow the appropriate time-like geodesics. We can solve for the point where they will intersect, and predictfrom the GTR alonehow far out of synchronization they will be when they meet. Or, in a more realistic case, predicting the apparent position of a star during a total eclipse does not demand significant input from outside the GTR proper. So it is a plain fact that the GTR makes deterministic predictions about observable physical magnitudes and about how those magnitudes can change. The only real question before us is how Bergmann's seemingly innocuous definition of an observable as any physical magnitude deterministically predictable from Cauchy data (irrespective of whether any instrument can, in the intuitive sense, observe it) can possibly

Tim Maudlin get us into trouble. Why don't things like the precession of the perihelion of Mercury or the reading of our clocks when they get back together turn out to be observables since they are clearly predictable? The devil, of course, is in the fine printand the first bit of fine print occurs in the artfully wrought statement cited above: Bergmann's criterion implies that there is no real change, "at least not for those quantities that are constructable in the most straightforward way from the materials at hand". Let's take a careful look at what exactly that means. The first candidates for Bergmann observables are
local field quantities which are constructed from the metric and its derivatives up to some finite order and which are evaluated at [a] spacetime point, e.g. the Ricci curvature scalar R. Is the value of this quantity at some point to the future of an initial hypersurface predictable from initial data on the hypersurface, or even from data on the entire past of the hypersurface? [p. 10]

Thoroughly Muddled McTaggart back the result. So surely this ought to count, on Bergmann's criterion, as an observable. But according to the analysis Earman offers, the Ricci curvature at a spacetime point is not observable by Bergmann's criterion, i.e., not predictable from initial data, unless the Ricci curvature is constant everywhere to the future of the initial hypersurface. How can that be? The trick is how to interpret the phrase "at a spacetime point". In my presentation, the relevant spacetime point is identified by a definite description: the point where the two geodesics meet. In Earman's approach, the point is not so identified. In fact, in Earman's account there is no story at all about how the relevant point is identified: it is just somehow given. The point is not given by a definite description such as the one offered above: the point is not identified by its spacetime relation to the initial hypersurface, or by its spacetime relation to any material object (e.g., as a point ten miles above the Earth), or by the object that occupies it. It is rather just (magically) given as a point in the "bare" manifold, i.e. as a point in the spacetime manifold before any metric has been specified for the manifold. Now, there is, in the first place, no coherent account about how such a point could be identified independently of the metric and contents of spacetime. And even if there were such an account, there is no account of how the relevant point could be observationally identified so that the Ricci curvature there could be checked. So for all intents and purposes, we are in the following situation. We are given the data on the initial surface, and then someone informs us that they have a spacetime point somewhere in the future in mind, but they provide us with no further information about which point it is, and then they ask whether we can predict what the Ricci scalar is at that otherwise unidentified point. And of course, we could in such a situation neither predict 15

One would initially think that the answer is clearly "yes". To fix ideas, let's take the case discussed above: the initial data include two objects (they may be clocks, but it does not matter for this example) that are being launched from Earth toward opposite sides of a gravitational lens (such as the Sun). Can we deterministically predict the value of the Ricci scalar at the space-time point in the future where the two objects will meet? Evidently, the answer is "yes". From the initial data and the EFE's, we can construct a complete four-dimensional solution, identify the point where the relevant geodesics (which originate at the Earth) meet, and find the Ricci scalar at that point. And if we wanted to do the experiment, we could construct rockets to be launched from the Earth that would measure the curvature when they meet and transmit

Tim Maudlin the scalar curvature there (unless, of course, the Ricci scalar is the same at every point in the future), nor could we later empirically determine what the relevant scalar curvature is. So if the spacetime point is only given to us as a point on the bare manifold, the Ricci scalar at that point is not predictable and so, by Bergmann's criterion, not observable. (Nota bene: the failure of observability has nothing whatever to do with a lack of instrumentation by which one can empirically determine the Ricci scalar; it has to do with the lack of instrumentation by which one can determine the identity of the relevant point.) Earman makes this argument using the technical machinery of diffeomorphism invariance. Take a solution to the EFE's, and choose a diffeomorphism that is the identity map on the initial data surface but not on the point p in the bare manifold. In particular, suppose that the Ricci scalar to the future of the initial surface is not everywhere constant. Then there is a point q to the future whose Ricci scalar differs from that of p in the solution. Choose a diffeomorphism that is the identity on the initial hypersurface and that maps q to p. Now "drag" the original solution along the diffeomorphism to get a new solution to the EFE's. In this new solution, the Ricci scalar at p will be different from the original: it will now be the value that was formerly at q. But since this is also a solution of the EFE's from the same initial data, that data and the EFE's cannot predict the Ricci scalar at p. Of course, if p happens to be the point where the two geodesics intersect in the first solution, it will not be the point where they intersect in the new solution. The geodesics will be "dragged along" by the diffeomorphism, and they will now intersect at a new point on the "bare" manifold, a point whose Ricci scalar is identical to that at p in the original solution. And so if we identify the relevant point by the definite description and empirically identify the point by 16

Thoroughly Muddled McTaggart where the two rockets collide, we have no problems. We only would have problems if we could per impossible identify points on the bare manifold as such. At this point, the attentive reader will begin to feel queasy. For what we have is just another incarnation of the notorious "hole" argument. That argument, recall, was supposed to establish that the GTR is indeterministic because the EFE's only determine a solution from initial data up to a diffeomorphism. Now, whatever one thinks of that argument (my own views have been expressed ad nauseum elsewhere5), all hands in that debate agreed that if there is any indeterminism, it is an unobservable indeterminism, since it concerns what happens at particular points of the bare manifold, but particular points of the bare manifold are not, per se, observable. Now that argument is being used, in conjunction with Bergmann's criterion, to a perfectly risible conclusion. The proposed logical form of a "local field quantity" is a quantity attached to a point on the bare manifold, and then the indeterminism is used to argue that these quantities are not observables (unless, of course, the quantity is the same everywhere, and so predictable). But even if the values of quantities attached to points of the bare manifold were observable in Bergmann's sense, they would not be observable in the normal sense, since we can't identify the relevant points by observation. What we can identify by observation are the points that satisfy definite descriptions such as "the point where these geodesics which originate here meet", and against these sorts of quantities Earman's diffeomorphism argument has exactly zero force. So if one were to start by reflecting on the logical form of the predictions we actually make using the GTR, or if one
E.g,. in "Substances and Space-Time: What Aristotle Would Have Said to Einstein", Studies in the History and Philosophy of Science Vol. 21, No. 4, pp. 531-561.
5

Tim Maudlin were to start by reflecting on what sorts of things we take to be actually observable in spacetime, then one would not begin with the logical form of a quantity attached to a point of the bare manifold. Earman's strategy instead is to start with quantities attached to bare points because they "are constructable in the most straightforward way from the materials at hand", and then to show, to no great surprise, that these are not Bergmann observables. There is nothing technically wrong with Earman's argument; it just seems like a rather senseless way to proceed once one reflects on the source of the difficulties it encounters. Earman does not claim that the examination of "local field quantities" (or the "quasi-local field quantities" one gets by integrating the local ones over patches of the bare manifold, which obviously inherit the same problems) is a decisive argument that there are no Bergmann observables that are different at different places in spacetime. But he casts a skeptical eye on the utility of any quantities which are not "attached" to points of the bare manifold (like the value of the Ricci scalar where the geodesics meet): "[I]t is worth remarking that it is not obvious how these unattached quantities could underwrite B-series change; for such change requires a subject, and since spacetime points and regions are the only obvious candidates for the subject role in GTR, these peculiar unattached quantities would seem to remove the subject of change from the picture" (pp. 10-11). So let's answer this concern before going forward. At the beginning of our discussion, it was agreed that according to the "naive" reading of the GTR, the GTR allows for B-series change. For the models of the GTR are fourdimensional manifolds with a relativistic metric, and that metric allows one to define certain timelike relations among events in the space-time. Thus, a clock in the GTR can change the time it indicates because (a) it indicates different 17

Thoroughly Muddled McTaggart times at different places in the space-time, and (b) the different places are timelike related to each other, so one can speak of which of a pair of indications is earlier and which later. Notice that the subject of the change is a material objecta clockwhich is represented by a spacetime worm that has different features at different points. Nothing at a spacetime point can change, for the simple reason that a spacetime point has no temporal extension: individual spacetime points are not even candidates for "subjects of change", i.e., for the things that change. Subjects of change must persist through time so they can have different properties at different times. And the things that persist through time are typically material objects like clocks or stars or galaxies. These are represented by spacetime worms that are identified by their material contents: the spacetime worm that is the clock is the collection of spacetime points that are occupied by the material of which the clock is made. And in this sense, the subjects of change are not "attached to" bare spacetime points: under an active diffeomorphism, the subject of change (e.g. the clock) will be "dragged along" to a new part of the bare manifold. So even if, in some sense, spacetime points are the ultimate "subjects of predication" in the GTR, it does not at all follow that they are the subjects of change, i.e., the things that change. This is just a confusion of two rather unrelated uses of the term 'subject'. In succeeding sections of the paper, Earman takes up the project of identifying some other "observables" beside his local and quasi-local field quantities. Unfortunately, he never considers anything as simple as "the value of the Ricci scalar where two given (i.e., given in the initial data) geodesics meet". He has some remarks about so-called "coincidence observables", and this looks promising: we may ask after the value of a quantity where the two geodesics coincide. But

Tim Maudlin even here things strike him as problematic. Adjusting his remarks to fit the case of the intersecting geodesics, one would get this:
Admittedly, however, it remains a bit obscure how the value of this coincidence observable is measured. For ... the measuring procedure cannot work by verifying that the coincidence of values ... does in fact take place by separately measuring the [bare manifold position of a point on one geodesic] and the [bare manifold position of a point on the other geodesic], and then checking for the coincidence. For [the positions on the bare manifold] are gauge dependent quantities, and by the Assumption these quantities are not fixed by measurement ... . [p. 13]

Thoroughly Muddled McTaggart to the Einstein field equations, which implies that the dynamics is implemented not by a mapping from one state to another state in the same solution, but from one solution to another solution". But the whole argument to date has been that complete solutions to the EFE'scomplete relativistic histories of the worldcan represent worlds in which things change. Why a mapping from one complete solution to anotherfrom one complete possible world to anothershould even be called "dynamics", or what it has to do with the physics of the one world we live in, is completely obscure. So in the end, we have three arguments against change in the GTR, two demonstrably inadequate and the third incomprehensible (at least to me). One might wonder why Earman would bother with three arguments if he thought that any one of them sufficed to establish the conclusion. The principle he seems to be following is: where there's smoke, there's fire. But sometimes where there's smoke, there's mirrors. The apparent difficulty for changeand therefore for time itselfwhich Earmans McTaggart discerns in the GTR is only an artifact of a bad choice of formalism or a bad choice for the logical form of an observable, not because of any intrinsic problem in the theory. Has our encounter with McTaggart yielded any positive results? There is, if not a moral, at least an intriguing suggestion of the possibility of a moral to our story. It is now a commonplace that there is a deep problem of time that arises when one tries to quantize the GTR, and that this implies a fundamental incompatibility between the GTR and quantum theory. McTaggarts puzzling claim was that he had found a problem of time in the purely classical theory. We have shown McTaggarts worries to be unfoundedbut are left with the intriguing possibility that the problem of time in quantum gravity is equally chimerical. If one casts the GTR 18

Of course, one would not tell where the geodesics coincide in anything like this way: one would tell by sending a rocket along each path and making the measurement when they collide. Nothing in any of Earman's arguments suggests any difficulty about this procedure. So the Observables Argument gets any traction only by considering candidates for observables (values at points of the bare manifold) which are neither the sorts of things one actually uses the GTR to predict nor the sorts of things one would expectquite apart from diffeomorphism invarianceto be observable. And as soon as one tries the arguments out on something that one would predict or observe, they fail. The Observables Argument runs on completely different principles from the Hamiltonian Argument, but it is equally broken-backed. The critical Section 4 ends with yet a third observation, drawn from an entirely different source: the "alternative approach of Ashtekar and Bombelli" (p 15). I can only admit that I have no first-hand knowledge of the theory and cannot make head nor tail out of Earman's description. He says the theory is defined on the space "of entire histories or solutions

Tim Maudlin in Hamiltonian formas is commonly donein order to quantize it, then the interpretative problems attending that form of the classical theory will likely arise again in the quantum version. And if tying "observables" to particular points in the bare manifold makes trouble in the classical version, then we should anticipate difficulties in defining observables in the quantum version. So McTaggart may have done us a favor after all, by redirecting our attention away from quantization back to the original theory as a source of various technical problems that might arise. But however things work out for quantum gravity, there is, after all, no problem of time or change in the GTR. Let's return McTaggart to his final resting place, and let him molder there in peace.

Thoroughly Muddled McTaggart (1) At the outset I want to emphasize a point generally accepted in the physics community but largely unappreciated in the philosophy of science community: There is a uniform method for getting a fix on gauge that applies to any theory in mathematical physics whose equations of motion/field equations are derivable from an action principle. I emphasize that any particle theories and field theories, Newtonian theories and relativistic theories, etc., all fallwithin the scope of the method.6 The first step in employing the method is to convert from the Lagrangian form of the theory to the Hamiltonian form. If non-trivial gauge freedom is involved in the theory, it reveals itself in the existence of Hamiltonian constraints. One then proceeds to identify the first class constraints and, following Dirac, these constraints are taken to generate the gauge transformation on the Hamiltonian phase space. Finally, the genuine physical magnitudes or "observables" are identified as the gauge invariant quantities. Now apply this method to some theory of physics. Suppose that the result offends your (or Tim's) intuitions. This might indicate that the method, despite its universally acknowledged success in providing a precise and systematic explication of the gauge concept across a vast range of cases, breaks down in the case at hand. But in the absence of any competing method for getting a fix on gaugeand I haven't heard a definite competing proposalyou (or Tim) should seriously consider the possibility that your intuitions have to be retrained. Tim doesn't want a competing method, but he does want to be able to cherry-pick the results of the constraint formalism. The motivation behind the technical apparatus is to
Here is one explicit expression of faith in the generality of the method: "It is well known that all the theories containing gauge transformations are described by constrained systems" (Gomis, Henneaux, and Pons 1990, p. 1089).
6

Response by John Earman


I am grateful to Tim for posing his disagreements with me in a form that more than matches my attempt to state the issues in a provocative way. Tim does a brilliant job of explaining the guts of some difficult technical issues. He takes his explanation to show that the sorts of considerations I adduced in favor of modern McTaggartism lead to a precipice below which lies absurdity. I see no precipice but rather a series of steps that lead to an understanding of the motivation and content of contemporary main-line research in the foundations of classical general relativity theory (GTR) and quantum gravity. This research may lead only to a dead end, but there is no a priori way to know this, much less that the research is based on absurd ideas. In what follows I will confine my comments to four points that lie at the heart of our disagreements. John Earman is University Professor of the History and Philosophy of Science at the University of Pittsburgh 19

John Earman Tim Maudlin detect when an apparent violation of determinism is merely a faux violation. Tim's intuitions tell him that some violations of determinism are tolerable, and in these cases he sees no need to save determinism by appeal to gauge freedom. But all the potential violations of determinism covered by the constraint apparatus are of a piece; namely, arbitrary functions of the independent variables show up in solutions to the equations of motion. Since time is normally the or one of the independent variables, what this means is not just that solutions to the equations of motion can agree on initial data while disagreeing at a later time but also that given any allowed initial value of a dependent variable, there is a solution of the equations of motion which has the prescribed initial value of the dependent variable but which gives to the dependent variable any value you like at any future time you choose. If this isn't a violation of determinism, it is hard to know what one would be. And it is hard to see any principled way to distinguished tolerable vs. intolerable violations of this kind. It is sometimes said that the kind of indeterminism that threatens GTR is uninteresting because it is unobservable. But this response buys into the gauge interpretation of the theory: the observables of the theory, in the guise of gauge independent quantities, do evolve deterministically. Furthermore, this response commits one to an ontology and ideology that is quite different from the extant proposals in the philosophical literature. More on this in (3) and (4) below. (2) In a sense there is an alternative to the Dirac constrained Hamiltonian formalism, but what this alternative gives isn't so much a rival account of gauge as a different nomenclature. The alternative works on the Lagrangian formulation of the theory, and it sees gauge freedom at work when Noether's second theorem applies, that is, when the action is invariant under an infinite dimensional Lie group 20

Response to Maudlin Thoroughly Muddled McTaggart of transformations which depend on arbitrary functions of the independent variables. Noether's second theorem implies that the Euler-Lagrange equations of motion are underdeterminedi.e., there is an apparent breakdown of determinism. The underdetermination is overcome if the elements of the invariance group are seen as gauge transformationsas relating different descriptions of the same physical situation. One now has to face the issue of how these Lagrangian gauge transformationswhich act on the independent and dependent variables of the Lagrangian and which map solutions of the Euler-Lagrange equations onto solutionsare related to the Dirac-Hamiltonian gauge transformations, which are point transformations on the Hamiltonian phase space and which map solutions of the Hamilton-Dirac equations onto solutions. In some cases the relation between the two concepts of gauge is transparent; in other cases, such as Einstein GTR, the relation is opaque and requires special effort to discern.7 I will return to the alternative Lagrangian approach below, but now I want to take up Tim's question of why one would want to put GTR in Hamiltonian form. The reason that physicists use the Dirac formalism to get a fix on gauge is that they always have one eye cocked towards quantization and because the standard route to quantization goes through the Hamiltonian formulation of a theory. Trying to travel this route with respect to GTR in order to produce a quantum theory of gravity is known as the canonical quantization program. To my knowledge, all of the leading research workers in this program (with the one notable exception of Karel Kuchar) accept the consequences of applying the Dirac formalism to GTRin particular, the consequence that the observables of GTR are "constants of the motion", a consequence that Tim labels as absurd and
7

See my (2002) for a discussion of this issue.

JohnMaudlin Tim Earman disastrous. The popular press would have you believe that the only viable approach to quantum gravity is via string theory or M-theory as it is now called. But the loop formulation of quantum gravitywhich falls within the canonical quantization programis currently an active research program. In contrast to M(ystery)-theory, loop quantum gravity is a definite theory rather than a wannabe theory, and it is a theory with notable theoretical success in the form of an explanation of black hole entropy and the prediction of area and volume quantization of space.8 I am not giving an argument from authority, although I do think that philosophers are on dangerous ground when they are dismissive of the prevailing opinions of physicists on matter of interpretation. Rather, my point is that interpretations of scientific theories are subject to empirical tests, albeit of an indirect sort. If the loop formulation of quantum gravity continues to make theoretical progress and, eventually, passes experimental checks, then I would take these successes to be confirmation of the gauge interpretation of GTR dictated by the Dirac constraint formalism. Before leaving the issue of quantization of gauge theories I want to emphasize that it is hard to see how the primacy of the Dirac account of gauge can be abandoned.9 As Tim mentions, one approach to quantizing a gauge theory is the brute-force method: impose a gauge condition to kill off the gauge freedom and then quantize in that gauge. What fixing a gauge means is explained in terms of the geometry of the Dirac constraint surface (the subspace of the Hamiltonian
For a review of the loop formulation of quantum gravity, see Rovelli (1998). This approach to quantum gravity takes advantage of a reformulation of classical GTR in terms of a new set of variables (due independently to Abhay Ashtekar and Amitaba Sen) that makes the Hamiltonian constraints easier to handle. 9 The standard reference on quantization of gauge theories is Henneaux and Teitelboim (1992).
8

Response to Maudlin Thoroughly Muddled McTaggart phase space where all of the Hamiltonian constraints are satisfied); namely, the gauge condition must define a transversal in the constraint surface, i.e., a lower dimensional surface that intercepts each of the gauge orbits (as generated by the first class constraints) exactly once. In some cases, such as Yang-Mills theories, familiar gauge conditions fail to define a transversal and, thus, the brute force attempt at quantization is defective. But what is more important is that when a gauge condition does succeed in fixing a global transversal, what one is getting on the cheap, so to speak, is an isomorphic copy of the reduced phase space which results when the Dirac gauge orbits are quotiented out. What this strongly suggests is that the theoretically desirable technique of quantization of a gauge theory would be to pass to the reduced phase space (where the new phase variables are gauge invariant quantities) and then to perform normal quantization on the resulting unconstrained system. Unfortunately, various technical obstructions can block the passage to the reduced phase space, and even if the passage is not blocked there remains the fact that the constraints may be too difficult for physicists to solve. This is why Dirac invented a short-cut method referred to as constraint quantization, which consists in promoting the first class constraints to operators on a suitable Hilbert space and then identifying the physical sector of this space in terms of the state vectors that are annihilated by the operator constraints. But whether one is performing reduced phase space quantization or Dirac constraint quantization, the philosophy is the same: only Dirac observables (= quantities which are constant along the Dirac gauge orbits or, equivalently, phase functions on the reduced phase space) get associated with quantum observables in the form of self-adjoint operators. (3) Leaving now the issues of quantization and eschewing the (3+1)-dimensional Hamiltonian approach in favor of 21

John Maudlin Tim Earman the 4-dimensional Lagrangian approach, the nomenclature changes but the essential conclusions remain the same; in particular, applying the considerations outlined at the beginning of (2) to GTR leads to the result that the EulerLagrange equations of GTR are underdetermined and, thus, that GTR is apparently an indeterministic theory. Tim is right that this is just Einstein's notorious "hole argument" dressed in a new guise. Tim feels that this argument should have been put to rest long ago. I agree, but for different reasons. The reactions in the philosophical literature to the hole argument are amazing in terms of their ingenuity and the extravagances they employ, and generally they have been skewed because philosophers want to seize the opportunity to ride a favorite hobby horsea favorite account of identity across possible worlds, a favorite account of how language and reference work, a favorite account of essential properties, etc. I ask them to pause for a moment and consider the fact there is an almost universally uniform reaction among practicing general relativists; namely, the lesson of the hole argument is (as the Lagrangian approach to gauge tells us) that the spacetime diffeomorphism group is a gauge group of GTR.10 Some philosophers mouth these words but they generally fail to work out the implications of their words, the most immediate of which is that the gauge invariant quantities of the theory must be diffeomorphic invariants. What are such quantities? Generally philosophers don't have an answer because they haven't bothered to ask the question. General relativists have asked, and the negative part of the answer they find is that the gauge invariants of GTR do not include
Again I would emphasize that the loop formulation of quantum gravity is a self-conscious attempt to accommodate the diffeomorphism invariance of classical GTR as a gauge symmetry. String theorist, in conversation if not in print, say that they hope that M-theory will display this accommodation.
10

Response to Maudlin Thoroughly Muddled McTaggart local field quantities (whether scalar, vector, or tensor). The positive part of their answer is that the gauge invariants include (at least) two different sorts of quantities: first, highly non-local quantities, such as volume integrals of local field quantities over all of spacetime; and, second, what I called coincidence events, which consist of a special kind of coincidence of values of two gauge-dependent quantities that go together to form a gauge independent one. Even apart from the issue of change and McTaggartism, this answer is interesting because the ontological picture that emerges from it lies outside the ambit of the normal discussion in the philosophy of space and time. Of course, the answer may be wrong. But it is remiss of philosophers to refuse to explore its ramifications on the grounds that their philosophy tell them it must be wrong. (4) Coming now to the main issue of modern McTaggartism, I am unrepentant in agreeing with Carlo Rovelli that, at base, what GTR describes is not evolution of the familiar kind, i.e., the change over time of observables in the sense of genuine physical magnitudes. Carlo tries to draw some of the sting of this position by saying that what the theory describes is relative change, the change of "partial observables" with respect to one another. I am not fond of this way of putting the matter since these partial observables are not gauge independent quantities and, thus, are not the kind of things whose values, or change of values, could be experimentally detected. My alternative for drawing the sting of the no-evolution view is threefold. First, I note that the form of McTaggartism that emerges from GTR does not support McTaggart's ultimate conclusion that time is unreal. Second, I point out that the gauge interpretation of GTR is compatible with an attenuated kind of change, for it is compatible with taking the history of the universe to be what I dub a D-series, i.e., a 22

Tim Earman John Maudlin time ordered series of coincidence events with different events occupying different places in the series. This does not go very far towards restoring normal change since the coincidence events do not consist of the occurrence of a change in a genuine physical quantity. Third, I recommend that we don't try to restore normal change and evolution either by changing the theory or by some clever interpretational ploy that rejects or bypasses the gauge interpretation of GTR. Rather, I recommend that if normal change and temporal evolution are wanted, we should seek them not in the intrinsic physics of classical GTR or quantum gravity but in the representations of the gauge invariant content of solutions to Einstein's field equations in terms of the standard textbook models of fields evolving on manifolds. Such a representational stance is nothing new; indeed, it should be familiar from the history of the debates over absolute vs relational accounts of space and time. For example, a good construal of Leibniz's relational account of space is to take him as saying that the Newtonians are welcome to talk about bodies being contained in and moving through space as long as such talk is taken not literally but rather as a way representing the actual and possible relative configurations of bodies. What I am suggesting is that a similar representational account be applied to GTR, and that if it is done in the proper way we can have our cake and eat it too: ordinary talk about change is accommodated in the representations while the gauge interpretation of the theory is respected by recognizing that what these representations are representations of is not of evolution in any ordinary sense. In conclusion, my major disappointment with Tim's response is really a disappointment with my presentation. For what his response reveals is that I failed to convey how the issues surrounding modern McTaggartism are not mere shuttlecocks to be batted back and forth in a philosopher's 23

Response to Maudlin Thoroughly Muddled McTaggart game of badminton. These issues connect directly to decisions that physicists working on the frontiers of research have to make, for example, in searching for a way to marry general relativity and quantum mechanics. If I could make Tim feel the excitement I experience when I see how philosophical concerns about time and change intertwine with contemporary research in physics, he might be willing to join me on the precipicea precipice not of absurdity but of a new understanding of old issues.

References Earman, J. 2002. "Getting a Fix on Gauge: An Ode to the Constrained Hamiltonian Formalism," to appear in K. Brading and E. Castellani (eds.), Symmetries in Physics. Cambridge: Cambridge University Press. Gomis, J., M. Henneaux, and J. M. Pons (1990). "Existence Theorem for Gauge Symmetries in Hamiltonian Constrained Systems', Classical and Quantum Gravity 7, 10891096. Henneaux, M., and C. Teitelboim (1992). Quantization of Gauge Systems (Princeton, NJ: Princeton University Press). Rovelli, C. 1998. " Loop Quantum Gravity," Living Reviews in Relativity. http://www.livingreviews.org

Das könnte Ihnen auch gefallen