0

Follow-up on this question about causality for Newton's second law.

In $F=ma$, the $=$ sign signifies proportionality, not causality. It makes sense, as the equal sign is invertible, whereas causality is not.

Does this mean that any formula with an equal sign is just a way to connect the r.h.s. with the l.h.s., and not to "explain an effect via a cause"? This seems a bit harsh to me, so I tried analyzing some cases. I've read the Wikipedia article about causality, but it doesn't really help with what I'm looking for, i.e. an explanation for which formulas (and why) show a causality relation, and which ones are simply rewritings of quantities.

  • The only safe ground I have for causality is relativity. For relativity, it's about how an event outside your past light cone can't be "causing" you. This is clear, but it is a reasoning of events: are there formulas that show causality in this sense?
  • There must be a way to include entropy, the clearest measure of time-passing processes, into this, but I can't see how.
  • In formulas like the one for the electric field $E=\frac{kqQ}{r^2}$, it is tempting to say that charges cause the electric field because there can't be a charge without an electric field, but there can well be an electric field without charge. Is this a good example? Why, how is it inherently different from Newton's law?
  • I thought of maybe non-invertible operation such as the gradient to explain how the gradient of a quantity comes necessarily "after" the quantity itself, meaning that for the gradient to be measured the quantity itself must be there first, but it seems off.

I'm very sorry if I'm making a mountain out of a molehill and may be missing some obvious points, but my head hurts and I wouldn't know where to look for an argument about causality in such simple terms.

Qmechanic
  • 201,751
  • 1
    No, causality is a tricky issue and not as settled as we would like. In mathematical physics it is discussed in terms of whether later states are uniquely determined by earlier states on some spacelike boundary; it gets very intricate in general relativity, while it is more straightforward (but by no means trivial) in electromagnetism. – Anders Sandberg Nov 14 '23 at 14:32
  • "the = sign signifies proportionality, not causality." Um, no. The = symbol indicates equality not proportionality and also not causality. For proportionality we usually use the symbol $\propto$. – hft Nov 14 '23 at 23:11
  • @hft you are correct and I was sloppy. I meant, following the reasoning of the linked post, that in "F=ma" the vector quantities "F" and "a" are proportional, and their proportionality constant is "m". – Mauro Giliberti Nov 15 '23 at 09:16

3 Answers3

3

An equation does not have a direction. It simply describes a relationship between different quantities. In $F=ma$, there is no directionality that says "force happens first and then causes acceleration" or vice versa. The statement is "simultaneous" -- if an object is accelerating, then it is experiencing a net force with magnitude $ma$, and vice versa.

However, the concept of causality can be built into equations as a feature. An example is the Green's function for the wave equation. If you have a field $\phi$ governed by the wave equation with a source $J$ $$ -\frac{1}{c^2}\frac{\partial^2 \phi}{\partial t^2} + \nabla^2 \phi = J $$ then a solution to this equation is given by integrating the source against the Green's function (technically the retarded Green's function) $$ \phi(\vec{x}, t) = \int dt' \int d^3 x' G(\vec{x},t; \vec{x}',t') J(\vec{x}',t') $$ If you have not seen this before, do not worry. The main idea is that the field $\phi$ "here and now" at the event $(t, \vec{x})$ is related to a superposition of the source $J$ at various events "there and then" $(t', \vec{x}')$, weighted by some factor $G(\vec{x},t;\vec{x}'t')$ that describes how influences propagate from the past to the present.

The key property that we need (you can prove this from the wave equation directly with a sufficient level of math) is that $$ G(\vec{x},t; \vec{x}',t') = 0 \ \ {\rm if}\ c^2(t-t')^2 < (\vec{x}-\vec{x}')^2 $$ In other words, the source $J$ at $\vec{x}', t'$ cannot affect the field at $\vec{x}, t$ if the events $(t', \vec{x}')$ and $(t, \vec{x})$ are spacelike separated.

The equals sign itself does not imply any "causal relationship". However, the form of the equation above does tell us about causality, by saying what events in the past can be related to the solution in the present. Most equations do not say anything in particular about causality, it is only encoded in special equations that describe time evolution.

Andrew
  • 48,573
  • I have indeed seen the Green's function, and your point is clear. Can we then say that any equation without any functions of some spacetime point (t,x) does not have any "built-in" causality? – Mauro Giliberti Nov 15 '23 at 09:26
  • 1
    @MauroGiliberti I hesitate to say any equation since maybe one can think of counterexamples :D But I would say causality is a physics concept that is present in some but not all equations -- generally it is relevant in some way in equations that describe a system evolving in time. It could appear in subtle ways like in the choice of boundary conditions (in my answer, the choice of a retarded Green's function instead of advanced). But in, say, algebraic geometry, time doesn't appear so there's no notion of causality. – Andrew Nov 15 '23 at 23:06
1

Great question! To start off, you’re exactly right that an “equals sign” in a physics equation has nothing to do with causation. The equations just tell us about correlations in both space and time. Correlation is not causation.

The realization that you can’t find causation in the ``bare correlations’’ has been a key piece of the huge increase in our understanding of causation over the past 30 years. So you can’t find causation in physics equations. Instead, along with the correlations, you need something called a causal model. Everyone uses causal models in physics, but they’re rarely explicitly represented. Physicists tend to just write down the equations and claim that this is all there is, but of course that’s not true: we use the equations in different ways and different contexts. Using the equations in some particular way is what we mean by a “model”, and physics models are almost always causal models. There absolutely is causation in physics models, and it deserves to be more widely understood. It’s not just random whims, and it’s not time-ordering (we have another set of words for describing time-order). Instead, it has to do with real or modeled interventions on a system.

When physicists have a working model of a finite system in spacetime, we intuitively make it a causal model. Specifically, we take some particular physical situation and then imagine how it might be made different. And we imagine the difference-making as originating from outside the system – some parameter on which it seems natural to “reach in” and “intervene” upon, changing it to some new value. This altered parameter is naturally thought of as a “cause”. All the other parts of the system that are correlated with this intervention is the “effect”. This is the interventionist account of causation in a nutshell: in the lab, the causes include our experimental interventions. More generally, causes are the variables one imagines directly intervening on, from outside the system. (Technical caveat referenced below: when we imagine intervening on a variable we also perform “arrow breaking”, preventing other causes of those same variables; we can’t always do this in the lab.)

Consider your example about how the electric field $E$ is correlated with $Q$ and $r$ and $k$. It’s probably intuitive that the position of the charge causes the E-field, and that this is true even if there are other E-fields around. That’s because it’s easy to imagine reaching in and intervening on the precise position of the charge, and this difference would be correlated with the corresponding E-field everywhere. So the charge position causes the E-field. Conversely, imagine reaching in and increasing the E-field in some region of space. The model you’re describing wouldn’t see an (instantaneous) change in the position of that charge, so one would not say that the E-field causes the charge position. (You could add a few more equations where an external E-field would cause a force on the particle, but then you’d have to distinguish between the various E’s in your various equations; each of them could have a different causal assessment.)

Interestingly, some people could look at your equation and will not see $Q$ as causing $E$. Other people might. That’s because people will disagree on whether it’s reasonable to imagine reaching into the model and “intervening” on the value of $Q$ directly. Some simulations let you do this, at which point it seems clear that $Q$ causes $E$. But if the particle is an electron, you might not see $Q$ as being “changeable”, even in principle, so $Q$ starts to be treated more like “k”. It's also barely possible to imagine “$k$” causing $E$, but such a model would be very unusual; most people won’t think of $k$ as a cause of $E$, since it’s hard to imagine some external way to change it.

On your point about relativity: don’t confuse the map for the territory. Relativity tells us where we might expect to see causal relations, and where we definitely do not expect to see causal relations. But it doesn’t say what causation is in the first place. Knowing where we might see causal relations is a far cry from a definition of causality. In fact, if you treat this as a definition of causality (as far too many people tend to do), it becomes circular; the restriction on direct causation outside the lightcone becomes meaningless, if you take that restriction to be the very definition of causation. Only by supplementing relativity with the above definition of interventionist causation can you actually draw some real conclusions.

Entropy considerations come into a causal model in two different ways. First, it comes in via the external agents in a causal scenario, real or imagined. Real external agents can intervene on a system, “causing” something in that system, and entropy considerations put constraints on what those agents can actually do. They can directly set the input of a device, but they can’t set the corresponding output. For imagined agents, as in most physics models, it gets a little trickier, but our understanding of entropy usually constrains our imagined interventions on a system. We can easily imagine changing the initial state of a system from outside the system, but it’s much less common to imagine imposing an external future boundary condition. If you did, you would find that the causal effect was opposite the order we usually expect (although note that the direct causes, even then, would still lie inside lightcones).

Second, entropy comes in when we “break arrows”, in the causal modelling lingo. When we imagine intervening on a parameter, it’s crucial that nothing else is also thought to cause that parameter, or it messes up our causal assessments. When we set an initial condition on some variable in a system, we don’t allow other things in the past to thwart our choice for that variable. So in our models we "break the arrows" linking the variable to events in the past, but not to events in the future.

Take the charge example. At $t=0$, I want to relocate it from $x=0$ to $x=1$ to see what will happen in the model. But I have equations telling me that its past position at $t=-1$ and the forces acting on the particle between $t=-1$ and $t=0$ actually determine its true location at $t=0$. But for causal purposes, I want to “break” these equations; ignore them, set them aside, and instead imagine I have complete freedom to choose $x=1$ at $t=0$. But we don’t generally do the same for the future; we don’t break the link between forces and motion after $t=0$, allowing the model to follow those equations. This seems perfectly reasonable to us entropy-increasing agents, but of course it breaks an arrow of time. If I instead broke the future equations, and left the past equations in place, I’d find that my intervention at $t=0$ was actually a final boundary condition on the $t<0$ side of the model, and I’d see entropy-decreasing behavior. Such a model doesn’t match reality, so it’s an empirically bad model, with an objectively poor choice of imagined interventions.

In short, entropy tells us which causal interventions to consider, and which causal arrows to break. If we pick the wrong interventions and the wrong arrow-breaking, we’ll get results that contradict our entropy-increasing observations, telling us that we probably shouldn’t consider those interventions in the first place. But the same is true for physics equations; if we pick the wrong equations, nature tells us that we’ve made a mistake. Using the scientific method to compare causal physics models with reality therefore guides us towards better equations, as well as guiding us towards the actual causes of a given system. We do this automatically, so that we hardly think about it; we just “know” that we should use initial conditions in our models, a lesson that we’ve already long since internalized. But just because this choice of initial-boundaries is now coming from us when we use our physics models doesn’t mean that it’s purely subjective. It’s the objective considerations described above that tell us which interventions are reasonable to consider in the first place. So in this sense, the direction of causation is just as objective as our most successful equations; they can both be determined empirically.

Ken Wharton
  • 1,515
0

The two axioms listed in the Wikipedia article, you cited, are:

  1. a cause may not lie in the future of an effect,
  2. a cause must lie in the past of an effect.

In Relativity,

  • Future Of = On Or In The Future Light Cone Of, But Not Coincident With,
  • Past Of = On Or In The Past Light Cone Of, But Not Coincident With,
That is: they lie on or in their respective light cones, except at the vertex.

In the world of non-relativistic physics:

  • Future Of = After The Hyper-Plane Of Simultaneity,
  • Past Of = Before The Hyper-Plane Of Simultaneity.

The $F = ma$ law, in the non-relativistic world, refers either to action-at-distance, which is simultaneous, or with direct action by contact, which is both coincident and simultaneous. Both of these relations lie outside of what are covered by the two axioms. For this reason, $F = ma$ can never, in itself, convey a cause-and-effect relation. There just isn't enough time for cause and effect to happen, when things are simultaneous.

This past-of and future-of relations were actually used as a basic element for an axiomatic treatment of the combined space and time geometry by A. A. Robb in 1914: A Theory Of Time And Space. He wrote out the contrastive properties of the Relativistic and non-relativistic worlds. In the case of Relativity, his α and β sets are the "after" and "before" sets - that which we call the future and past light cones today.