Key Concepts in Physical Theories: Determinism, Predictability, Stochasticity, Causality, and Reversibility

Start with a state and a law

A physical theory represents a system by a state space $S$: the set of states that count as complete descriptions according to that theory. It then supplies a dynamical rule describing how states evolve.

For a time interval $\Delta t$, write the evolution rule as

A theory is deterministic when its complete state and laws fix one and only one later state. Equivalently, ${\Phi }_{\Delta t}$ is a genuine function: every allowed input state has a unique output.

For continuous dynamics, the same idea appears as an initial-value problem. If the state $s(t_0)$ is specified and the law has a unique solution, then there is exactly one trajectory $s(t)$ compatible with that state and law.

Determinism is a claim about the theory’s ontology and dynamics, not about our practical ability to forecast.

This definition immediately makes two qualifications essential:

1. Completeness matters. If position alone does not fix a particle’s motion, that does not make Newtonian mechanics indeterministic; its state includes both position and momentum. A description that omits momentum is incomplete relative to that theory.

2. System boundaries matter. A subsystem can appear random because omitted environmental degrees of freedom influence it. Whether that randomness is fundamental or merely due to an incomplete system description is a central question in quantum foundations.

The relevant question is therefore never merely “does the observed variable have a predictable future?” It is:

Five notions, five different questions

The distinctions can be stated compactly:

• Determinism concerns the number of physically allowed histories through a complete state.
• Predictability concerns epistemic access: measurement precision, model adequacy, computational tractability, and forecast horizon.
• Stochasticity concerns whether a law itself gives probabilities for alternative transitions.
• Causality concerns explanatory or counterfactual dependence, not merely mathematical entailment.
• Reversibility concerns whether evolution loses information about prior states.

A single theory can possess any combination of these properties. That fact will keep us from drawing premature conclusions about quantum mechanics.

Determinism is not predictability

Laplace’s ideal predictor is rhetorically powerful: if an intelligence knew every position, momentum, and law, it could calculate all future events. But that picture packages together two distinct claims:

1. the laws yield a unique future; and
2. some agent can obtain, represent, and compute from the required state.

Only the first is determinism.

A forecast requires at least:

• a sufficiently accurate dynamical model;
• sufficiently precise information about the present state;
• sufficient computational resources;
• and, for a physical agent embedded in the world, enough time to perform the computation.

Failure of any one condition can defeat prediction without introducing multiple physically possible futures.

Chaos: unique evolution with rapidly vanishing forecasts

Chaos is the cleanest counterexample to the equation

For a chaotic system, two initially nearby trajectories may separate approximately as

where $\lambda >0$ is a Lyapunov exponent. The exact initial state still follows one unique trajectory. But a finite uncertainty in the measured initial state is exponentially amplified.

The image does not depict a ball randomly deciding how to move. Given its exact position and velocity, together with the idealized billiard-table geometry, Newtonian dynamics fixes every collision and every later location. The obstacle’s curved surface amplifies minute differences in incoming direction. An experimentally unavoidable uncertainty in the launch state eventually becomes a macroscopic uncertainty in the ball’s location.

This yields a useful three-level distinction:

1. Microscopic dynamics: deterministic, under the ideal model.
2. Individual long-run trajectory: not practically predictable because of amplified uncertainty.
3. Coarse-grained statistics: often predictable; one may forecast distributions, averages, or invariant measures even when a particular trajectory is inaccessible.

So “looks random” is not yet evidence for fundamental stochasticity. It may reveal coarse-graining, ignorance of initial conditions, environmental coupling, or deterministic chaos.

Stochasticity: probabilities in the law or in the description?

A deterministic law provides one successor state:

A stochastic law instead supplies probabilities:

where several successor states can have nonzero probability. If, for the complete state $s_t$,

for more than one possible $s^{\prime }$, then the model is stochastic at that state-description level.

The phrase at that level matters. Consider a coin toss:

• At a coarse level—“the coin begins somewhere near here, with roughly this angular velocity”—we model heads or tails probabilistically.
• Under an ideal classical microscopic description of the coin, hand, air, and surface, the outcome would be fixed by the exact initial conditions and Newtonian dynamics.

The probability in the first description can represent epistemic uncertainty: ignorance or deliberate coarse-graining. It does not automatically represent objective chance built into nature.

In quantum theory, this distinction becomes sharper. The Born rule gives highly successful probabilities for experimental outcomes. But whether those probabilities are irreducible features of reality, arise from hidden variables, express branching structure, or reflect an agent-centered account of quantum states depends on the interpretation. The operational success of probabilistic predictions alone does not settle that ontological issue.

There is also a useful quantum-engineering version of the point. A quantum channel acts deterministically on a density operator:

For a specified input $\rho$ and channel $\mathcal{E}$, the output density operator is uniquely fixed. Yet an individual measurement performed on that output can have multiple outcomes with Born-rule probabilities. Thus “deterministic state transformation” and “unique observed outcome” are different claims—one we will develop carefully later in this module.

Causality is not determinism

The word causal often appears in “causal determinism,” but it should not be treated as redundant with determinism.

A deterministic law says:

This is a relation of mathematical entailment. Causation, by contrast, normally adds a claim of directed physical dependence: changing or intervening on one factor would alter another, perhaps via an identifiable mechanism.

A causal model may be stochastic. For example, a preparation procedure might causally affect the probability distribution of a detector click without fixing whether a particular run clicks. In interventionist notation, one can express a causal influence through a difference such as

That a cause changes probabilities does not require it to force one outcome.

Conversely, deterministic equations do not by themselves settle the causal interpretation of their variables. A time-symmetric equation may let us infer both future from present and past from present. That mathematical symmetry does not mean the future causes the past. The ordinary causal arrow is tied to further structure: boundary conditions, intervention possibilities, relativistic causal order, and thermodynamic asymmetries.

This distinction will matter later when Bell’s theorem is described as ruling out a certain kind of local causal explanation. Bell violations do not simply say, “the world has no causes,” nor do they directly rule out all deterministic theories.

Reversibility: can the past be recovered?

A deterministic evolution map has a unique output for each input. A reversible evolution has the stronger property that each output also has a unique input. Formally, $\Phi$ must be invertible:

So reversibility requires both:

• forward uniqueness: one future for each present state;
• backward uniqueness: one past for each present state.

A simple discrete model makes the distinction clear:

This law is deterministic: each $x_t$ has one successor. But it is not reversible. If $x_{t+1}=4$, the preceding state could have been $x_t=2$ or $x_t=-2$. Information about the sign has been erased.

By contrast, classical Hamiltonian mechanics, with a suitably complete microstate, is generally reversible. So is closed-system unitary quantum evolution. In both cases, the state at one time determines states in both time directions.

This does not mean real-world reversal is feasible. To reverse a macroscopic process, one would need an astronomically precise control of the full microscopic state. Practical irreversibility can therefore coexist with reversible fundamental equations. Coarse-graining and environmental entanglement often make the inverse map inaccessible for a subsystem even if the global evolution is reversible.

The quantum case: one formalism, several dynamical claims

For a closed nonrelativistic quantum system with a specified Hamiltonian, the Schrödinger equation is

Given $|\psi (t_0)⟩$, this equation has a unique solution:

For time-independent $H$,

Because $U$ is unitary,

so the closed-system state evolution is both deterministic and reversible. This is a precise mathematical result, not an interpretive preference.

The familiar projection postulate introduces a distinct rule. If an observable has projectors $P_i$, then outcome $i$ is assigned probability

and, conditional on that outcome, the state updates to

If this collapse is taken as a fundamental physical process, it is stochastic: the pre-measurement state does not select one unique outcome. It is also non-reversible as a state update, because the components removed by projection cannot generally be reconstructed from the post-measurement state alone.

Thus the usual textbook package contains two importantly different dynamical notions:

At this stage, do not draw the conclusion that quantum theory is simply deterministic or simply indeterministic. The accurate preliminary statement is narrower:

Closed-system unitary evolution is deterministic and reversible at the quantum-state level. Whether a complete quantum theory has one stochastic collapse outcome, many decohered branches, additional deterministic variables, or another ontology is an interpretive and theoretical question.

A diagnostic protocol for later lessons

When encountering any physical theory—including a proposed quantum completion—classify it in this order:

1. Specify the state space. What variables constitute a complete state?
2. State the dynamics. Is the rule a function $\Phi$, a probability kernel $K$, or a combination of rules?
3. Test forward uniqueness. Can one complete present state yield more than one allowed future?
4. Test invertibility. Does one present state correspond to more than one possible past?
5. Separate theory from epistemology. Is unpredictability caused by incomplete information, chaos, computational limits, or genuinely stochastic dynamics?
6. Add causal claims only explicitly. What causal structure, locality condition, or intervention relation is being assumed beyond the bare evolution law?

This protocol prevents several recurring mistakes: treating randomness as proof of indeterminism, treating unpredictability as proof of randomness, treating time-reversibility as backward causation, or treating a deterministic state law as a guarantee of one observed measurement outcome.

The core distinctions are now in place:

• Determinism means unique evolution from a complete state.
• Predictability is an agent-relative ability to forecast, which chaos can defeat without defeating determinism.
• Stochasticity means that the dynamical law assigns probabilities to alternative transitions; such probabilities may be fundamental or may arise from incomplete descriptions.
• Causality adds a directed explanatory or intervention-based structure and can be deterministic or probabilistic.
• Reversibility requires unique evolution in both temporal directions, a stronger condition than determinism.

Next, we will use this vocabulary operationally: given a toy theory’s state space and dynamical law, you will determine whether it is deterministic—and identify exactly which missing state variables or non-unique transitions matter.