A system can have a unique next state without allowing an observer to predict its distant future. Determinism concerns the rule governing a complete state. Predictability also depends on what we know, what we can measure, and what question we are trying to answer.
A rule with no random choice
Take a number x between zero and one, excluding one. Double it and discard the integer part. Repeat. This defines a deterministic update:
For each exact input there is one output. We also display a single bit before each update: zero when x is less than one half, and one when x is at least one half. That display tells us which half of the interval currently contains the state.
In binary notation, the update shifts the digits after the point one place to the left and removes the first digit. The display reads the digit that is about to be removed. The rule is therefore particularly easy to follow without simulation.
Two close starts, different later records
Choose two exact starting values, 22/64 and 23/64. Their binary forms are 0.010110 and 0.010111. The gap between them is just 1/64. Their first five displayed bits are identical, but their sixth bits differ.
| Readout number | Starting at 22/64 | Starting at 23/64 |
|---|---|---|
| 1 | 0 | 0 |
| 2 | 1 | 1 |
| 3 | 0 | 0 |
| 4 | 1 | 1 |
| 5 | 1 | 1 |
| 6 | 0 | 1 |
An observer who knows only the first five binary digits cannot distinguish these starts. That observer can nevertheless predict the first five readouts exactly. Uncertainty becomes relevant at a particular horizon and for a particular output. “Unpredictable” need not mean that nothing useful can be said.
Both chosen fractions have terminating binary expansions, so both eventually reach zero. This finite example illustrates a limit imposed by incomplete initial information; it is not an example of two indefinitely nonperiodic trajectories. Other initial values can have nonterminating binary expansions. The distinction matters when moving from a teaching construction to a claim about chaos.
More precision buys a longer horizon
In this idealized bit-shift example, each additional known binary digit supplies one additional exact future readout. To predict a very long sequence, we need correspondingly detailed initial information. A fixed precision is not enough for arbitrarily many exact predictions.
A computer introduces another issue. Stored numbers have finite precision, and rounding can change a later result. That numerical limitation is different from uncertainty in the physical initial state, although both can affect a computed forecast. A calculation should distinguish the system being modelled, the observations used to initialize it, and the algorithm used to propagate it.
The essay on simulation time makes a related distinction between the model's duration, the integration step and the computer's execution time. A more powerful computer can reduce some computational limits. It cannot reconstruct information that the measurement never supplied.
Why the distinction matters in physics
In his 1963 paper, Edward Lorenz studied a small system of deterministic differential equations arising from a simplified convection model. He exhibited bounded nonperiodic behaviour and emphasized the forecasting implications of sensitivity to starting conditions. The example showed why an exact evolution rule need not provide dependable long-range forecasts from imperfect observations.
Useful statistical prediction may remain possible even when a detailed trajectory cannot be forecast. Conversely, some deterministic systems are easy to predict over the times and observables of interest. Determinism alone establishes neither universal predictability nor universal practical unpredictability.
Quantum probabilities introduce additional interpretive questions. A classical example of sensitivity to initial conditions does not explain quantum measurement or reproduce entanglement correlations. Any proposed deeper account still has to meet the constraints discussed in entanglement across separation.
A complete state for which question?
A present display can omit distinctions that matter later. That is why a model's operational state must be assessed against the predictions it is intended to support. The predictive-memory essay explores this through a finite system in which retaining additional information changes what can be forecast.
For the Single Monad Model and Duality of Time Theory, a claim about a complete generative description should therefore be accompanied by a specified readout and update rule. Which hidden distinctions can safely be ignored? Which ones affect future observable probabilities? How much history must an effective description retain? The published completion-principles paper provides a route into those questions.
Sources and further reading
- Edward N. Lorenz, “Deterministic Nonperiodic Flow,” Journal of the Atmospheric Sciences 20 (1963), 130–141. MIT-hosted full text.
- Mohamed Haj Yousef, “Completion principles, real commutants, and positive-frequency geometry for orthogonal Hilbert representations,” Open Journal of Mathematical Sciences 10 (2026), 996–1012. Published paper.
- Claims and evidence: separate exact calculations, empirical tests and interpretive conclusions.