Source-linked AI summary
Occam's Quantum Razor: How Quantum Mechanics can reduce the complexity of classical models
Mile Gu, Karoline Wiesner, Elisabeth Rieper, Vlatko Vedral
TL;DR
Provably optimal classical models can still retain more information than is predictive for many stochastic processes. The paper constructs quantum simulators that reproduce the same statistics with lower input entropy when the classical model is non-ideal. The result establishes a quantum route to simpler models, while leaving open whether ideal simulation is always possible.
Problem
For many stochastic processes, provably optimal classical models require more input entropy than the predictive information in the past, leaving room to reduce model complexity.
Source: INTRODUCTION · p. 1, RESULTS · p. 3Method
The paper constructs quantum systems that generate identical statistics with less input entropy than the optimal classical model when its input entropy exceeds the excess entropy.
Source: RESULTS · p. 4Results
For processes with Cµ > E, a quantum model exists with Cµ > Cq ≥ E.
Source: RESULTS · p. 3Takeaways & Limitations
Processes without reversible classical models can be further simplified by quantum processing, and perturbed coins admit a single-qubit quantum epsilon machine.
Source: DISCUSSION · pp. 4–5The construction is not wholly reversible, and it remains open whether ideal simulators are always possible within quantum theory.
Source: DISCUSSION · p. 5
Abstract
from arXiv · showhide
Mathematical models are an essential component of quantitative science. They generate predictions about the future, based on information available in the present. In the spirit of Occam's razor, simpler is better; should two models make identical predictions, the one that requires less input is preferred. Yet, for almost all stochastic processes, even the provably optimal classical models waste information. The amount of input information they demand exceeds the amount of predictive information they output. We systematically construct quantum models that break this classical bound, and show that the system of minimal entropy that simulates such processes must necessarily feature quantum dynamics. This indicates that many observed phenomena could be significantly simpler than classically possible should quantum effects be involved.
INTRODUCTION
Occam’s razor favors models that make the same predictions with less input, and the paper treats minimal input entropy as a measure of model efficiency. It argues that quantum models can reduce the input required by provably simplest classical models, with implications for simulator storage.
INTRODUCTION
- Occam’s razor rejects models that require unnecessary or redundant input when another model makes identical predictions.For an apple in free fall, the trajectory requires only its current location and velocity, not its color or entire past trajectory.Source: INTRODUCTION · p. 1
- A mathematical model is an algorithmic abstraction of a system’s observable output, accurate when its outputs are statistically indistinguishable from the system’s.The paper frames the system and model as black boxes whose outputs an external observer cannot distinguish.Source: INTRODUCTION · p. 1
- The paper interprets an optimal model as one with minimal input entropy; its two-switch example needs one bit rather than two by tracking the previous emission.The example’s outputs alternate between 0 and 1, so the previous emission suffices to model them.Source: INTRODUCTION · p. 1
- Reducing a model’s input entropy can reduce simulator storage requirements because a simulator’s initial entropy matches the model’s input entropy.The figure describes encoding the input, evolving the physical system, and measuring it to retrieve predictions.Source: INTRODUCTION · p. 1, Figure 1 · p. 2
- For any stochastic process, the paper outlines its provably simplest classical model and shows that a lower-entropy quantum model can be constructed unless that improvement violates the second law of thermodynamics.The paper identifies the optimal classical model as the benchmark for this comparison.Source: INTRODUCTION · p. 2
RESULTS
For stochastic processes whose optimal classical model stores more than the excess entropy, the paper constructs a quantum simulator that reproduces the same statistics with lower entropy. The improvement is tied to an irreversibility condition in the classical model.
Results
- Cµ > E exactly when two causal states can emit the same symbol and transition to the same next state with nonzero probability.Such shared futures erase information distinguishing the original states in the classical model.Source: RESULTS · p. 3
- Quantum causal states can be nonorthogonal, allowing the simulator to retain enough distinctions for prediction with less entropy than the classical model.The construction addresses the classical model’s loss of state information by partially discarding distinctions while preserving statistical differences.Source: RESULTS · pp. 3–4
- When Cµ > E, a quantum simulator reproduces the process with E ≤ Cq < Cµ and never uses more than Cq entropy while generating its statistics.This establishes a lower-entropy simulator for every process meeting the condition.Source: RESULTS · pp. 3–4
- For the perturbed coin, Cµ = 1 while E approaches zero as p approaches 0.5; quantum memory improves substantially but remains above E.The figure contrasts the excess entropy, classical complexity, and quantum complexity as the perturbation changes.Source: Figure 2 · p. 4, RESULTS · p. 4
- For K independent perturbed coins, the classical model requires entropy K while the quantum model requires KCq; for p > 0.4, Cq < 0.1.The paper notes that appropriate lossless compression could allow K ≪ N qubits to simulate N coins.Source: RESULTS · p. 4
DISCUSSION
The paper argues that quantum processing can simplify stochastic processes that lack reversible classical models, and illustrates the result with perturbed coins. It also leaves open whether an ideal, fully reversible simulator is possible.
DISCUSSION
- Processes with no reversible classical model can be further simplified by quantum processing.Source: DISCUSSION · p. 4
- A single qubit can simulate perturbed-coin statistics with reduced entropy, offering potential for experimental validation with present-day technology.Source: DISCUSSION · p. 5
- Quantum memory reduction implies that even purely classical processes may be less complex than their classical statistical-complexity bound suggests.Source: DISCUSSION · p. 5
- The constructed quantum simulator is not wholly reversible, leaving open whether more efficient quantum models exist or ideal simulators are forbidden by quantum theory.Source: DISCUSSION · p. 5
METHODS:
The methods characterize when classical causal-state transitions lose information and introduce a refined quantum prediction protocol that maintains memory Cq throughout operation while generating correct statistics.
METHODS:
- A valid emission pair allows a causal state to emit an output and transition onward, while a valid reception pair allows some prior state to reach that state on the output.Source: METHODS · p. 5
- The transition function f maps each valid emission configuration (Sj, r) to its resulting reception configuration (Sk, r).Source: METHODS · p. 5
- The proof links shared-output transitions from distinct causal states to Cµ differing from E, identifying a structural condition associated with classical irreversibility.Source: METHODS · p. 5
- The protocol outputs R1 as the classical emission r and retains R2 × K as the next timestep’s initial condition.Source: METHODS · p. 6
- The protocol does not require knowledge of the measurement result and can execute correctly even when outputs remain unmeasured.Source: METHODS · p. 6
- The refined protocol generates correct prediction statistics while requiring no more than Cq memory throughout operation.Its circuit representation is identified as Fig. 3.Source: METHODS · p. 6, Figure 3 · p. 6