Can a simulation be the same reality?
S–I–C–T, category theory and the line between a mathematically equivalent model and a physically identical world.
Structural equivalence is not ontological identity.
Two dynamical systems can be isomorphic in a chosen mathematical category: a structure-preserving map can match their evolution and specified observables. That is a precise mathematical statement. It does not establish that the systems occupy the same physical substrate, possess the same thermodynamic implementation or constitute one and the same reality.
The S–I–C–T research programme distinguishes what is preserved under a model mapping from what the mapping deliberately forgets. In a second-order audit of an earlier category-theoretic paper, the principal category and error-propagation arguments survive under explicit assumptions. Important overstatements about logical erasure, physical heat and continuum state spaces require corrections. A new exact theorem on finite factors, together with a corrected six-bit benchmark, strengthens the mathematical boundary.
Four coordinates, four distinct questions
The S–I–C–T framework separates a system's state-carrier description from its information, invariant constraints and dynamical evolution. In the present application these are proposed measurement lenses—not four new laws of physics.
S — Structure
What is the state space? Record its topology, dimension, representation, precision and the encoder's domain. A finite register and a real-valued continuum need not be the same mathematical object.
I — Information
Which ensemble is being measured? Report Shannon entropy, uncertainty under decoding and conditional information loss. Entropy claims are meaningless without the relevant distribution.
C — Cohesion
Which independently chosen quantities are conserved? Test invariants such as energy or magnetisation, and avoid using the same data to construct and validate a result.
T — Transformation
How do states evolve? Measure local consistency error, error propagation, merging of representable states and validated prediction horizons. Distinguish observed behaviour from proven guarantees.
Three results that survive scrutiny
1. Isomorphism preserves the chosen dynamics
Given a continuous invertible state map h with a continuous inverse, exact conjugacy requires the source transformation F and model transformation G to commute with that map, while specified observables are preserved. When a category omits spatial carriers or probability measures from its morphisms, an isomorphism cannot establish their equality.
2. A finite exact factor of surjective dynamics is a permutation
Theorem. Let F:A→A be onto, let Z be finite, and let h:A→Z be onto. If h ∘ F = G ∘ h, then G:Z→Z must be a permutation. Proof: since both F and h are onto, G is onto; every surjection of a finite set onto itself is bijective.
This rules out a specific possibility: a lossy finite-state update cannot be an exact surjective factor of a reversible target. It does not rule out approximate emulation, finite-horizon agreement or a different infinite-state implementation.
3. The error budget depends on reachable states
Let δ₀ be initial encoding error, ε a one-step consistency bound valid throughout the reachable simulator states and L a Lipschitz constant for the target. Then the decoded model error is bounded by:
Evaluating ε only on the initial energy shell is not sufficient if an integrator subsequently leaves that shell. The distinction is measurable in the oscillator benchmark.
Reproducibility rather than analogy
The supplied Python file was rerun, and a separate deterministic extension adds a better reversible-target test. These numerical findings are implementation-specific, not measurements of the underlying universe.
Experiment A — A reversible shift with a lossy register
The target cyclic shift permutes all 64 six-bit states. The simulator shifts and then clears one bit, losing one bit per step until only one state remains.
Remaining Shannon entropy: 6 bits
Cumulative information lost: 0 bits
Green: reversible target (always 6 bits). Gold: lossy simulator (6 → 0). Logical entropy loss does not measure total device heat.
Experiment B — Explore a proved upper bound
Adjust the parameters of the error-budget theorem to see how precision and target expansion change a guaranteed mathematical bound. The output is not an empirical forecast for a particular physical system.
Computed upper bound:
—
Formula: Lⁿδ₀ + ε·ΣLᵏ. Only valid when the error and Lipschitz assumptions hold on the required states.
Original study: what was actually observed?
| Benchmark | Supported finding | Epistemic status |
|---|---|---|
| Shift vs Rule 90 (6, 7, 8 bits) | Exhaustive shift conjugacy; finite Rule-90 ensemble entropy loss. Rule 90 is an irreversible target and not the appropriate Landauer asymmetry test. | GROUNDED |
| Harmonic oscillator | Explicit Euler exhibits growing energy drift; Verlet stays near its modified invariant. At 200,000 steps, a path-weighted error bound is nearly sharp. | GROUNDED |
| Lorenz–63, RK4 | Small machine-level differences can disappear into identical digital trajectories. The supplied code records first coincidences after 8, 14 and 34 steps for 1-, 2- and 3-ulp displacements. | GROUNDED |
| Corrected six-bit benchmark | A reversible target and a noninjective simulator are compared directly; exactly six bits of logical entropy are lost over six steps, then the rate becomes zero. | GROUNDED |
Important qualification: the Lorenz Jacobian norm of approximately 30.3 is a sample observation, not a certified global Lipschitz constant. First-merger times can vary across machines or numerical libraries.
Logical erasure is not a permanent heating law
For a fixed deterministic update on at most 2b states and an ensemble distribution μn, the per-step logical erasure equals the conditional entropy of the previous state given its image. Summing over a run produces a telescoping identity:
This does not say hardware dissipates no energy after the ensemble settles. Real processors repeatedly compute intermediate operations, create or erase scratch data and consume power. The familiar Landauer energy expression applies only under specified physical assumptions; an upper bound on state-level entropy loss is not an upper bound on actual heat release. Bennett's reversible-computation construction highlights how dissipation may be traded against extra information storage.
The computability obstruction also has a precise scope. A countable collection of finite strings cannot bijectively represent every point in a continuum state space. That is a theorem about a chosen continuum model, not empirical proof that fundamental physical states really form such a continuum.
Download the paper, Python source and data
The following files constitute a complete, rerunnable source package. The original supplied audit and scripts are preserved separately from the revised manuscript. Run the extension locally with Python and NumPy; browser-based interactive sliders are explanatory visualisations and do not execute Python on the site.
Revised research paper
14-page PDF: corrected thermodynamic caveats, finite-factor theorem, revised benchmark, limitations and references.
Python and machine-readable results
Original three benchmark families, new six-bit benchmark, executable assertions and numerical output.
python -m pip install numpy
python extension_benchmarks.py
python benchmarks.pyTo cite this work, use the revision date and manuscript title shown in the PDF. The work is a Lab working paper, not an independently refereed publication.
Frequently asked questions
Does mathematical equivalence prove the universe is simulated?
No. Exact equivalence proves preservation of the explicitly selected mathematical structure. The physical substrate and ontological status are not inferred from that theorem.
Can finite digital computation exactly reproduce a reversible flow?
It may closely approximate one for a finite time or reproduce a finite factor. But a noninjective finite update cannot be an exact, onto factor of a surjective target. Stronger comparisons require definitions of the state mapping and observation metrics.
Is erasure entropy equal to hardware energy consumption?
No. A state-level conditional-entropy calculation does not include all processor operations, memory access, interconnects and physical implementation details. Ideal heat bounds require additional thermodynamic assumptions.
What would falsify a claimed S–I–C–T model equivalence?
Pre-register observables and tolerances, test outside the calibration settings, and measure whether simulations violate invariant drift, error budgets, orbit correspondence or reversible dynamics. Failure refutes that particular equivalence claim, not the existence of every conceivable simulation.
Sources and further reading
- Original supplied working audit, First-Principles Validation and Extension of the S–I–C–T Category-Theoretic Reconstruction (2026), and its Python benchmark script.
- Mac Lane, S. (1998), Categories for the Working Mathematician, Springer.
- Bennett, C. H. (1982), The thermodynamics of computation—A review, International Journal of Theoretical Physics.
- Bennett, C. H. (1989), Time/Space Trade-Offs for Reversible Computation, SIAM Journal on Computing.
- Hairer, Lubich & Wanner (2006), Geometric Numerical Integration, 2nd edition, Springer.
- Read a related S–I–C–T assessment of quantum and classical physics.