Constructor Theory

The Program in Brief

Constructor theory begins with a complaint about the shape of physical explanation. The prevailing form — specify initial conditions, specify laws of motion, integrate — privileges trajectories, and quietly makes everything that did not happen second-class. Deutsch’s proposal is to invert the hierarchy: the fundamental statements of physics should be about which transformations are possible and which are impossible, and why. Laws of motion then become derived, subsidiary facts about a world whose primary description is counterfactual.

The vocabulary is spare and well-chosen. A substrate is a physical system with attributes. A task is a specification of ordered pairs of input and output attributes. A constructor is an entity that can cause a task to be performed on a substrate and retains the ability to cause it again. A task is possible if a constructor for it can be approximated arbitrarily well; impossible if some law forbids it. Everything else — information, work, heat, measurement, life — is built from these.

The payoff is that whole regions of physics can be stated exactly rather than statistically. The constructor theory of information (Deutsch and Marletto) defines information without reference to probability or ensembles, via an interoperability principle, and characterizes quantum theory as the theory of “superinformation” — a medium carrying two information-like attributes that cannot be jointly measured. The constructor theory of thermodynamics (Marletto) recovers a second law that is exact and scale-independent. The constructor theory of life shows that high-fidelity self-reproduction is compatible with a no-design physics. And constructor-theoretic reasoning about information has produced at least one genuinely experimental proposal — the gravitationally induced entanglement test of Bose, Marletto and Vedral, now a live programme in tabletop quantum gravity.

It is a young programme, explicitly foundational, and unusually honest about what it is and is not trying to do. It is also, of all the frameworks compared on this site, the one whose central object comes closest to the framework’s own.

The Constructor and the Observer

Set the two definitions side by side.

A constructor is an entity that can cause a transformation and remains capable of causing it again. An observer in this framework is a triple of a state space, a conserved invariant, and a self/non-self boundary, whose internal dynamics closes into a loop: it interacts, and its state returns to a configuration from which it can interact again. Both definitions are functional — they specify what the thing does, not what it is made of, and both apply indifferently to an electron, a catalyst, an enzyme, or a galaxy. Both make repeatability, rather than composition or scale, the criterion for being a fundamental kind of thing.

This is not a loose analogy. The framework’s operational grounding derives loop closure from exactly the requirement constructor theory writes into its definition: an entity that acts once and is consumed cannot be the subject of a lawlike statement about what is possible, because the possibility must be re-instantiable. Where constructor theory takes “retains the ability to do it again” as a definitional stipulation, the framework derives it — an observer with finite resources whose state drifts by a small amount per cycle has a lifetime proportional to the reciprocal of that drift, and only exact return gives indefinite persistence. Approximate constructors are mortal constructors, and the framework computes how mortal.

The framework then goes somewhere constructor theory has not. If an observer absorbs any relational structure from its interactions — if it records anything — the recording permanently enlarges its state space, and the loop that closed exactly on the smaller space no longer closes on the larger one. The memory-persistence tradeoff makes this a theorem: exact persistence and accumulated memory are structurally incompatible, and the only observer that escapes is the minimal one, whose state space is too small to hold an extra degree of freedom. Read against constructor theory, this says that the constructor which retains the ability to do it again is a genuine limiting case — it must either be informationally trivial, or its repeatability must degrade. That is a substantive refinement of constructor theory’s central idealization rather than a disagreement with it, and it is the sort of thing constructor theory’s own machinery might be able to express.

Possible, Impossible, and Conserved

The two frameworks share a taste for exhaustive classification of the possible, and differ in what does the excluding.

The framework’s Three Interaction Types is a constructor-theoretic result in everything but vocabulary. It asks what can happen when two observers interact, builds a decision tree over whether each invariant survives, and proves the tree exhaustive: Passage, Fusion, Resonance, and nothing else. It is explicitly a statement about what is possible, and it is explicitly silent about rates and probabilities — those are deferred to the Born rule and the quantum formalism downstream. Deutsch’s separation of possibility-statements from dynamical-statements is the same separation, made in the same place.

What differs is the exclusion principle. In constructor theory, impossibility is primitive: a task is impossible because some law says so, and the laws are the statements of impossibility. In the framework, impossibility is always downstream of a conserved quantity. Case D of the interaction table — both invariants destroyed — is eliminated because the coherence they carried has nowhere to go. The conservation of distinguishability derives no-cloning, no-deleting, unitarity, and the monotonicity of coarse-graining from a single fact: an admissible transformation cannot change any coherence value, hence cannot change any quantity built from coherence values. Constructor theory reaches strikingly similar prohibitions from its interoperability principle without ever invoking a conserved measure. The overlap in conclusions across such different primitives is the interesting datum.

The framework’s viability triplet is the most constructor-theoretic object in the whole derivation chain. An observer exists only if it can form, be preserved, and be detected; the Standard Model spectrum is read as the simultaneous-satisfaction set of three independent possibility conditions. Stated in Deutsch’s language: the particle content of the world is the set of substrates for which all three tasks are possible. Free quarks fail exactly one condition — detection — and confinement becomes an impossibility-statement about a specific task rather than a dynamical accident.

Information, Thermodynamics, and Error Correction

Both programmes want a second law that is exact rather than statistical, and both get one — by opposite routes.

Constructor theory’s second law is scale-independent and observer-free: a statement about which work-and-heat tasks are possible, holding as sharply for one atom as for a mole. The framework’s entropy as inaccessible coherence is equally non-statistical — no ensembles, no ergodic hypothesis, no past hypothesis — but it is irreducibly indexed to an observer. Entropy is the coherence a bounded observer cannot reach, the second law follows from boundedness plus conservation, and the entropy of the universe relative to itself is exactly zero. Constructor theory removes the observer to get exactness; the framework gets exactness by taking the observer seriously enough to make the accounting sharp. The two therefore make different statements about who the law is a law for.

The convergence resumes at error correction. Marletto’s constructor theory of life turns on resilience — the ability of a self-reproducer to keep its ability intact under noise — and the framework’s observer as an error-correcting code gives that idea a concrete mechanism: the observer-projection is a channel from a noisy substrate onto an integer-invariant profile, and the observer is the code space that profile cuts out. “Retains the ability to do it again” becomes, in the framework’s reading, a statement that the code distance exceeds the distance the substrate noise demands. Repeatability is not free; it is purchased with error correction, and the framework prices it.

The Pattern: One Primitive, Two Ambitions

Across every point of contact the shape is the same. Both frameworks put a persistent, repeatably-acting entity at the foundation, and both get further than one would expect from so austere a starting point — exact information theorems, exact thermodynamics, classification of the possible. But constructor theory is designed as a mode of explanation: a language in which existing and future theories are to be expressed, deliberately neutral about which dynamical theory turns out to be true. The framework is designed as a first-order theory: the observer is not a way of talking about physics, it is what physics is made of, and the chain is meant to terminate in three spatial dimensions, a gauge group, three generations, and a strain spectrum a real interferometer can measure. Same primitive, opposite ambitions — one wants to be the grammar, the other the sentence.

Where They Genuinely Part Ways

Three divergences are real rather than terminological.

First, the primitives. The framework cannot state an impossibility without exhibiting the conserved quantity that forbids it — a real constraint on what it can say, and a real source of content, since the same conservation law that forbids cloning also fixes the geometry of state space. Constructor theory’s impossibilities need no such warrant, which makes it more general and correspondingly less predictive.

Second, the level of the theory. Constructor theory does not, and does not attempt to, derive a particle spectrum, a spacetime, or a coupling constant; asking it for one is a category error. The framework does attempt exactly that, and therefore carries falsification risk constructor theory has deliberately declined — including a quantitative interferometer signature and a commitment to exact unitarity with no room to retreat. Neither posture is wrong; they are different bets about where foundational progress comes from.

Third, the status of the observer. Constructor theory’s constructor is defined by its effect on a substrate and carries no boundary and no self/non-self partition. The framework’s observer carries both, and its non-triviality conditions — a non-trivial symmetry group, the existence of genuine threats, an informative invariant — do real work downstream. A constructor need not be threatened by anything. An observer must be, or it is not an observer.

What the Framework Takes From It

Constructor theory is the strongest available argument that possibility-statements are a legitimate and fruitful primary layer of physics, and the framework should be read as agreeing. The interaction-type classification, the viability triplet, and the no-go results derived from distinguishability conservation are all possibility-statements in Deutsch’s sense, and they are among the framework’s cleanest results precisely because they do not depend on dynamics.

There is also a discipline worth borrowing. Constructor theory is scrupulous about distinguishing what its formalism can express from what it can predict. The framework, which claims more, is under a correspondingly heavier obligation to be explicit about which of its results are theorems, which are structural arguments, and which are conjectures.

Finally, the near-identity of constructor and observer should be taken as weak evidence for both. Two programmes with different motivations, different formal apparatus, and no shared lineage independently concluded that the right fundamental object is a thing that acts and remains able to act. That convergence is not proof of anything. But when two people surveying from different hills draw the same ridgeline, it is reasonable to suspect there is a ridge.