Everettian Quantum Mechanics

The Program in Brief

Everett’s move was one of subtraction. Quantum mechanics as taught has two dynamical laws — smooth unitary evolution, and a discontinuous collapse that fires when someone measures — and no principled account of which applies when. Everett deleted the second law and asked what the first one, applied without exception, actually says. The answer: the measuring apparatus, the laboratory, and the physicist all become entangled with the system, and the universal state develops into a superposition of decohered components, each containing a version of the observer who saw one definite result.

The program’s modern form is considerably more sophisticated than the popular slogan. Decoherence supplies the branching structure: interaction with an environment rapidly diagonalizes the density matrix in an einselected pointer basis, and the branches are dynamically autonomous thereafter. Wallace’s The Emergent Multiverse argues that branches are emergent structures in the universal state — real patterns, not extra postulated entities — so the multiverse is a consequence of unitary quantum mechanics rather than an addition to it. And the Deutsch–Wallace decision-theoretic program aims to show that a rational agent facing branching, constrained by symmetry and consistency axioms, must set credences equal to squared amplitudes.

Whatever one concludes, the temperament deserves credit. Everettianism takes the formalism at face value and refuses to bolt on machinery to make the answer come out comfortable. The framework shares that instinct completely — and then makes a different set of trades.

Unitarity Without Collapse

The deepest agreement is also the most easily stated: neither program has a collapse postulate, and both regard exceptions to unitarity as a symptom of an incomplete description rather than a feature of nature.

The framework does not adopt this as a stance; it inherits it. Coherence is conserved at every node of the dependency graph (Coherence Conservation), admissible transformations preserve every coherence value exactly, and conservation of distinguishability turns that into isometry of the state-space geometry — which is unitarity, with no-cloning and no-deleting falling out of the same argument. The resulting commitment is stated without hedging as a prediction: exact unitarity at all scales, no objective collapse, no nonlinear correction, no mass threshold where the Schrödinger equation quietly stops applying.

So the two programs sit on the same side of the sharpest live experimental question in quantum foundations. If matter-wave interferometry ever finds a genuine spontaneous-collapse signature — GRW, CSL, gravitationally induced reduction — both are falsified together. That is real shared ground, and it means their disagreement must be settled somewhere other than at the interferometer.

The frameworks also agree on the structure of the measurement itself. The von Neumann coupling $\sum_k \psi_k |d_k\rangle_{\mathcal{O}} \otimes |k\rangle_S$ that Everettians read as the branching event is the same state the framework writes down in Measurement — where it is derived as the unique unitary consistent with a Type III interaction rather than assumed. The disagreement is not about the state. It is about what the sum over $k$ means.

The Preferred Basis

Why is the world carved into position-definite pointer states rather than superpositions of them? Everettianism’s answer is decoherence: environmental monitoring einselects a pointer basis, and the branch structure is whatever the decoherent-histories machinery delivers. Wallace’s defence of this is careful and, on its own terms, effective — the branching structure is emergent and approximate, and asking for a sharp branch count is like asking for the exact number of tigers in a jungle.

The framework locates the basis one level earlier. Every measurement is a Type III interaction that generates a relational invariant between observer and system; that invariant is a conserved, real-valued quantity, hence a self-adjoint operator, hence — by the spectral theorem — carries a unique eigenbasis. The preferred basis is that eigenbasis. Basis selection is exact rather than asymptotic, complementarity follows from non-commuting invariants generated by different interactions, and Zurek’s einselection is recovered as the macroscopic limit where many environmental invariants share a common eigenbasis.

This is a real structural difference, and it should be stated with its price attached. The framework’s derivation is provisional: it rests on a structural postulate identifying which interaction Hamiltonian governs a given configuration. The correspondence between relational invariants and Noether charges is forced; the explicit map from a laboratory setup to its Hamiltonian is not yet derived. Everettianism’s reliance on decoherence is a different kind of debt — approximateness rather than incompleteness — and it is not obvious that one debt is strictly smaller. What is fair to say is that the two answers differ in kind: one says the basis is what survives environmental monitoring, the other says the basis is what the interaction conserves.

Deriving the Born Rule

Both programs regard $P = |\psi|^2$ as something to be derived rather than postulated, and both have invested heavily in doing so. The routes have almost nothing in common.

The Everettian derivations are agent-centred. Deutsch and Wallace ask what credences a rational agent must adopt when facing branching, and show that decision-theoretic axioms — measurement neutrality, branching indifference, and their relatives — force the Born weights. Sebens and Carroll route instead through self-locating uncertainty: an observer post-branching but pre-observation is genuinely ignorant of which branch they occupy, and an epistemic separability principle pins the credences. Zurek’s envariance argument works from entanglement symmetry alone. The persistent objection is that these arguments must import something amplitude-flavoured to get an amplitude-flavoured answer, and whether they succeed is genuinely unresolved rather than settled either way.

The framework’s route is not about agents at all. Three constraints — normalization from coherence conservation, phase covariance from the $U(1)$ loop structure, and consistency of two-stage measurements from the interaction graph — leave exactly one probability rule standing (Born Rule). The uniqueness is a functional equation, confirmed independently by Gleason’s theorem for $d \geq 3$ and covering $d = 2$ where Gleason does not. The same argument fixes the coherence functional itself to $\langle\psi|\psi\rangle$ and forces the Hilbert space structure, so probability is read as the fraction of conserved coherence flowing through an outcome channel — a statement about distribution, not about credence.

That reframing is the substantive point. In the framework, probability never becomes a question about what a rational agent should bet, because there is a conserved quantity being partitioned and the partition is what the number measures.

What Branching Costs

Everett’s payoff for deleting collapse is that observer-independent realism survives: there is a single universal state, and every outcome is equally real somewhere in it. The framework declines that payoff, and the reason is structural rather than aesthetic.

Coherence is finite on every Cauchy slice, integer-quantized in units of $\hbar\omega_0$, and accounted per observer — the ontology is closed, with no external reservoir. Within that accounting, the sheaf structure of the observer network gives a sharp result: the coherence and probability sheaves have unique global sections, and temporal branching is ruled out by acyclicity of the dependency graph. There is one coherence budget and one evolution forward. What is multiple is the outcome sheaf’s local sections — several mutually incompatible but individually consistent outcome assignments, all living inside the single global coherence section, weighted by the unique Born measure. Indeterminacy is contextual, not temporal. Nothing is duplicated.

The corresponding cost is that the framework gives up the view from nowhere. Observer-Relative Objectivity replaces it with a trichotomy: some facts are observer-invariant (total coherence, conservation laws, network topology), most are observer-relative but uniquely constrained, and some are observer-undefined. Wigner’s friend then dissolves without branching, because time is phase ordering on a graph and there is no observer-independent “now” at which two descriptions must be reconciled — only vertices where observers actually meet, and there entanglement structure enforces agreement.

The Pattern: One Ledger, Many Perspectives

The two programs answer the same question — how do you get definite experience out of exact unitary evolution? — with opposite economies. Everett spends ontology to buy an observer-independent world: multiply the branches, and every observer’s definite outcome is a true description of a real part of a single wavefunction. The framework spends observer-independence to buy a finite ontology: keep one closed coherence ledger, and let “the state of the system” be irreducibly indexed by whose relational invariants you are asking about.

Everything else follows from that trade. Because Everett has a universal state and no preferred perspective, the basis has to be recovered from dynamics (decoherence) and the probabilities from agents (decision theory). Because the framework has a conserved quantity and per-observer accounting, the basis comes from what an interaction conserves and the probabilities come from how that conserved quantity partitions. Neither is getting something for nothing. They are paying in different currencies.

Where They Genuinely Part Ways

The disagreement is not about any experiment anyone knows how to run. Both predict exact unitarity, both deny collapse, both reproduce standard quantum statistics.

Where they genuinely part ways is on three structural questions. Is branch structure emergent and approximate, or is basis selection exact? Is probability a rational credence under branching, or a conserved-quantity fraction? And is the multiplicity of outcomes temporal (many branches, one perspective each) or contextual (one evolution, many perspectives on it)? The framework’s answers are the second option in each case, and the sheaf result makes the third one a derived claim rather than a preference — within this framework, temporal branching is not merely unattractive but structurally excluded.

There is also an asymmetry worth naming plainly. Everettianism is an interpretation of quantum mechanics and does not aspire to predict physics beyond it; that is a legitimate scope, not a deficiency. The framework is not an interpretation, and it therefore cannot rest on internal coherence alone — it stakes falsifiable claims downstream, in interferometer noise with a derived angular signature (holographic noise) and elsewhere. Its quantum foundations are judged by whether the derivations hold; the program as a whole is judged at the detector.

What the Framework Takes From It

Everett’s central discipline — apply the dynamics without exception and follow it wherever it goes — is a debt the framework should acknowledge openly, because the framework’s own measurement analysis makes exactly that move: the joint evolution is unitary throughout, and every appearance of collapse is a conditional description of a subsystem. Half a century of Everettian work in making that stance survive contact with actual physics is why the position is available to be inherited at all.

The decoherence program in particular is absorbed rather than displaced: einselection is recovered, not refuted, and the framework’s claim is only that the selection is exact one level down. And the Deutsch–Wallace derivation sets a standard the framework’s own Born-rule argument should be measured against, not exempted from. Both are attempts to remove the same postulate; both should face the same hard question — what did you assume that already knew the answer? — and the framework’s answer is better for having Wallace’s version to compare with.