The Many-Worlds Interpretation

The Many-Worlds Interpretation confronts a structural embarrassment in quantum mechanics: the theory ships with two dynamical rules that contradict each other. Between measurements, states evolve smoothly and deterministically under $i\hbar\,\partial_t|\psi\rangle = \hat H|\psi\rangle$ . At a measurement, the orthodox projection postulate replaces the superposition with a single eigenstate, chosen at random with Born-rule weight. Everett's 1957 proposal was radical in its economy: delete the second rule entirely and ask whether pure unitary evolution, applied even-handedly to observers and apparatus as quantum systems, already reproduces everything we see. His answer was yes — at the price of taking the universal wavefunction literally as a superposition of coexisting, non-interacting outcomes. The mechanism is entanglement, recalled by name from Quantum Entanglement and Non-Separability. When an apparatus measures a system in a superposition $\sum_n c_n|s_n\rangle$ , unitary evolution produces the entangled sum $\sum_n c_n|s_n\rangle|A_n\rangle$ , and extending the same logic to the observer and environment gives a vast entangled state in which each term carries a fully correlated record of one outcome. Everett called $|A_n\rangle$ the state of the apparatus relative to $|s_n\rangle$ — hence "relative-state formulation." No term is privileged; all coexist as branches. This article owns the relative-state theorem — that a "branch" is exactly a Schmidt term of the entangled state — and shows why decoherence (developed in Quantum Decoherence and the Density Matrix by name) makes the branches stable and quasi-classical. One thing this article does not claim to do is derive the Born rule. In a theory where every outcome happens, the meaning of "outcome $n$ has probability $|c_n|^2$ " is a genuine open problem. We state the leading attempts — Everett's typicality, the Deutsch–Wallace decision-theoretic program, Zurek's envariance — and show, honestly, exactly where envariance fixes the probabilities (the symmetric, equal-amplitude case) and where it does not. Many-worlds trades an ill-defined collapse for a sharp ontology and a sharp puzzle about probability, while remaining experimentally identical to textbook quantum mechanics.

Key takeaways

The Many-Worlds Interpretation (Everett, 1957) takes the unitary Schrödinger equation as the complete and only dynamics — there is no collapse postulate.
Relative-state theorem (owned): a measurement is ordinary entangling unitary evolution $\sum_n c_n|s_n\rangle|A_n\rangle$ ; each "branch" is a Schmidt term, and the observer's reduced state is one copy per branch (recalled from The Schmidt Decomposition and Quantum Entanglement and Non-Separability).
Decoherence defines branches (owned): environmental entanglement diagonalizes the reduced state in the pointer basis, so branches stop interfering and become autonomous and quasi-classical — the answer to the preferred-basis objection (developed in Quantum Decoherence and the Density Matrix and The Pointer Basis and Environment-Induced Superselection by name).
The Born rule is an OPEN problem in MWI: with every branch realized, "probability $|c_n|^2$ " needs justification. Envariance forces $p=\tfrac12$ in the symmetric equal-amplitude case; the general $|c_n|^2$ rule is not settled — this is stated, not claimed derived.
The global state stays pure (no collapse); a subsystem is an improper mixture — the same distinction that separates decoherence from real collapse.
Worked: a $z$ -measurement gives $(|\!\uparrow\rangle|A_\uparrow\rangle + |\!\downarrow\rangle|A_\downarrow\rangle)/\sqrt2$ with $\rho_O = \mathrm{diag}(\tfrac12,\tfrac12)$ ; the equal-amplitude Bell state is envariant under a label swap, forcing $p=\tfrac12$ .
Key idea: MWI keeps only unitary evolution; measurement is branching by entanglement, decoherence makes branches classical, and the meaning of probability in a many-branch world is the theory's central unsolved problem.

The core insight

Everett's proposal is to keep the Schrödinger equation and drop collapse. Measurement is then entanglement: the joint system–apparatus–observer state branches into non-interacting Schmidt terms, each holding a definite record. Decoherence selects the pointer basis and makes branches autonomous, answering the preferred-basis objection. What remains unsolved is probability: deriving the Born weights $|c_n|^2$ in a deterministic theory where all branches occur — envariance handles the symmetric case, the general rule is open.

Where this lives today

Everett's picture, ignored for a decade, is now a mainstream option in the foundations community and a working assumption in parts of quantum cosmology, where there is no external observer to collapse the wavefunction of the universe — a many-worlds or decoherent-histories reading is almost forced. In quantum computing, the intuition that a quantum computer "explores many branches in parallel" is an Everett-flavored heuristic (to be handled with care, since interference, not literal parallel worlds, does the work). And the Born-rule debate is genuinely live: the Deutsch–Wallace decision-theoretic derivation and its critics, Zurek's envariance program, and self-locating-uncertainty approaches (Sebens–Carroll) are active research, each trying to close the gap this article marks as open. Experimentally, MWI is identical to textbook quantum mechanics — no experiment distinguishes them — so the debate is over explanatory economy and the meaning of probability, not data. The rivals against which it is measured — a real dynamical collapse, or added trajectories — are drawn by name in Objective Collapse Theories and De Broglie–Bohm Pilot Wave Theory.

The physical picture

Start from the one move that defines the interpretation: refuse to make an exception for measuring devices. A measuring device is made of atoms; atoms obey the Schrödinger equation; therefore a measuring device obeys the Schrödinger equation. If a system is in a superposition and the device is a quantum system too, then their interaction — a perfectly ordinary unitary — cannot collapse the superposition. It can only correlate the device with each branch of the system. The pointer does not choose a value; it enters a superposition of pointing at each value, each term glued to its own system state. Push the same logic to the experimenter reading the pointer, and to the air molecules and photons in the room, and you get one enormous entangled state whose terms are self-consistent, fully correlated records of distinct outcomes — the "worlds." The natural objection is: why do I only ever see one outcome? The Everettian answer is that "I" am also in the superposition — there is a copy of me in each branch, each seeing a single definite result and each equally real. The branches do not interact because decoherence has diagonalized the reduced state in a robust pointer basis, so the interference terms that would let one branch "feel" another are suppressed to unmeasurable levels. This is what makes the worlds stable and quasi-classical: each looks like a single classical history obeying the Born statistics, and none can recohere with another on any realistic timescale.

Figure

Measurement as branching: one unitary entangles system, apparatus, and observer

The formalism: branching is entanglement

The technical heart of the interpretation is a single identification: a "branch" is a term in the Schmidt decomposition of the entangled measurement state. Once that is granted, everything the interpretation says about definite records and observer copies is bookkeeping on an entangled state — objects already owned by the formalism.

Derivation AThe relative-state theorem: a measurement branches into Schmidt terms

We show that unitary "measurement" turns a product state into an entangled sum whose terms are Everett's relative states, and that the observer's reduced description is one definite record per branch. The strategy is to run the measurement as an entangling unitary (a CNOT for a qubit) and read off the reduced states — no collapse anywhere.

Step 1. Set up the pre-measurement product. Take a system qubit in $|s\rangle = (|\!\uparrow\rangle + |\!\downarrow\rangle)/\sqrt2$ and an apparatus in a ready state $|A_0\rangle$ , so the joint state is the product $|s\rangle|A_0\rangle$ . Unitary evolution acts on the whole; nothing is measured "from outside."

|s\rangle|A_0\rangle = \tfrac{1}{\sqrt2}\big(|\!\uparrow\rangle + |\!\downarrow\rangle\big)|A_0\rangle \;\longmapsto\; \tfrac{1}{\sqrt2}\big(|\!\uparrow\rangle|A_\uparrow\rangle + |\!\downarrow\rangle|A_\downarrow\rangle\big) \qquad (1)

an entangled state — the two terms are Everett's relative states: $|A_\uparrow\rangle$ is the apparatus state relative to $|\!\uparrow\rangle$ . Why this is the whole idea: linearity forbids the unitary from selecting one outcome; it can only correlate the pointer with each, producing a superposition of "pointer reads up" and "pointer reads down." These are the branches, and (1) is literally a Schmidt decomposition with equal coefficients (recalled from The Schmidt Decomposition by name).

Step 3. Add the observer, read the reduced state. Extending the same unitary to an observer $O$ gives $\tfrac{1}{\sqrt2}(|\!\uparrow\rangle|A_\uparrow\rangle|O_\uparrow\rangle + |\!\downarrow\rangle|A_\downarrow\rangle|O_\downarrow\rangle)$ . Tracing out system and apparatus (partial trace recalled from The Reduced Density Matrix and Partial Trace) leaves the observer in $\rho_O = \tfrac12(|O_\uparrow\rangle\langle O_\uparrow| + |O_\downarrow\rangle\langle O_\downarrow|) = \mathrm{diag}(\tfrac12,\tfrac12)$ — a mixed reduced state with one copy of the observer per branch, each holding a definite, self-consistent record. Verified numerically: the CNOT maps $|s\rangle|A_0\rangle$ to $(|\!\uparrow\rangle|A_\uparrow\rangle + |\!\downarrow\rangle|A_\downarrow\rangle)/\sqrt2$ ; the system reduced state is $\mathrm{diag}(\tfrac12,\tfrac12)$ , the three-way observer reduced state is $\mathrm{diag}(\tfrac12,\tfrac12)$ , and the global state stays pure ( $\mathrm{Tr}\,\rho^2 = 1$ ).

Go deeper · Derivation B · 1 minDecoherence defines the branches — answering the preferred-basis objection

Derivation BDecoherence defines the branches — answering the preferred-basis objection

We address the classic objection that "branches" are ambiguous because an entangled state can be re-expanded in any basis. The resolution is that the environment picks out one basis dynamically: the reduced state is diagonal in the pointer basis, and it is only in that basis that branches are stable and non-interfering. The engine is the suppression of the reduced off-diagonal by the overlap of environmental records.

Step 1. Write the state with an environment record. After a system–environment interaction the joint state is $\tfrac{1}{\sqrt2}(|\!\uparrow\rangle|E_\uparrow\rangle + |\!\downarrow\rangle|E_\downarrow\rangle)$ , and the system's reduced density matrix is $\rho_S = \tfrac12\!\begin{pmatrix}1 & \langle E_\downarrow|E_\uparrow\rangle\\ \langle E_\uparrow|E_\downarrow\rangle & 1\end{pmatrix}$ . The off-diagonal is weighted by the environment-record overlap.

Step 2. Let the records become orthogonal. As the environment monitors the system, $\langle E_\uparrow|E_\downarrow\rangle\to0$ , so $\rho_S\to\mathrm{diag}(\tfrac12,\tfrac12)$ — diagonal in the pointer ( $z$ ) basis. Why this fixes the basis: the interference terms that would let the up-branch and down-branch recohere are exactly the off-diagonals, and they are gone. Only the pointer basis, singled out by the interaction, gives stable branches; a superposition basis (say $|\pm\rangle$ ) is not preserved and decoheres. The selection of that basis is developed in The Pointer Basis and Environment-Induced Superselection by name.

Step 3. Draw the conclusion. Branches are not a matter of choice: they are the terms diagonalized by decoherence in the robust pointer basis, and they are autonomous because their mutual interference is suppressed to unobservability. This is the modern repair of Everett's picture — the "worlds" are the decohered pointer-basis branches, not an arbitrary re-expansion. Verified numerically: with orthogonal records $\langle E_\uparrow|E_\downarrow\rangle=0$ the reduced off-diagonal is $0$ ; with partial overlap $0.3$ the off-diagonal is $0.15$ (residual coherence) — the branches are only fully autonomous once the records are orthogonal.

The Born-rule problem — stated honestly

In a theory where every branch happens, saying "outcome $n$ occurs with probability $|c_n|^2$ " is not obviously meaningful: from the god's-eye view, all outcomes occur with certainty. This is the deepest open problem of the interpretation, and the honest position is to say so.

Common trapMWI does not, as of now, derive the Born rule — beware claims that it does

It is tempting to think that because decoherence is "already in the theory," the Born probabilities must be too. They are not. Decoherence is itself unitary and produces all branches alike; it cannot by itself explain why the branch with amplitude $c_n$ should be experienced with weight $|c_n|^2$ rather than, say, counted equally. Three programs attack the gap: Everett's original typicality argument (most observers in a suitable measure see Born statistics), the Deutsch–Wallace decision-theoretic derivation (a rational agent indifferent to branch labels is forced to weight by $|c_n|^2$ ), and Zurek's envariance (environment-assisted invariance, which fixes equal probabilities in symmetric cases by symmetry). Whether any succeeds without hidden circularity is actively debated. The correct claim is therefore modest: MWI reproduces Born statistics if you grant one of these accounts; it does not yet derive the Born rule from unitary dynamics alone. The full measurement problem — why definite outcomes at all — is developed in The Quantum Measurement Problem by name.

Figure

The Born-rule frontier: envariance earns the symmetric case, the general rule is open

Worked examples

Example 1 — a spin- $\tfrac12$ measurement with no collapse (confidence builder). Measure $S_z$ on a system in $|s\rangle = (|\!\uparrow\rangle+|\!\downarrow\rangle)/\sqrt2$ using an apparatus and an observer. Track the branches and the observer's reduced state.

Step 1. Branch. The measurement unitary produces $\tfrac{1}{\sqrt2}(|\!\uparrow\rangle|A_\uparrow\rangle|O_\uparrow\rangle + |\!\downarrow\rangle|A_\downarrow\rangle|O_\downarrow\rangle)$ : two equal-weight, fully correlated branches, one "saw up," one "saw down."

Step 2. Reduce to the observer. Tracing out system and apparatus gives $\rho_O = \mathrm{diag}(\tfrac12,\tfrac12)$ (machine-verified) — a maximally mixed reduced state, one definite-record copy of the observer in each branch. Neither copy is privileged.

Step 3. Check the global purity. The full state remains pure, $\mathrm{Tr}\,\rho^2 = 1$ (machine-verified), while the system reduced state has purity $\tfrac12$ — an improper mixture. This is the exact signature that distinguishes the no-collapse account: nothing has actually reduced; the mixedness is entirely in the ignored correlations, the same improper-mixture structure that decoherence produces.

Example 2 — envariance fixes the equal-amplitude probability (exam level). Take the maximally entangled Bell state $|\Phi^+\rangle = (|00\rangle + |11\rangle)/\sqrt2$ shared between a system and its environment. Show that a symmetry forces the two outcomes to be equiprobable, and explain why this argument does not extend to unequal amplitudes.

Step 1. Build the swap. Relabel the two system outcomes by acting with $X$ on qubit 1: this sends $|0\rangle\leftrightarrow|1\rangle$ on the system. Counteract it by acting with $X$ on qubit 2 (the environment). The combined operation is $(X\otimes X)$ .

Step 2. Check envariance. Compute $(X\otimes X)|\Phi^+\rangle = (|11\rangle + |00\rangle)/\sqrt2 = |\Phi^+\rangle$ (machine-verified): the global state is invariant. A relabeling of the outcomes on the system was undone by a counter-operation on the environment, leaving the world unchanged. Because the two outcomes can be swapped by a symmetry of the global state, they must be assigned equal probability — $p(0) = p(1) = \tfrac12$ (machine-verified: the reduced weights are $(\tfrac12,\tfrac12)$ ).

Step 3. See where it breaks. For an unequal superposition $\sqrt{0.8}\,|00\rangle + \sqrt{0.2}\,|11\rangle$ , the reduced weights are $(0.8, 0.2)$ (machine-verified), and no such label-swap symmetry exists — the two branches are not exchangeable, so envariance gives no verdict. The general $|c_n|^2$ rule must be recovered by a further argument (fine-graining into equal-amplitude sub-branches, or the decision-theoretic route), and whether that closure is non-circular is exactly the open question flagged above. The honest scope: envariance earns the symmetric case; the rest is still contested.

The thesis nobody wanted

Origins · 1 minhugh everett iii and how this was discovered

hugh everett iii

1930–1982 · Princeton, Arlington

The relative-state formulation was the doctoral thesis of a graduate student who then left physics for good. Hugh Everett III, working under John Wheeler at Princeton, proposed in his 1957 dissertation (published in Reviews of Modern Physics the same year) that the collapse postulate could be dropped and the universal wavefunction taken literally, with measurement outcomes as relative states within one never-reducing superposition. The reception was cold — Bohr's circle in Copenhagen was dismissive, and Wheeler, caught between his student and his mentor, softened the presentation. Everett, discouraged, took a career in defense operations research and weapons analysis and never published on quantum mechanics again, dying at 51. His idea was rescued in 1970 by Bryce DeWitt, who gave it the vivid, contentious name "many-worlds" and championed it into visibility. The interpretation's modern strength — the decoherence account of branch stability and the decision-theoretic attack on the Born rule — was built decades after its author had walked away.

History

1957

— Everett's "'Relative State' Formulation of Quantum Mechanics" (Rev. Mod. Phys. 29, 454)

1970

— DeWitt names it "many-worlds" and revives it in Physics Today

1985

— decoherence (Zurek, Zeh) supplies the preferred-basis and branch-stability account

1999

— Deutsch's decision-theoretic argument for the Born weights; Wallace develops it

2005

— Zurek's envariance derivation of equal probabilities in symmetric cases

Going further

References. H. Everett III, Rev. Mod. Phys. 29, 454 (1957) · B. S. DeWitt & N. Graham (eds.), The Many-Worlds Interpretation of Quantum Mechanics (1973) · D. Wallace, The Emergent Multiverse (2012) · D. Deutsch, Proc. R. Soc. A 455, 3129 (1999) · W. H. Zurek, Phys. Rev. A 71, 052105 (2005) (envariance) · Saunders, Barrett, Kent & Wallace (eds.), Many Worlds? (2010).

Provenance. All results are standard. The unitary Schrödinger evolution is owned by Time-Dependent Schrödinger Equation; entanglement and separability by Quantum Entanglement and Non-Separability; the Schmidt decomposition by The Schmidt Decomposition; the partial trace and reduced states by The Reduced Density Matrix and Partial Trace; the pointer-basis selection by The Pointer Basis and Environment-Induced Superselection; decoherence and the improper-mixture distinction by Quantum Decoherence and the Density Matrix — all recalled by name. Owned and shown here: (1) the relative-state theorem, that measurement is an entangling unitary whose Schmidt terms are the branches and whose observer reduced state is $\mathrm{diag}(\tfrac12,\tfrac12)$ , verified numerically with the global state staying pure; (2) decoherence defines the branches, the reduced off-diagonal going to $0$ as environment records become orthogonal (and $0.15$ at overlap $0.3$ ), verified numerically; (3) the envariance argument that a label swap leaves $|\Phi^+\rangle$ invariant and forces $p=\tfrac12$ in the symmetric case, with the unequal-amplitude case shown to escape it. The Born-rule derivation is stated as an open problem, not claimed; the measurement problem itself is developed in The Quantum Measurement Problem, and the rival interpretations in Objective Collapse Theories and De Broglie–Bohm Pilot Wave Theory — all by name.

Builds on: Quantum Entanglement and Non-Separability · The Schmidt Decomposition · The Reduced Density Matrix and Partial Trace · Quantum Decoherence and the Density Matrix

Leads to: The Quantum Measurement Problem · The Pointer Basis and Environment-Induced Superselection · Objective Collapse Theories

If all outcomes happen, in what sense do I "observe" just one?

Because "you" are also in the superposition. After the measurement there is a copy of you in each branch, each having observed a single definite result and each equally real. There is no fact of the matter about which one is "really" you — the question presupposes a single self that the theory denies. Each copy's experience is a single classical outcome with the right Born statistics; the appearance of a unique result is a fact about being in a branch, not about one branch being selected. What remains genuinely puzzling is why the branch you find yourself in should be Born-weighted, which is the open probability problem.

Isn't postulating infinitely many unobservable worlds extravagant?

Everettians argue the opposite: MWI removes a postulate (collapse) rather than adding one, so it is ontologically extravagant but axiomatically economical — you keep only the Schrödinger equation. The "worlds" are not an extra assumption bolted on; they are what the unmodified formalism describes if you take the universal wavefunction literally. Whether trading axiomatic simplicity for a vast ontology is a good deal is exactly the philosophical crux, and reasonable physicists disagree. Rival interpretations make the opposite trade: keep one world, pay with an extra dynamical rule (collapse) or extra variables (Bohmian positions).

Does decoherence solve the measurement problem for MWI?

It solves part of it and not the rest. Decoherence explains why branches are stable, non-interfering, and expressed in a definite (pointer) basis — this answers the "preferred-basis" and "why don't we see superpositions" questions. What it does not do is explain why there are definite outcomes at all (in MWI, there aren't — all branches persist) or why they follow Born statistics. So decoherence is essential machinery for the interpretation but not a complete solution; the residual problems are the probability puzzle and the very meaning of "an outcome," developed in The Quantum Measurement Problem by name.

Can any experiment distinguish MWI from Copenhagen or collapse theories?

Not from Copenhagen or Bohmian mechanics — all three reproduce standard quantum statistics exactly, so no experiment separates them; the choice is interpretive. Genuine dynamical collapse theories (developed in Objective Collapse Theories by name) are different: they modify the Schrödinger equation and therefore predict tiny deviations (spontaneous localization, faint heating) that are being searched for. If such a deviation were found, it would rule out MWI (and standard QM); if collapse is excluded to ever-tighter bounds, the no-collapse interpretations gain ground. So the MWI-vs-collapse question is, in principle, empirical, while MWI-vs-Copenhagen is not.

What is "envariance" and why does it only fix the equal-amplitude case?

Envariance (environment-assisted invariance) is Zurek's observation that for an entangled state, a transformation on the system can sometimes be exactly undone by a transformation on the environment, leaving the global state unchanged. When two outcomes can be swapped this way — as for the maximally entangled $|\Phi^+\rangle$ — they are physically interchangeable, so they must carry equal probability by symmetry. The argument is clean and assumption-light in the symmetric case. But for unequal amplitudes there is no such swap symmetry, so envariance gives no direct verdict; one must fine-grain the unequal case into equal-amplitude sub-branches, and whether that step is non-circular is debated. Hence envariance earns $p=\tfrac12$ for symmetric splits but not the full $|c_n|^2$ rule.

Is the "splitting" a physical event that happens at a definite time?

Not sharply. Branching is not an instantaneous cut but a continuous process: as the system entangles with more and more environmental degrees of freedom, the reduced off-diagonals decay smoothly toward zero, and the branches become effectively autonomous once the environment records are essentially orthogonal. There is no privileged instant of "splitting" and no exact branch count — both are approximate, decoherence-timescale notions. This fuzziness is sometimes raised as an objection (what exactly is a "world"?), and the honest answer is that worlds are emergent, approximate structures in the wavefunction, sharp enough for all practical purposes but not fundamental — much as "a wave" or "a table" is emergent in classical physics.