The Einstein–Podolsky–Rosen paradox, usually abbreviated EPR paradox, is an argument about the completeness of quantum mechanics presented in 1935 by Albert Einstein, Boris Podolsky, and Nathan Rosen. It considers two systems whose measurable properties are strongly correlated even after the systems become widely separated. Under a proposed criterion for identifying elements of physical reality, and the assumption that an action on one system does not immediately change the physical situation of the distant system, the authors concluded that the quantum-mechanical description leaves some real properties unspecified.
The paradox is not an internal mathematical contradiction in quantum theory. It exposes a tension among its statistical predictions, the possibility of a complete description of individual systems, and a principle of locality. Its central resource is quantum entanglement: a joint state that cannot be reduced to independently specified states of its constituent systems. The original argument concerned position and momentum; later versions involving particle spin made its logic easier to analyze and its consequences easier to test.
EPR helped transform a dispute over the meaning of the wave function into an experimental research program. Bell’s theorem, established in 1964, showed that a broad class of locally causal explanations cannot reproduce all quantum predictions. Subsequent experiments have observed the relevant violations of Bell inequalities. These findings sharply constrain local explanations of quantum correlations, while leaving important distinctions among nonlocality, faster-than-light signaling, and interpretations of quantum theory.
Historical setting and the question of completeness
Quantum mechanics acquired its principal mathematical forms in 1925–1926 through matrix mechanics and wave mechanics. Its predictive success was accompanied by disagreement over what its formalism described. According to the statistical interpretation associated with the Born rule, the quantum state determines probabilities for measurement outcomes. It does not ordinarily assign definite values to every measurable quantity before measurement. The resulting question was whether those probabilities reflected an irreducibly indeterminate world or an incomplete description of a more detailed physical situation.
Einstein accepted the empirical power of quantum theory but resisted treating its probabilistic description as the final account of individual systems. His objections were not simply objections to randomness. He also emphasized the independent physical status of spatially separated systems. If two systems no longer interact, a choice of measurement on one should not determine what is physically real about the other. This concern connected quantum foundations with the causal structure of special relativity, although the EPR paper did not derive its argument directly from relativistic field equations.
The paper “Can Quantum-Mechanical Description of Physical Reality Be Considered Complete?” appeared in Physical Review on 15 May 1935. Its authors proposed two requirements that should be kept distinct. A complete theory should represent every element of physical reality, while a sufficient criterion for recognizing such an element is the ability to predict a quantity with certainty without disturbing the system concerned. The argument then used a correlated two-particle state to identify properties allegedly satisfying that criterion but not jointly specified by a quantum state.
The authors did not propose a detailed replacement theory. They argued that a more complete description should be possible, but the paper did not supply equations for additional variables or a mechanism producing the observed correlations. Its conclusion was therefore a diagnosis of incompleteness rather than a developed alternative to quantum mechanics. Moreover, Einstein’s later explanations often emphasized the incompatibility between a complete wave-function description and the independent reality of distant systems more directly than the published paper did.
The debate quickly expanded. Niels Bohr published a response in 1935, defending quantum mechanics and challenging the EPR use of “without in any way disturbing a system.” Erwin Schrödinger, also writing in 1935, introduced the terminology of entanglement and developed the related concept of steering. These exchanges established distinctions among correlation, physical disturbance, state assignment, and completeness that remain central to the subject.
The original position–momentum argument
The original EPR construction uses two particles with correlated positions and momenta. Denote their position observables by and their momentum observables by . An idealized joint state can have a precisely specified relative position and total momentum:
Here and are fixed constants. Although each particle’s own position and momentum are incompatible observables, these two collective quantities commute:
This follows from , with the reduced Planck constant. Quantum mechanics therefore permits the collective quantities to be jointly sharp in the idealized construction.
Suppose the particles have separated sufficiently that a measurement on particle 1 can be regarded as not disturbing particle 2. Measuring gives a prediction with certainty. Alternatively, measuring gives with certainty. Under the EPR reality criterion, the first procedure establishes position as an element of reality for particle 2, and the second establishes momentum as an element of reality for that same particle.
Only one of these alternative measurements is performed in a particular run. The argument does not claim that an experimenter can jointly measure and with arbitrary precision. Instead, it reasons that the distant particle’s real properties cannot depend on which measurement is freely chosen on particle 1. If neither choice changes particle 2, both quantities apparently qualify as real before either measurement. Yet no ordinary one-particle wave function assigns simultaneously sharp position and momentum. Their spreads obey the uncertainty principle,
EPR therefore concluded that the wave function does not completely represent the physical reality of particle 2.
The construction involves idealizations. A state with exactly sharp relative position and total momentum is not a normalizable finite-energy two-particle wave packet on an infinite line. It is a generalized state, analogous to an exact momentum eigenstate. Physical experiments instead use normalizable states with finite correlation widths. These approximations preserve the essential issue, but perfect certainty must then be replaced by quantitatively bounded inference errors.
The apparent clash with uncertainty is consequently not a direct violation of quantum mechanics. The uncertainty relation constrains the quantum state of a system, whereas EPR draws conclusions about underlying properties by combining alternative measurements, certainty, and locality. The paradox arises from that combination, not from a prediction that standard quantum measurements themselves evade the uncertainty relation.
Reality, locality, and the logical structure of the argument
The EPR criterion states, in substance, that if a physical quantity can be predicted with certainty without disturbing the system, a corresponding element of physical reality exists. It is a sufficient criterion, not an exhaustive definition of reality. EPR did not claim that only perfectly predictable quantities are real. The criterion was intended to identify at least some properties whose existence should not depend on controversial assumptions about the measurement process.
The phrase “without disturbing” carries much of the argument’s force. In a classical example, opening one of two boxes containing oppositely colored objects reveals the other object’s color without changing it. Quantum entanglement permits similarly certain predictions for different, incompatible kinds of measurement. If spatial separation guarantees that the distant system is not physically altered by the local choice, then those alternative predictions seem to reveal different pre-existing properties of one and the same system.
Locality and separability are related but not identical ideas. Locality concerns constraints on causal influence between regions. Separability concerns whether the physical state of a composite system is fully determined by the physical states of its parts. An entangled quantum state is nonseparable in this latter mathematical sense. Whether that nonseparability also entails a particular kind of nonlocal physical process depends on additional assumptions about what quantum states represent and how outcomes arise.
The EPR reasoning also uses a counterfactual comparison: a position measurement was possible, and a momentum measurement was possible, even though only one was made. The inference is not merely that every imaginable unperformed measurement has an outcome. Rather, it is that certainty of remote prediction, combined with the absence of remote disturbance, licenses an attribution of an existing value. Debates about EPR often become confused when this inferential structure is replaced by a blanket premise called “realism.”
Accordingly, the conclusion is conditional. If the EPR reality criterion is accepted, and if the relevant locality assumption holds, the standard wave function is incomplete. A defender of completeness may reject the criterion’s application to incompatible experimental arrangements, reject the independent-state picture of distant systems, or adopt an interpretation in which the premises require reformulation. EPR alone does not experimentally decide among these responses.
This distinction became especially important after Bell. Expressions such as “local realism has been disproved” can compress several logically different assumptions into one label. Bell-type results concern precise restrictions on joint probabilities, not a general experimental refutation of the existence of an external world. EPR’s lasting contribution was to identify how remote predictability could turn apparently philosophical commitments about reality and locality into constraints on physical explanation.
Bohr’s response and Schrödinger’s steering
Bohr’s 1935 reply accepted that there need be no ordinary mechanical disturbance of the distant particle during the final measurement. His objection concerned the meaning of the conditions under which a physical quantity could be predicted. In his account, an experimental arrangement defines which statements about a system can be made unambiguously. Alternative arrangements for position and momentum cannot simply be combined into a single classical description of independently existing attributes.
This response drew on complementarity, Bohr’s view that mutually exclusive experimental arrangements can be necessary for a full account of quantum phenomena. Position and momentum experiments reveal different aspects of a system, but their results do not license simultaneous assignment of all the corresponding classical properties. For Bohr, the EPR criterion overlooked the dependence of meaningful predictions on the complete experimental context, even where that dependence was not a physical impulse transmitted to the remote object.
The reply has remained controversial as an interpretation of the argument. It does not provide an additional dynamical interaction that mechanically disturbs particle 2. Instead, it disputes the inference from alternative prediction procedures to jointly attributable properties. Consequently, the disagreement between EPR and Bohr was not simply about whether one particle exerted an unnoticed force on the other. It involved the role of experimental context in defining physical quantities and the status of the quantum description itself.
Schrödinger identified another striking feature of the same situation: by choosing a measurement on one subsystem, an observer can prepare different conditional state descriptions for the other. He called this effect steering. For example, measurements in one basis can sort a distant ensemble into one collection of pure states, while a different measurement sorts it into another collection. Both collections can correspond to the same unconditioned density matrix.
For a maximally entangled pair of two-level systems, the remote state averaged over all local outcomes is , where is the identity operator. A measurement along one spin direction can conditionally prepare eigenstates of that component; another measurement can conditionally prepare eigenstates of a different component. Without receiving the outcome information, however, the distant observer still describes the same maximally mixed ensemble. Steering therefore concerns correlations and conditional preparation, not controllable transmission of a chosen pure state.
Modern steering theory gives this idea an operational definition. A state is steerable from one party to another when the observed conditional states cannot be explained by a model in which the receiving party already possesses a local quantum state and the other party merely reveals classical information about it. This produces a hierarchy: Bell-nonlocal correlations imply steering, and steering implies entanglement, while the converses do not hold in general. Unlike ordinary entanglement, steering can also be directional.
Bohm’s spin formulation and entangled pairs
In 1951 David Bohm presented a particularly clear version of the EPR argument using spin. Instead of continuous position and momentum variables, it employs two spin- particles in a total-spin-zero state. For a chosen reference axis, the state is
This singlet state is rotationally invariant: its total spin vanishes regardless of the axis used to describe it. If both observers measure spin along the same axis, their results are perfectly anticorrelated.
For an electron, an ideal measurement of a spin component yields either or . If Alice measures the first particle and obtains , she can predict that Bob will obtain along the same axis. She could instead choose a different axis and make an equally certain prediction for the corresponding component of Bob’s spin. Applying EPR’s locality reasoning therefore appears to assign definite values to multiple incompatible spin components.
The spin version avoids the nonnormalizable states used in the ideal position–momentum argument. Its state belongs to a finite-dimensional Hilbert space and is physically preparable to high accuracy. It also makes the connection with Bell tests direct. Perfect anticorrelation along matching axes motivates the EPR inference, while correlations between different axes provide the additional information needed to distinguish quantum predictions from local models.
Assign the normalized outcomes to the two spin results. For measurement directions represented by unit vectors and , quantum mechanics predicts the singlet correlation
Thus matching directions yield , orthogonal directions yield , and opposite directions yield . Each individual observer nevertheless sees an unbiased sequence of positive and negative results. The distinctive structure lies in how the two sequences correlate for different pairs of settings.
Experiments often use entangled photons rather than massive spin- particles. Photon polarization provides another two-level degree of freedom, but the angular dependence differs: for an appropriate polarization-entangled state, correlations involve twice the relative analyzer angle. Other platforms include trapped ions, atomic ensembles, superconducting circuits, and electron spins in solids. They share the abstract structure of joint states and incompatible local measurements without being physically identical realizations.
“EPR pair” is consequently used more broadly than the original 1935 setup. In quantum-information terminology it often means a maximally entangled pair of two-level systems, especially one of the four Bell states. This usage identifies a useful entanglement resource rather than a literal implementation of the original position–momentum thought experiment.
Bell’s theorem and quantitative tests of locality
Bell’s 1964 result changed the problem from whether a more complete account was desirable to whether a locally causal account could reproduce quantum predictions. Consider observers Alice and Bob, measurement settings , outcomes , and a variable representing a sufficiently complete specification of relevant prior conditions. A Bell-local model has the factorized form
An additional assumption is measurement independence: the distribution is independent of the settings, so that .
Factorization expresses more than the absence of an observable signal. It requires the local outcome probabilities, once the relevant prior specification is supplied, not to depend on the distant setting or outcome. Both deterministic models, in which fixes outcomes, and stochastic models, in which it only fixes probabilities, can satisfy this condition. Bell’s theorem is therefore not restricted to a narrow picture of particles carrying simple, predetermined instruction lists.
A widely used test is the CHSH inequality, introduced in 1969 by John Clauser, Michael Horne, Abner Shimony, and Richard Holt. Each observer chooses between two settings. For outcomes normalized to , define
Every Bell-local model satisfying the stated assumptions obeys
For deterministic outcomes, this follows because is always or ; averaging cannot exceed those bounds. Stochastic local models obey the same inequality.
Quantum mechanics predicts violations. For a singlet, suitably selected coplanar spin directions give , the maximum quantum value, known as Tsirelson’s bound. For example, Alice can choose two orthogonal directions, while Bob chooses their normalized sum and difference. The resulting correlations cannot all be reproduced by a model with Bell’s factorization and measurement independence.
Bell’s theorem is stronger than EPR’s original incompleteness claim. Adding locally acting hidden variables is not enough to recover all quantum predictions. Indeed, Bohm’s 1952 pilot-wave theory, an explicit hidden-variable theory reproducing quantum phenomena, is nonlocal in its dynamics. Conversely, rejecting deterministic outcomes by itself does not evade the inequalities, because stochastic local theories are also constrained.
An experimental violation excludes the conjunction of the assumptions used to derive the tested bound, subject to the validity of the experimental analysis. It does not by itself select a unique ontology, prove a usable faster-than-light influence, or decide how quantum states should be interpreted. The strength of Bell’s work lies in making a clearly specified class of explanations empirically distinguishable from quantum mechanics.
Experimental development and the closure of loopholes
Early experiments tested whether polarization correlations agreed with quantum predictions and violated appropriate Bell inequalities. In 1972 Stuart Freedman and John Clauser reported a test using photons from an atomic cascade. Alain Aspect and collaborators conducted a series of experiments in the early 1980s, including a 1982 experiment with time-varying analyzer settings. These studies strongly supported the quantum predictions and helped establish Bell testing as an experimental discipline.
Two practical issues shaped later work. The detection loophole arises when only a small fraction of emitted systems are detected. If detected events are not representative of the full ensemble, a local model can sometimes imitate a violation obtained by assuming fair sampling. The locality loophole arises when setting choices or outcomes on one side could be communicated to the other through an ordinary light-speed-or-slower process during the relevant interval. Closing it requires appropriate spatial separation and carefully timed choices and measurements.
In 1998 Gregor Weihs and collaborators performed a photon experiment with rapid, independently generated analyzer choices and spacelike-separated measurement events. Other experiments achieved high detection efficiencies in systems such as trapped ions but did not simultaneously provide the separation needed to close the locality loophole. Progress therefore involved combining technologies that had initially solved different parts of the problem.
In 2015 three prominent experiments closed the main detection and locality loopholes within individual tests. Bas Hensen and collaborators used electron spins in diamond defects separated by approximately 1.3 kilometres, with entanglement heralded through an intermediate optical station. Marissa Giustina and collaborators and Lynden Shalm and collaborators used high-efficiency photonic arrangements. Their results violated suitable Bell inequalities without relying on the fair-sampling assumptions that had limited many earlier photon experiments.
“Loophole-free” is a conventional term with a specific scope. Such experiments still require assumptions about the experimental devices, the recorded settings and outcomes, and the statistical procedure. They also cannot rule out every conceivable correlation between measurement settings and hidden conditions in their remote common past. Later tests have used astronomical sources to constrain some possible common-cause explanations, but no finite experiment can eliminate unrestricted conspiratorial correlations by observation alone.
Modern analyses also account for finite sample sizes, imperfect state preparation, and possible memory between successive trials. A valid statistical test need not assume that every trial is independent and identically distributed. In 2022 the Nobel Prize in Physics was awarded to Aspect, Clauser, and Anton Zeilinger for work on entangled photons, violations of Bell inequalities, and quantum information science, reflecting the connection between foundational tests and new experimental capabilities.
No-signaling and competing interpretations
Entanglement does not allow an observer to send a chosen message instantaneously. The relevant constraint is the no-signaling principle: the probability distribution for outcomes at one location is unchanged by the measurement setting selected at a spacelike-separated location. For a joint distribution this requires, for example,
independent of Alice’s setting . Quantum correlations obey this requirement even when they violate Bell inequalities.
The density-matrix formalism shows why. If is a joint state, Bob’s local state is
Any local trace-preserving quantum operation performed by Alice, with its outcomes ignored, leaves unchanged. Conditioning on a particular outcome can change Bob’s assigned state, but Alice cannot choose which random outcome occurs. Bob must receive an ordinary classical message identifying that outcome before he can sort his data into the corresponding conditional ensemble.
This distinction separates an operational prediction from an interpretation of state change. A collapse description updates the joint state when an outcome is obtained. Whether this update represents a physical event, a change in information, or an effective description depends on the interpretation. In relativistic discussions, assigning a collapse to a global instant is especially delicate because different observers need not agree on the time ordering of spacelike-separated events.
Different interpretations respond to EPR in different ways. Bohmian mechanics supplements the wave function with particle configurations and uses explicitly nonlocal dynamics. Everettian approaches retain unitary evolution and describe measurement through correlations among systems, apparatus, and observers rather than a unique dynamical collapse. Operational approaches emphasize probabilities for experimental outcomes without requiring a classical-like inventory of underlying properties. Relational and agent-centered interpretations further revise what it means to assign a state to a distant system.
None of these brief characterizations constitutes a separate experimental resolution of every issue raised by EPR. In particular, Bell violations do not show that all interpretations make different empirical predictions. Many recover the same laboratory statistics while differing in ontology, the status of probabilities, and the account of individual outcomes. Quantum decoherence explains how interaction with an environment suppresses observable interference between certain alternatives, but does not by itself uniquely settle the measurement problem.
No-signaling is also weaker than Bell locality. Hypothetical probability distributions can satisfy no-signaling while violating CHSH more strongly than quantum mechanics permits, up to the algebraic limit . Thus the relativistic prohibition on controllable superluminal communication does not alone determine the quantum correlation bound. Explaining why physical correlations occupy the quantum range, rather than the larger no-signaling range, remains an important subject in quantum foundations.
Continuous-variable realizations and quantum-information uses
The original position–momentum structure survives directly in continuous-variable experiments. Optical fields possess conjugate quadrature observables analogous to position and momentum. In two-mode squeezed states, selected combinations of the two modes’ quadratures have reduced fluctuations, allowing one observer to infer the other’s results more accurately than would be possible for an uncorrelated system. Unlike the ideal EPR state, laboratory states have finite squeezing, finite energy, and nonzero measurement noise.
A useful quantitative criterion was developed by Margaret Reid in 1989. Let and denote root-mean-square errors in estimating Bob’s position-like and momentum-like observables from Alice’s respective measurements. For canonical variables satisfying , an EPR-steering demonstration can take the form
The bound concerns inference errors based on alternative remote measurements, not a single system’s simultaneous intrinsic position and momentum spreads. Its violation excludes a local-quantum-state explanation for Bob’s system; it does not violate Bob’s ordinary uncertainty relation.
In 1992 Z. Y. Ou and collaborators reported an optical realization of this continuous-variable EPR criterion using correlated field quadratures. Subsequent experiments extended steering tests to other optical configurations and to material systems. Such work also made clear that entanglement, EPR steering, and Bell nonlocality require different experimental certifications: demonstrating a nonseparable state is not automatically a demonstration of every stronger form of nonclassical correlation.
Entangled pairs became resources for quantum teleportation, proposed in 1993 by Charles Bennett and collaborators. In the standard protocol, two parties share an entangled pair; the sender jointly measures an unknown input system and one member of that pair, then communicates a classical result. The receiver applies a result-dependent operation to recover the input state. For a single qubit, the standard protocol requires two classical bits. The receiver cannot recover the unknown state before that message arrives, so teleportation does not bypass no-signaling.
In entanglement swapping, a joint measurement on members of two initially separate entangled pairs can conditionally entangle the remaining systems, even though those systems have not interacted directly. This extends the EPR structure to networks and underlies proposals for distributing entanglement over long distances. Quantum repeaters seek to combine such operations with storage and entanglement purification or error-correction methods to overcome channel losses.
Bell correlations also support device-independent quantum information, in which an observed violation can certify properties without a detailed model of the internal measurement devices. Applications include randomness certification and certain cryptographic protocols. These certifications still require explicit assumptions, including suitable isolation of the devices, valid setting choices, and adequate statistical treatment. Their distinctive feature is that the experimentally observed correlation pattern, rather than a complete microscopic description of the apparatus, supplies part of the guarantee.