Post-Selected von Neumann Framework

• The Post-Selected von Neumann Framework is a quantum-measurement method that uses both pre- and post-selection to condition measurement outcomes on future success.
• It generalizes conventional measurement theory by employing subnormalized POVMs and Kraus operators to redefine probabilities and state evolution.
• Operational examples, pointer state engineering, and refined entropy measures demonstrate its potential to enhance estimation accuracy and information flow.

The post-selected von Neumann framework denotes a class of quantum-measurement constructions in which the standard von Neumann scheme—preparation, system–apparatus interaction, and readout—is supplemented by conditioning on a later successful outcome. In its most explicit time-symmetric form, the intermediate system is described not by a single state but by a pair $\langle \psi_f(\theta)||\psi_i(\theta)\rangle$, and measurement statistics are conditioned on both an initial and a final boundary condition. This alters both the admissible measurement objects and the interpretation of information flow: probabilities become conditional on post-selection success, subnormalized operations become physically meaningful, and some outcomes can be excluded from the successful ensemble altogether (Massar et al., 2011).

1. Two-boundary description of measurement

In the standard von Neumann picture, a system is prepared in an initial state, coupled unitarily to a measurement register, and then read out. In the post-selected extension developed for pre- and post-selected ensembles, the corresponding object is a two-time ensemble specified by a pre-selected ket and a post-selected bra,

$$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$

written compactly as

$\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$

This is the operational form of the Aharonov–Bergmann–Lebowitz / two-state-vector formalism: the Measurer acts between a Pre-selector at time $t_i$ and a Post-selector at time $t_f$. A convenient Hilbert-space decomposition uses a system space $H_S$, a measurement register $H_R$ with outcome basis $|k\rangle_R$, and, when needed, an ancillary post-selection qubit $H_P$ initially in $|0\rangle_P$. A unitary $$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$0 acts on $$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$1, and a logical gate forwards the outcome $$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$2 only if the prescribed post-selection succeeds (Massar et al., 2011).

This two-boundary architecture preserves the recognizable structure of von Neumann measurement while changing its semantics. The initial state no longer exhausts the specification of the experiment; the final state participates directly in the operational probabilities. A plausible implication is that the framework is best understood not as a single alternative formalism but as an umbrella description for time-symmetric, conditionally normalized, or explicitly post-selected versions of measurement theory.

2. From POVMs to Kraus operators

For pre-selected-only estimation, the effective measurement is a POVM $$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$3 with

$$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$4

and outcome probabilities

$$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$5

If a fixed post-selected state is added, the appropriate objects become subnormalized POVMs,

$$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$6

with conditional probabilities

$$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$7

The inequality $$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$8 encodes discarded runs: outcomes associated with failed post-selection are removed from the successful ensemble. This is the mechanism by which “certain measurement outcomes can be forced never to occur” among successful runs (Massar et al., 2011).

When both pre- and post-selected states are unknown and part of the estimation problem, POVMs are no longer the natural primitives. The basic measurement objects are Kraus operators $$|\psi_i(\theta)\rangle,\qquad \langle \psi_f(\theta)|,$$9, normalized by

$\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$0

or, with an additional fixed post-selection available as a dump channel,

$\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$1

The central Born-like rule is then

$\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$2

In the unitary implementation,

$\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$3

so the Kraus operators are literally the compression of the global unitary between the initial and final boundaries. In this sense, the post-selected von Neumann framework is a Kraus-operator generalization of the ordinary system–apparatus model rather than a simple relabeling of POVM theory (Massar et al., 2011).

3. Estimation theory and time-symmetric information flow

The estimation problem is formulated by choosing a parameter $\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$4 with prior $\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$5, assuming the functional dependence $\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$6 and $\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$7 is known, and optimizing a decision rule $\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$8 against an average merit function. In the fully pre- and post-selected case,

$\langle \psi_f(\theta)||\psi_i(\theta)\rangle.$9

A central structural result is the relation between information from the future and complex-conjugated information from the past. If $t_i$0 is the complex conjugate of $t_i$1 in a fixed basis, then, with fixed post-selection available, estimating $t_i$2 via a subnormalized CP map is equivalent to estimating the ordinary pre-selected state $t_i$3 via a subnormalized rank-1 POVM. Without fixed post-selection, the implication becomes one-way: the pre-/post-selected task is never harder than the conjugate pre-selected task, and can be strictly easier (Massar et al., 2011).

This asymmetry is operationally nontrivial. The explicit counterexample is a variant of unambiguous state discrimination in which the probability of inconclusive outcomes can be made $t_i$4 with future information but remains $t_i$5 using only pre-selected information. In covariant spin estimation, the distinction can disappear or persist depending on encoding: for parallel spins $t_i$6 with fixed post-selection, the fidelity remains

$t_i$7

exactly as in the ordinary case, whereas for anti-parallel encodings $t_i$8 the reported optimal fidelities $t_i$9, $t_f$0, and $t_f$1 for $t_f$2 exceed the usual anti-parallel-only values. This suggests that post-selection is not uniformly advantageous; its effect depends on the geometry of the encoding and on whether discarded runs can be exploited as a resource (Massar et al., 2011).

4. Operational examples and optical realizations

The simplest operational illustration is unambiguous state discrimination with fixed post-selection. The Measurer performs the usual USD measurement on two nonorthogonal states, but arranges that successful conclusive outcomes are followed by a post-selection-compatible system state while “I do not know” is routed to a state that fails post-selection. Conditional on success, the inconclusive outcome is thereby absent from the observed ensemble. A second canonical example is Aaronson’s PP construction: for the promise problem with $t_f$3 identical qubits in $t_f$4, the sample complexity is $t_f$5 without post-selection but

$t_f$6

with fixed post-selection, reflecting the use of discarded branches as a computational resource (Massar et al., 2011).

A distinct but closely related operational line studies explicit post-selected von Neumann interactions with continuous-variable pointers. In the optical formulation, the impulsive Hamiltonian is

$t_f$7

with unitary

$t_f$8

measurement strength

$t_f$9

and weak value

$H_S$0

For Hermite–Gaussian and Laguerre–Gaussian pointer states, exact all-orders formulas were obtained for the final pointer shifts when $H_S$1 and when $H_S$2. The fundamental Gaussian remains optimal for $H_S$3 in the coupling direction, but LG modes with nonzero OAM generate an induced orthogonal shift $H_S$4 and can improve $H_S$5 for imaginary weak values, especially in the projector case and in the weak regime (Turek et al., 2014).

Recent pointer-engineering variants push this logic further. Superpositions of Gaussian and Laguerre–Gaussian OAM modes yield post-selected pointer states with quadrature squeezing, Wigner-function negativity, and a signal-to-noise ratio advantage in the weak regime when the weak value is anomalous and the coupling is moderate (Yuanbek et al., 2024). Two-mode entangled coherent states used as pointers lead, after post-selected von Neumann coupling to $H_S$6 and $H_S$7, to a superposition of four displaced ECS components; the reported effects include enhanced sum squeezing, negative Hillery–Zubairy entanglement witness values, interference-rich joint Wigner functions, and increased quantum Fisher information as the couplings $H_S$8 increase (Yuanbek et al., 18 Nov 2025).

5. Conditional states and entropy of post-selected ensembles

A complementary formulation represents a post-selected ensemble by a generalized density operator

$H_S$9

which is trace one but generally non-Hermitian. In this representation, weak values become ordinary trace expressions,

$H_R$0

so the generalized density operator plays the same formal role for post-selected weak values that a standard density matrix plays for ordinary expectation values (Salek et al., 2013).

The associated conditional entropy for a fixed post-selection is defined by

$H_R$1

and the average conditional entropy by

$H_R$2

For pure pre-selection this simplifies to

$H_R$3

These quantities can be negative. The paper states the upper bound $H_R$4, attained in the trivial no-post-selection case, and the lower bound

$H_R$5

for a $H_R$6-dimensional Hilbert space, attained when all overlaps are equal. In the three-box example, one post-selection gives

$H_R$7

and the full average is reported as $H_R$8 in base-3 logarithm units. Within this construction, negative conditional entropy quantifies the informational peculiarity of conditioning a pure ensemble on a future outcome rather than on a subsystem (Salek et al., 2013).

6. Projection, refinement, and alternative dynamical reductions

The relation between post-selection and the traditional projection postulates is subtle in the presence of degeneracy. For an observable

$H_R$9

with degenerate spectral subspaces, von Neumann’s original position was that a measurement of $|k\rangle_R$0 does not by itself determine a unique post-measurement state; one must refine $|k\rangle_R$1 by a commuting nondegenerate observable

$|k\rangle_R$2

Under natural assumptions on the hidden post-measurement state and consistency with the Born rule for all such refinements, the coarse-grained state conditioned on outcome $|k\rangle_R$3 is uniquely

$|k\rangle_R$4

namely Lüders’ rule. Because local measurements on one subsystem of an entangled state correspond to degenerate observables on the composite Hilbert space, this reconstruction is directly relevant to teleportation and one-way quantum computing (Khrennikov, 2009).

A more explicitly apparatus-based derivation appears in supmech, where both system and apparatus are quantum systems but the apparatus admits a classical phase-space approximation. Pointer observables

$|k\rangle_R$5

correspond to macroscopically distinct stability domains $|k\rangle_R$6, and observations are made only after the apparatus has settled into one of these domains. Under these observational constraints, off-diagonal terms in the post-interaction system–apparatus state are suppressed, yielding an effective settled state

$|k\rangle_R$7

Conditioning on a pointer sector then reproduces the usual selective projection on the system. In this treatment, the collapse rule is not postulated but emerges from Hamiltonian dynamics, classical approximation of the apparatus, and coarse-grained pointer observables (Dass, 2010).

A different proposal replaces instantaneous projection by a quasilinear evolution law,

$|k\rangle_R$8

with exact solution

$|k\rangle_R$9

This nonlinear scheme preserves ensemble equivalence and therefore no-signalling, retains the Born rule for outcome selection, and for two-level systems converges to the standard projected state outside very narrow structurally unstable parameter regions. In that sense it turns post-selected reduction into a continuous conditional evolution rather than a discontinuous collapse (Rembieliński et al., 13 May 2026).

7. Broader abstractions and formal limits

Post-selection can also be abstracted away from laboratory measurement into computational semantics. In real-time probabilistic and quantum finite automata with postselection, one designates a set of postselection states $H_P$0, requires that the final probability of landing in $H_P$1 be strictly positive, and defines acceptance by conditional normalization,

$H_P$2

This is equivalent to a restart model in which all nonpostselection branches are rerun until a postselection branch occurs. The resulting language classes satisfy

$H_P$3

and zero-error classical postselection collapses back to $H_P$4. Although this setting is not a measurement theory in the physical sense, it isolates the same formal operation: conditioning on a designated successful sector after ordinary probabilistic or quantum evolution (Yakaryilmaz et al., 2011).

At the same time, the literature places sharp limits on how far one can extend the von Neumann rule as an exact functional principle. For a quantization map

$H_P$5

the “Neumannian” condition

$H_P$6

for all $H_P$7 and all $H_P$8 is strong enough that every $H_P$9-linear Neumannian map is Abelian, and any local Neumannian map is Abelian as well. The stated conclusion is that a physically valid quantization map should be neither linear nor local if one insists on exact von Neumann post-composition invariance. For post-selected von Neumann frameworks built directly from self-adjoint operator observables, this functions as a boundary result: strong functional-calculus invariance, linearity, locality, and genuine noncommutativity cannot all be retained simultaneously (Müller, 2019).

Taken together, these developments show that the post-selected von Neumann framework is not a single closed formalism but a technically rich domain organized around one common move: an otherwise standard measurement, state-update, or computational process is redefined by conditioning on a later successful event. In some formulations this produces a two-time Kraus theory of estimation; in others it yields generalized conditional states and entropies, refined derivations of Lüders update, continuous quasilinear reductions, or structured-pointer metrology. What unifies them is the replacement of unconditional readout by conditional normalization on a selected final sector.