Quantum Consensus Principle: A Thermodynamic Theory Of Quantum Measurement

---

Full Text

The Quantum Consensus Principle

A Thermodynamic Information Principle for Quantum Measurement

Emergent Collapse from Large-Deviation Consensus Dynamics in Open Quantum Systems

Sven Düring

Quantum Consensus Principle Project

March 2026

Abstract

Standard formulations of quantum measurement do not by themselves provide a universally accepted dynamical account of single-outcome selection. Within stan- dard open-system quantum mechanics, the Quantum Consensus Principle (QCP) proposes a unied framework connecting microscopic detector dynamics, operational outcome statistics, and apparatus dependence. Starting from a microscopic Hamiltonian description of system, apparatus, and environment dynamics, we derive a deformed positive operator-valued measure (POVM)

d X

d X

j=1 Sij Πj, Sij ≥0,

i=1 Sij = 1 ∀j, (1)

Ei =

whose weights depend on two canonical functionals of the apparatus response: the redundancy rate ˜Ri and the noise susceptibility χi. These quantities, dened in the BogoliubovKuboMori (BKM) information geometry of the apparatus, are tied to microscopic Hamiltonian parameters via GreenKubo transport coecients a(T, η, J) and b(T, η, J). Within the admissible class specied in the companion Supplement [companion Supplement]causal CPTP dynamics on process-tensor path space, data-processing inequality (DPI) monotonicity, exact concatenation additivity, convexity/lower semi- continuity, and the stated KMS/regularity assumptionsthe eective selection func- tional is unique up to ane gauge and, after calibration of the canonical sucient statistics, takes the linear form Φi = a ˜Ri + bi χi. Competing nonlinear selectors are excluded within that explicit admissible class. We establish that the conditioned system state forms a Hellinger contractive supermartingale, converging almost surely (i.e. with probability one) to a unique pointer state Πi∗. Outcome probabilities follow from a path-space large-deviation principle, reproducing Born's rule in the apparatus-neutral limit and predicting mea- surable deviations for biased apparatuses. QCP yields a non-monotonic collapse timescale τcoll(κ) with a unique optimal measurement strength κopt = a/b. These results identify quantum measurement as a thermodynamic consensus phenomenon,

deriving collapse, outcome statistics, and apparatus dependence directly from Hamil- tonian physics without modifying quantum mechanics.

Scope. The present main paper states the physical architecture of QCP and its front-end consequences in the Markovian pointer-preserving regime. The full ad- missibility, uniqueness, calibration, process-tensor extension, and counterexample- exclusion proofs are carried by the companion Supplement [companion Supplement] and notes [companion Born addendum; Necessity note]. The strongest claims are conditional on those standing assumptions and calibration conditions, rather than unrestricted statements about arbitrary detector models.

Contents

1 Introduction 3

2 Microscopic Model and Conditional Dynamics 3 2.1 Total Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2.2 Lindblad generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 Canonical Scores: Redundancy and Susceptibility 5

4 The Quantum Consensus Potential 5

5 Large-Deviation Derivation of Outcome Probabilities 6 5.1 Operational probability law and neutral limit . . . . . . . . . . . . . . . . 6

6 Emergent Collapse Dynamics 7

7 Non-Markovian Robustness 8

8 Predictions and Falsiability 8 8.1 Non-monotonic collapse rate . . . . . . . . . . . . . . . . . . . . . . . . . . 8 8.2 Born deviations for biased apparatus . . . . . . . . . . . . . . . . . . . . . 8 8.3 Microscopic auditability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

9 Comparison with Existing Frameworks 9

10 Discussion 9

11 Conclusion 10

1 Introduction

Quantum measurement presents an unresolved structural gap: the unitary Liouvillian dy- namics of quantum systems does not produce single outcomes, yet experiments do. Projec- tion postulates [1, 2, 3] formalize collapse without mechanism; Everettian branches [4] pre- serve unitarity at the cost of ontic multiplicity. Objective collapse theories [5, 6, 7] modify the Schrödinger equation, and Bohmian mechanics [8] introduces hidden conguration- space trajectories. None explain how real detectors amplify microscopic information into macroscopic consensus. Open quantum systems theory [9, 10, 11, 12] and decoher- ence/einselection approaches [13] provide precise dynamical tools, but no existing frame- work derives the emergence of a unique pointer state, the Born rule, and the apparatus dependence of measurement statistics from a single principle. Here we introduce the Quantum Consensus Principle (QCP): a universal information- theoretic mechanism by which measurement outcomes arise from large-deviation consen- sus selection in macroscopic detectors. QCP produces a mathematically rigorous collapse, quantitative timescales, auditability from microscopic Hamiltonians, and experimentally accessible deviations from Born statistics. Detailed derivations and mathematical proofs are provided in the companion Supple- ment [companion Supplement]. A separate companion note [companion Necessity note] provides the complementary necessity statement: under explicit bare recorded unitary hypotheses, deterministic state-only one-world outcome selection is excluded on the reach- able recorded state space.

2 Microscopic Model and Conditional Dynamics

2.1 Total Hamiltonian

Consider a measured system S, a macroscopic apparatus A, and a thermal environment E governed by the total Hamiltonian

H = HS + HA + HE + Hint. (2)

Denition 2.1 (Pointer projectors). The pointer projectors {Πi}d i=1 ⊂B(HS) form an orthogonal projection-valued measure (PVM) on the system Hilbert space HS:

d X

ΠiΠj = δijΠi, Πi = Π† i,

i=1 Πi = 1S. (3)

Denition 2.2 (Positive operator-valued measure (POVM)). A POVM on HS is a set {Ei}d i=1 of positive semidenite operators satisfying Ei ≥0 and Pd i=1 Ei = 1S. The probability of outcome i for a system in state ρ is P(i|ρ) = Tr(Eiρ).

The interaction term has the pointer-selective coupling form

d X

j=1 Πj ⊗Bj, (4)

Hint =

where Bj ∈B(HE) are bath operators satisfying KMS detailed-balance relations. Each channel j couples the pointer sector Πj to its own bath operator Bj.

2.2 Lindblad generator

Under the standard weak-coupling, rotating-wave, and coarse-graining approximations (collectively the BornMarkovSecular approximation [9, 14, 17]), the conditional dy- namics of the reduced system state ρt := TrE(ρtot,t) is governed by the stochastic master equation

 ΠiρtΠi

d X

 dNi(t) −λi(ρt) dt
, (5)

dρt = L(ρt) dt +

Tr(Πiρt) −ρt

i=1

with Lindblad generator

d X

2{L† iLi, ρ}  , (6)

 Li ρ L† i −1

L(ρ) = −i[Heff, ρ] +

i=1

and jump operators

Li = √κi Πi, κi = 2π ∥Ai∥2 SE(ωi) > 0. (7)

Here SE(ω) is the bath spectral density, Ai are the coupling operators entering the interac- tion Hamiltonian, and {dNi(t)} are Poisson increments with intensities λi(ρ) = κi Tr(Πiρ). Remark 2.3 (Coupling operators). The Golden-Rule rates (7) involve the coupling opera- tors Ai that appear in the Supplement's factored interaction form Hint = P

j Πj ⊗Aj ⊗B, where B is a collective bath operator. Identifying Bj := Aj ⊗B in the appropriate contraction recovers the channel-specic form (4) used here, with κi = 2π∥Ai∥2SE(ωi).

Assumption 2.4 (QND / Pointer invariance). The eective Hamiltonian Heff = HS+HLS (where HLS is the Lamb shift) commutes with every pointer projector:

[Heff, Πi] = 0 for all i = 1, . . . , d. (8)

This quantum non-demolition (QND) condition ensures that the pointer basis is preserved under the coherent part of the dynamics. The jump rates κi follow directly from Fermi's Golden Rule and the bath spectral density SE(ω); upward/downward rates for non-QND couplings obey KMS relations [9]. QCP uses no phenomenological parameters: all appara- tus asymmetries enter through κi and through correlation functions dening the canonical scores below.

Standing assumptions. The following assumptions, inherited from the companion Sup- plement, are in force throughout this paper: (A1) Valid Lindblad dynamics.

The reduced system state ρt ∈S(HS) evolves via (6) with jump operators (7). (A2) Pointer invariance / QND.

[Heff, Πi] = 0 for all i (Assumption 2.4); equivalently [HS, Πi] = 0, since [HLS, Πi] = 0 under the Secular approximation. (A3) Dephasing dominance.

O-diagonal elements of ρ in the pointer basis decay exponentially under the dy- namics, guaranteeing classicalization of trajectories. (A4) Identiability.

The KullbackLeibler divergence rate between trajectory statistics generated by any two distinct pointer states Πi ̸= Πj is strictly positive. In addition, the bath assumptions (BKM/KMS) and the regularity assumption (A∗) of the companion Supplement remain in force.

3 Canonical Scores: Redundancy and Susceptibility

Macroscopic detectors amplify microscopic alternatives via large networks of degrees of freedom. For each output channel i, QCP denes two canonical scores in terms of current operators on the apparatus Hilbert space. The information current Ii measures the rate at which the apparatus subsystem coupled to channel i produces distinguishable records; concretely, it is the operator whose BKM norm gives the mutual-information production rate between pointer sector i and the environment fragment. The noise current Bi mea- sures the rate at which thermal uctuations and entanglement-induced noise degrade the recorded information in channel i.

Denition 3.1 (Redundancy rate). The redundancy rate ˜Ri is the asymptotic rate at which stable, distinguishable information about outcome i is produced in the apparatus:

˜Ri = ⟨Ii, Ii⟩BKM, (9)

where Ii is the information-current operator and ⟨·, ·⟩BKM denotes the BogoliubovKubo Mori (BKM) inner product [15] on the apparatus state space.

Denition 3.2 (Noise susceptibility). The noise susceptibility χi is the sensitivity of the apparatus record to thermal and entanglement-induced noise:

χi = ⟨Bi, Bi⟩BKM, (10)

where Bi is the noise-current operator.

Denition 3.3 (GreenKubo coecients). The transport coecients a(T, η, J) and b(T, η, J) entering the QCP potential are given by linear-response integrals [22]:

a = Z ∞

0 ⟨Ii(t) Ii(0)⟩ρE dt, b = Z ∞

0 ⟨Bi(t) Bi(0)⟩ρE dt, (11)

where ⟨·⟩ρE denotes the expectation with respect to the KMS equilibrium state of the bath at inverse temperature β. These link Φi directly to the microscopic Hamiltonian. The scores ( ˜Ri, χi) are static information-geometric quantities, while a, b quantify dynamical transport.

Remark 3.4 (Notation convention). We use the standard Lindblad normalization Li = √κi Πi, where κi are Golden-Rule jump rates. For channel-resolved outcome competition we write Φi = a ˜Ri + biχi. In symmetric or one-parameter calibration scans we use the reduced notation Φscan(κ) := b χ(κ) −a ˜R(κ) and dene the scalar scan functional as a coupling-tuning functional. This is not a redenition of the channel selector Φi; the sign inversion reects the cost-minus-gain interpretation appropriate for locating the minimum of the scan [companion Supplement, Notation convention, p. S-3].

4 The Quantum Consensus Potential

Theorem 4.1 (Admissible linear selector; main-paper form). Within the explicit ad- missible class dened in the companion Supplement [companion Supplement]causal CPTP/process-tensor dynamics, DPI monotonicity, exact concatenation additivity, con- vexity/lower semicontinuity, and the stated support/KMS regularity assumptionsthe

QCP selection functional is unique up to ane gauge. After identication of the two canonical sucient statistics ˜Ri and χi, this yields the calibrated linear form

Φi = a ˜Ri + bi χi. (12)

The stronger path-space uniqueness and no-competing-divergence statements are proved in the Supplement [companion Supplement, Theorem 10.1, Corollary 10.2]. Competing nonlinear selectors are excluded within the above admissible class; outside that class, no such claim is made.

Remark 4.2 (Non-monotonic collapse time). When scanning over measurement strength κ, the eective scan potential Φscan(κ) = b χ(κ)−a ˜R(κ) has a U-shaped prole with a unique minimum at κopt = a

b. (13)

Within the calibrated scan family considered here, this structure underlies a non- monotonic collapse time τcoll(κ) ∼1/γmin(κ). This prediction is absent from all stan- dard models (Copenhagen, Many-Worlds, Bohm, GRW). See [companion Supplement, Section 9.2] for the detailed derivation.

5 Large-Deviation Derivation of Outcome Probabilities

Let ρt denote the conditional state trajectory. Using a Doob h-transform of the path mea- sure [18] (see [companion Supplement, Section 4] for the full construction), we establish the trajectory-level large-deviation principle [16]

P(ρT →Πi) ≍exp  −T Φi  , (14)

where ≍denotes logarithmic equivalence in the sense of large-deviation theory [16]:

lim T→∞ 1 T log P(ρT →Πi) = −Φi.

Normalization denes a column-stochastic response matrix S and the QCP POVM:

d X

d X

i=1 Sij = 1 ∀j, P(i|ρ) = Tr(Eiρ). (15)

j=1 Sij Πj, Sij ≥0,

Ei :=

5.1 Operational probability law and neutral limit

In QCP, outcome probabilities are dened operationally as asymptotic frequencies of stochastic trajectories (see [companion Born addendum, Denition 3.1]). They are not postulated at the level of measurement outcomes. For a general calibrated instrument, the exact output law is

P(I∗= i) = Tr(Eiρ0), Ei = X

j Sij Πj, (16)

with S column-stochastic.

Denition 5.1 (Neutral apparatus). The apparatus is neutral if S = I (the identity matrix), equivalently Ei = Πi for all i. By Corollary 2.6 of the Supplement [companion Supplement], this is the limit in which Φi = const across channels.

In the neutral case, the posterior process Mi(t) := Tr(Πiρt) is a bounded martingale ([companion Born addendum, Theorem 4.1]), and together with almost-sure collapse to pointer sectors this yields P(I∗= i) = Tr(Πiρ0). (17)

Thus the Born rule is the exact trajectory-frequency law of the neutral instrument, not an independent probability postulate [companion Born addendum]. Away from neutrality, rst-order deviations are stated only for smooth response-matrix families S(ε); see [com- panion Born addendum, Theorem 6.3]. Remark 5.2 (Born is not presupposed). Corollary 6.5 of the Born addendum [companion Born addendum] establishes: P(I∗= i) = Tr(Πiρ0) for all states and all i if and only if S = I. Since QCP predicts apparatus-dependent deviations whenever S ̸= I, it follows by modus tollens that Born is not an axiom of QCP but a derived xed-point property of the neutral instrument.

6 Emergent Collapse Dynamics

The stochastic process {ρt} is a supermartingale under the natural ltration generated by measurement events. We construct a Lyapunov functional from the squared Hellinger distance to the nearest pointer-state vertex:

Denition 6.1 (Hellinger-Lyapunov functional). Let pk(ρ) := Tr(Πkρ) be the pointer probabilities and ek the k-th standard basis vector. Dene

V(ρ) := min k∈{1,...,d} D2 H p(ρ), ek
= 2  1 −max k

pk(ρ)  , (18)

p

i(√pi−√qi)2 is the squared Hellinger distance [23]. The functional V is non-negative and equals zero if and only if ρ is a pointer state (p = ek for some k).

where D2 H(p, q) := P

Theorem 6.2 (Consensus collapse). Under the standing assumptions (A1)(A4) [com- panion Supplement, Section 3], the conditioned state process admits a Hellinger-type Lya- punov functional and converges almost surely (i.e. with probability one, in the measure- theoretic sense of Doob's supermartingale convergence theorem [25]) to a pointer projector:

ρt −→Πi∗ almost surely, (19)

where i∗is a random index determined by the trajectory. The exponential contraction bound is E[V(ρt)] ≤e−κmin t V(ρ0), κmin := min i κi > 0. (20)

The contraction rate ¯κ(p) := P

i κi pi(ρ) ≥κmin is a monotone decreasing function of the potential: smaller Φi implies larger spectral gap and faster collapse.

Proof sketch (full proof: [companion Supplement, Theorem 3.1, Lemma A.3]). The QND condition (8) ensures the Hamiltonian part of the generator preserves pointer probabilities. Each quantum jump into channel i sends the post-jump state to σi = ΠiρΠi/ Tr(Πiρ), for which V(σi) = 0 exactly. Therefore the dissipative generator satises GV(ρ) = −¯κ(p) V(ρ) ≤−κminV(ρ). Since V ≥0, Doob's supermartingale convergence theorem gives V(ρt) →0 a.s., hence ρt →Πi∗.

Remark 6.3 (Non-monotonic collapse time). Because Φscan(κ) is U-shaped, γ(κ) is ∩- shaped, and τcoll(κ) ∼1/γmin(κ) has a unique minimum at κopt = a/b. For weak coupling (κ ≪κopt), the process is information-limited (insucient redundancy amplication). For strong coupling (κ ≫κopt), it becomes noise-limited (excessive thermal backaction). The optimal point κopt identies the thermodynamically optimal consensus rate [companion Supplement, Section 9.2].

7 Non-Markovian Robustness

The Markovian pointer-preserving regime serves as the reference front-end model. The companion Supplement formulates the corresponding path-space uniqueness, POVM va- lidity, and no-signalling statements on process-tensor path space [19, 20] under explicit causality, DPI, additivity, support, and KMS/regularity assumptions [companion Supple- ment, Section 2.12]; cf. [21] for the process-tensor framework. Accordingly, robustness beyond Markov/QND is claimed only in that conditional sense. The formal derivation of the process-tensor formulation and the proof of invari- ance of the RSK-linear potential under non-Markovian dynamics are given in [companion Supplement, Section 2.12].

8 Predictions and Falsiability

QCP makes several experimentally accessible predictions.

8.1 Non-monotonic collapse rate

In continuous-measurement platforms (superconducting qubits [29], NV centres, trapped ions), varying the measurement strength κ must yield a U-shaped collapse time τcoll(κ) with a device-specic optimum at κopt = a/b (13).

8.2 Born deviations for biased apparatus

Denition 8.1 (Apparatus bias parameter). Let β 7→S(β) be a C1 family of column- stochastic response matrices with S(0) = I. The scalar parameter β ≥0 quanties the deviation of the apparatus from neutrality; its value is determined by the microscopic apparatus data (T, η, J) through the calibrated map (T, η, J) 7→S of the companion Supplement. The neutral limit corresponds to β = 0, where Born statistics are recovered exactly.

For asymmetrically structured detectors with bias parameter β > 0,

Dtr(PQCP, PBorn) = O(β), (21)

where Dtr denotes the trace distance between the QCP outcome distribution and the Born distribution. Modern POVM tomography [27, 28] can resolve such deviations.

8.3 Microscopic auditability

All parameters of QCP are operationally measurable:

ˆ The microscopic coupling strengths κi can be determined from spectroscopy of the environmental response. ˆ The transport coecients a and b are obtained from GreenKubo integrals (11). ˆ The redundancy rates ˜Ri and noise susceptibilities χi can be extracted through partial monitoring of the environment. ˆ The eective measurement operators Ei are reconstructable via multi-strength quan- tum tomography [26, 27]. Together, these render QCP a fully falsiable physical theory.

9 Comparison with Existing Frameworks

Framework Outcome selection Born rule Distinctive predictions Assumptions

Copenhagen Postulated collapse Axiom None Classical observer cut

Many- Worlds None (all branches persist) Derived (decision- theoretic)

None beyond standard QM Ontology of branches

Bohmian mechanics Deterministic pilot-wave Reproduced Nonlocal hidden variables Additional particle ontology

GRW / CSL Spontaneous localization Modied Heating, localization events New fundamental constants

Assumed Spectrum-broadcast structure Standard decoherence

Redundant environment encoding

Quantum Darwin- ism [24]

Standard open QM dynamics

Apparatus-dep. POVM deformation

QCP Thermodynamic attractor selection Derived (martin- gale collapse)

Table 1: Comparison of quantum measurement frameworks. QCP is the only framework that (i) derives outcome selection from standard open-system dynamics, (ii) derives Born's rule as a neutral xed point, and (iii) predicts apparatus-dependent deviations.

10 Discussion

QCP reconceives measurement as an information-driven nonequilibrium phase-selection process. Collapse is not an exception inserted into the formalism; it is the inevitable attractor of redundancy-amplifying dynamics. The Born rule becomes a statement of apparatus neutrality, not a primitive axiom. The theory oers:

ˆ no new ontology, ˆ no hidden variables, ˆ no nonlinear Schrödinger corrections, ˆ no branching universes.

Instead, outcome selection results from thermodynamic consensus, governed by univer- sal constraints on CPTP maps, DPI, and information geometry. The formalism unies quantum measurement with statistical physics: macroscopic measurement devices act as large-deviation engines, computing the most stable macroscopic alternative.

11 Conclusion

We have presented the Quantum Consensus Principle (QCP), a rst-principles framework that derives quantum measurement as an emergent thermodynamic phenomenon. Start- ing from microscopic Hamiltonian dynamics and extending through information-geometric and path-space large-deviation formalisms, QCP provides a unied and quantitatively predictive description of measurement. In the Markovian pointer-preserving regime emphasized here, and under the standing assumptions carried by the companion materials [companion Supplement; Born adden- dum], QCP yields:

(i) an almost-sure collapse architecture (Theorem 6.2); (ii) an apparatus-dependent POVM description (15); (iii) a non-monotonic calibrated collapse-time prediction (Remark 6.3). In the neutral instrument limit S = I, the Born law is recovered as the exact opera- tional trajectory-frequency law (17) [companion Born addendum]; away from neutrality, calibrated response matrices generate controlled POVM-level deviations that are, in prin- ciple, experimentally testable. By grounding outcome selection in the universal thermodynamics of information pro- duction, QCP transforms the measurement problem from an interpretational postulate into a derived physical process. A separate companion note [companion Necessity note] supplies a narrower neces- sity statement: under bare recorded unitary dynamics, explicit record hypotheses, sup- port faithfulness, and reachable relabelling covariance, no deterministic state-only single- outcome selector exists on the reachable recorded state space. This closes one direct state-only route within the stated formal setting, while the positive uniqueness of the admissible path-space selector remains the independent content of the companion Sup- plement [companion Supplement].

References

[1] J. von Neumann, Mathematische Grundlagen der Quantenmechanik, Springer, Berlin (1932).

[2] N. Bohr, Can quantum-mechanical description of physical reality be considered com- plete? Phys. Rev. 48, 696702 (1935).

[3] W. Heisenberg, Über den anschaulichen Inhalt der quantentheoretischen Kinematik und Mechanik, Z. Phys. 43, 172198 (1927).

[4] H. Everett, Relative state formulation of quantum mechanics, Rev. Mod. Phys. 29, 454462 (1957).

[5] G. C. Ghirardi, A. Rimini, and T. Weber, Unied dynamics for microscopic and macroscopic systems, Phys. Rev. D 34, 470491 (1986).

[6] A. Bassi and G. C. Ghirardi, Dynamical reduction models, Phys. Rep. 379, 257426 (2003).

[7] P. Pearle, Combining stochastic dynamical state-vector reduction with spontaneous localization, Phys. Rev. A 39, 22772289 (1989).

[8] D. Bohm, A suggested interpretation of the quantum theory in terms of `hidden' variables, Phys. Rev. 85, 166193 (1952).

[9] H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford University Press (2002).

[10] H. M. Wiseman and G. J. Milburn, Quantum Measurement and Control, Cambridge University Press (2010).

[11] K. Jacobs, Quantum Measurement Theory and Its Applications, Cambridge Univer- sity Press (2014).

[12] D. Chru±ci«ski and S. Pascazio, A brief history of the GKLS equation, Open Syst. Inf. Dyn. 24, 1740001 (2017).

[13] W. H. Zurek, Decoherence, einselection, and the quantum origins of the classical, Rev. Mod. Phys. 75, 715775 (2003).

[14] E. B. Davies, Markovian master equations, Commun. Math. Phys. 39, 91110 (1974).

[15] D. Petz, Monotone metrics on matrix spaces, Linear Algebra Appl. 244, 8196 (1996).

[16] H. Touchette, The large deviation approach to statistical mechanics, Phys. Rep. 478, 169 (2009).

[17] G. Lindblad, On the generators of quantum dynamical semigroups, Commun. Math. Phys. 48, 119130 (1976).

[18] R. Chetrite and H. Touchette, Nonequilibrium Markov processes conditioned on large deviations, Ann. Henri Poincaré 16, 20052057 (2015).

[19] F. A. Pollock, C. Rodríguez-Rosario, T. Frauenheim, M. Paternostro, and K. Modi, Operational Markov condition for quantum processes, Phys. Rev. Lett. 120, 040405 (2018).

[20] F. A. Pollock, C. Rodríguez-Rosario, T. Frauenheim, M. Paternostro, and K. Modi, Non-Markovian quantum processes: Complete framework and ecient characteriza- tion, Phys. Rev. A 97, 012127 (2018).

[21] G. Chiribella, G. M. D'Ariano, and P. Perinotti, Theoretical framework for quantum networks, Phys. Rev. A 80, 022339 (2009).

[22] R. Kubo, Statistical-mechanical theory of irreversible processes. I, J. Phys. Soc. Jpn. 12, 570586 (1957).

[23] A. Uhlmann, The `transition probability' in the state space of a ∗-algebra, Rep. Math. Phys. 9, 273279 (1976).

[24] W. H. Zurek, Quantum Darwinism, Nat. Phys. 5, 181188 (2009).

[25] J. L. Doob, Stochastic Processes, Wiley, New York (1953).

[26] J. Fiurá²ek, Maximum-likelihood estimation of quantum measurement, Phys. Rev. A 64, 024102 (2001).

[27] G. M. D'Ariano, M. G. A. Paris, and M. F. Sacchi, Quantum tomography, Adv. Imaging Electron Phys. 128, 205308 (2003).

[28] G. M. D'Ariano, L. Maccone, and P. Lo Presti, Quantum calibration of measurement instrumentation, Phys. Rev. Lett. 93, 250407 (2004).

[29] K. W. Murch et al., Observing single quantum trajectories of a superconducting quantum bit, Nature 502, 211214 (2013).