Replica Markov Processes & Channel Entropy

• Replica Markov processes are resource-theoretic frameworks that quantify uncertainty in classical channels using operational games and majorization techniques.
• They extend probability vector majorization to stochastic maps, linking channel simulation, pre-/post-processing, and minimal output Shannon entropy.
• The framework uniquely characterizes channel entropy as a monotone that is additive and asymptotically continuous, unifying classical state and channel resource theories.

Replica Markov processes refer to a resource-theoretic and operational framework for quantifying uncertainty and entropy in classical information-processing scenarios, with a particular focus on describing and comparing noise and uncertainty in classical channels via games of chance and majorization techniques. The formalism generalizes the majorization-based approach to entropy from probability vectors to stochastic maps, capturing the convertibility and preorders induced by channel simulation and pre-/post-processing structure. This approach serves as an operational backbone for understanding channel entropies, establishing the minimal output Shannon entropy as the unique asymptotically continuous, additive channel entropy monotone under channel majorization (Brandsen et al., 2021).

1. Majorization, Conditional Majorization, and Channel Majorization

The foundational principle is that operational tasks—such as guessing games—define preorders on probabilistic objects, with entropy corresponding to monotones of these preorders. In the classical setting, three core levels exist:

1. State majorization: For probability vectors $p$ and $q$, $q\precsim p$ if and only if $q$ can be obtained from $p$ via a doubly stochastic (bistochastic) map. The canonical operational scenario is the w-gambling game, where the player names $w$ faces before a die roll; the optimal probability is the sum of the largest $w$ probabilities in $p$.
2. Conditional majorization: For joint distributions $P$ and $Q$, $Q\precsim_c P$ if $Q$ can be realized from $P$ by conditional relabelings on the side information index, operationalized by correlated dice games with side information and adjustable guess-set sizes.
3. Channel majorization: For classical channels $M$ and $N$, $M\precsim N$ if $M$ can be simulated from $N$ via pre-processing, post-processing (controlled isometries), i.e., a classical superchannel. This order expresses the maximal winning probability in channel-guessing games, establishing a resource theory for classical channels and their entropies (Brandsen et al., 2021).

2. Operational Characterization: Games of Chance

The fundamental methodology behind replica Markov processes is the use of operational games to define preorders and hence resource theories:

• Dice games (no side information, standard majorization): The player's chance to win is the Ky-Fan norm $\sum_{i=1}^w p_i^{\downarrow}$ for the $w$ largest entries of $p$.
• Correlated dice games (conditional majorization): The player has access to side information $x$, chooses a guess set size based on $x$, and the host draws $w$ from a sub-stochastic rule, with the win probability involving maxima over possible guess-set strategies per $x$.
• Channel-input games (channel majorization): The player chooses the channel input $x$ (possibly adaptively) and wins if the output $y$ falls within a top-$w$ set, matched for each scenario by maximal success strategies.

These operational definitions translate majorization and its extensions into physically meaningful games, connecting resource theory, channel simulation, and entropy characterization (Brandsen et al., 2021).

3. Channel Entropy as a Resource Monotone

Within this formalism, entropy is uniquely identified as a monotone for the corresponding preorder. For channels, a channel entropy $H(M)$ satisfies:

• Monotonicity: $M\precsim N \implies H(M) \geq H(N)$.
• Additivity: $H(M \otimes N) = H(M) + H(N)$.

The central theorem is that the minimal output Shannon entropy is the unique, additive, asymptotically continuous classical channel entropy that reduces to Shannon entropy on states:

$H(M) = \min_x H_S(M(\ketbra{x}{x})) = \min_x \left[ -\sum_y p_{y|x} \log p_{y|x} \right]$

This result, originally conjectured and established through connections to dynamical resource theories and asymptotic channel convertibility, means that classical channel entropy is fundamentally determined by minimal output uncertainty under optimal input choice (Brandsen et al., 2021).

4. Resource-Theoretic and Preorder Equivalences

The replica Markov process viewpoint makes explicit the equivalence between operational convertibility, preorders, and resource monotones:

Resource	Preorder	Operational Monotone	Operational Task
States	Majorization	Schur-concave entropy (Shannon, Rényi)	Guessing probability in dice games
Sources	Conditional majorization	Conditional entropy	Correlated dice games with side info
Channels	Channel majorization	Minimal output entropy	Channel-input games/guessing

Majorization relates to the potential for (noisy) conversion between resources, with monotones such as entropy providing the associated measure of resourcefulness or uncertainty.

5. Uniqueness and Continuity

Classical channel entropy is uniquely pinned down (in the asymptotic regime) by being the only monotonic, additive, and asymptotically continuous function under channel majorization that recovers the usual Shannon entropy for deterministic/stochastic states. These properties are closely related to continuity bounds established by Andreas–Winter and follow from deep connections to dynamical divergences (Gour–Tomamichel, Gour–Winter). This establishes the minimal output Shannon entropy as the robust quantifier across different operational regimes and channel preorders (Brandsen et al., 2021).

6. Connections to Resource Theories and Outlook

The replica Markov process framework supports resource theories beyond the classical setting, illuminating:

• State conversion under noisy operations (majorization).
• Dynamical resource theories for processes (channels).
• Unification of entropies: in each setting, entropy arises as the unique, additive, operationally relevant monotone.

Further directions include extending these results to the quantum domain—where entanglement and nonclassical side information become relevant in the construction of quantum channel monotones and entropies—and exploring whether analogous uniqueness theorems hold for quantum channel entropies (Brandsen et al., 2021). The approach provides a powerful bridge between the mathematical structure of Markov processes, operationally motivated game-theoretic definitions, and the axiomatic foundation of entropy in information theory.