{"id":"dd136a00-4f4d-4ef1-8eb4-fd0503b8ae0d","arxiv_id":"1908.08942","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a generic choice of the matrix-valued function L, the purification condition holds and the Lyapunov spectrum of the associated quantum channel is well defined; in a Markov-chain example the top exponent equals -h/2.","lead":"This paper studies quantum channels built by averaging random matrix multiplications, and shows that for almost every choice of a matrix-valued function L (with a fixed sampling measure) the Lyapunov exponents of the associated stochastic process are well defined. It also computes the top Lyapunov exponent for a Markov-chain example and finds it equals minus half the entropy of the chain.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 7.3's 'generic L' need not be stochastic, so the probability P_rho used for the Lyapunov limits is not a probability; the genericity results in B(M_k) never intersect the stochastic set.","rationale":"Reading in good faith, the paper's aims are to prove that purification is generic for fixed mu and to use this, together with earlier Phi-Erg genericity, to obtain Lyapunov exponents for generic L. The purification argument is a plausible Baire-category construction, and the Markov-chain example does point to the claimed -h/2 relation. The load-bearing weak point is the junction between the genericity results and the Lyapunov theorem. Theorem 7.2 requires L to be stochastic, and this is essential: it is what makes P_rho a probability, makes the quantum-trajectory kernel a Markov operator, and makes Proposition 6.5 a martingale statement. Theorem 7.3 omits this hypothesis, and the cited genericity results live in B(M_k), where the stochastic condition is not residual. The scalar example shows the gap concretely: for mu = (delta_0 + delta_1)/2, the stochastic set is a closed nowhere dense subset of B(M_1), so a Baire-generic L fails the normalization condition and P_rho is not a probability. This is an internal hypothesis gap, not a disagreement with any external consensus. It is likely repairable by either proving a relative genericity statement for stochastic L, which would require reworking the density step of Proposition 9.6 since L + epsilon Q^* D Q breaks stochasticity, or by formulating the Lyapunov theorem directly for the i.i.d. product measure mu^N and separating it from the quantum-trajectory probability P_rho. Until one of these repairs is made, Theorem 7.3 is not established as stated. Since the reader's CONDITIONAL verdict already reflects this concern and the paper's ideas are plausibly salvageable, no verdict adjustment is needed.","tokens_in":14723,"tokens_out":13844,"duration_ms":151046,"concrete_test":"Take k = 1 and mu = (delta_0 + delta_1)/2. For the constant function L_t(z) = t, compute the total mass of P_rho on Omega: P_rho(Omega) = |t|^{2n}; if |t| != 1, this is not a probability and not shift-invariant. Then test whether a residual subset of B(M_1) can be chosen stochastic: show S = { L : (|L(0)|^2 + |L(1)|^2)/2 = 1 } is closed with empty interior, e.g. by perturbing any L in S with a small continuous bump supported away from {0,1}. If no residual subset of S is produced, Theorem 7.3 as stated is not supported by Theorem 7.2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 7 begins by assuming L is a Phi-Erg stochastic map, and the probability P(On) = int_{On} tr(W_n rho W_n^dagger) d mu^n is a probability only because Phi*_L(Id) = int L(v)^dagger L(v) d mu(v) = Id. Theorem 7.3 replaces this with 'L is generic for mu' and cites the genericity results of Section 9 and [12], which are Baire statements in B(M_k), the space of all bounded continuous L. No argument is given that the stochastic condition is generic or that a residual subset of B(M_k) can be intersected with the stochastic set. In fact the condition is not generic: for k = 1 and mu = (delta_0 + delta_1)/2, the stochastic set S = { L in B(M_1) : int |L|^2 d mu = 1 } is closed and nowhere dense, so a Baire-generic L is outside S. Without stochasticity, P_rho is not a probability (for scalar constant L_t, P_rho(Omega) = |t|^{2n}, which depends on n unless |t| = 1), the martingale Proposition 6.5 and all 'P_nu-a.s.' statements have no probabilistic meaning, and Theorem 7.2 does not apply. Moreover, the density perturbation in Proposition 9.6, L_epsilon = L + epsilon Q^* D Q, leaves the stochastic set even if L was stochastic, so the purification genericity proof does not automatically adapt to the stochastic subset.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies quantum channels of the form φ_L(ρ)=∫ L(v)ρL(v)† dμ(v) for a fixed measure μ on M_k and a bounded continuous map L. It claims three main results: (i) the purification condition is generic in L for fixed μ; (ii) for generic L, under an integrability condition, the Lyapunov exponents γ_1≥...≥γ_k are well defined; and (iii) in a Markov-chain example, the top Lyapunov exponent equals −h/2, where h is the entropy of the associated Markov measure, and the second exponent is −∞. The paper adapts the framework of [10] to variable L and relies heavily on the authors' earlier work [12] for the Φ-Erg property and entropy definitions.","tokens_in":15078,"tokens_out":10724,"duration_ms":118512,"significance":"If the advertised results were correct, the paper would be a useful extension of the Lyapunov theory of quantum trajectories from the constant-identity case of [10] to a generic set of variable L, and the Markov-chain example would provide a concrete bridge between channel entropy and dynamical Lyapunov exponents. The explicit computation of γ_1=−h/2 is a valuable test case and is essentially correct modulo an indexing issue in the displayed formula for W_n^*W_n. However, the central genericity-to-Lyapunov implication is not established: Theorem 7.3 as stated is false, and the proof of the generic purification property rests on an invalid covering argument. These are load-bearing issues, not presentation problems.","major_comments":[{"comment":"Theorem 7.3 drops the stochasticity assumption that the rest of Section 7 uses to define P as a probability. Earlier in the section, P(O_n)=∫_{O_n} tr(W_n ρ W_n†) dμ^n is a probability only because φ_L^*(Id)=∫ L(v)†L(v) dμ(v)=Id. For a Baire-generic L in B(M_k) this identity fails; for example, for k=1 and μ=(δ_0+δ_1)/2, the stochastic set S={L:∫|L|^2 dμ=1} is closed and nowhere dense, so a generic L is outside S. Consequently the martingale Proposition 6.5, all 'Pν-a.s.' statements, and the invocation of Theorem 7.2 have no valid probability measure behind them, and Theorem 7.3 is false as stated. The authors must either add the stochasticity assumption to the theorem and prove the needed genericity statements inside the stochastic subset, or substantially revise the claim.","section":"Section 7, Theorem 7.3"},{"comment":"The proof of Lemma 9.14 asserts that the balls B(π,ε(π)) centered at a countable dense set K2⊂P2 cover P2. This does not follow from density: in a complete metric space, a union of variable-radius balls around a dense set may fail to cover the whole space, since the radii ε(π) can shrink too quickly. The open set {π: L∈Pur_π} can be a proper open dense set containing K2, as in the standard example of R∖{√2} containing Q. Since Lemma 9.14 is the direct justification of Proposition 9.1, the proof of generic purification is incomplete. A different argument, such as uniform control of ε(π) or a different Baire-category scheme, is needed.","section":"Section 9, Lemma 9.14"},{"comment":"Even if generic purification in B(M_k) were proved, the perturbation used in Proposition 9.6, L_ε=L+εQ^*DQ, does not preserve the stochastic condition ∫L†L dμ=Id. The denseness results of Section 9 are therefore statements about the full space B(M_k), not about the stochastic subset needed for the probability interpretation in Section 7. No argument is given that the residual set of Φ-Erg and purifying L can be intersected with the stochastic set to produce a residual subset there; indeed the stochastic set is typically meager. Thus the advertised conclusion 'generically on L ... Lyapunov exponents are well defined' is not supported by the provided genericity proofs.","section":"Section 9, Proposition 9.6; Section 7"},{"comment":"The proof of Theorem 7.2 is not supplied; the text states only that the proof from [10] 'works here in our setting'. Since this theorem is the probabilistic heart of the Lyapunov-exponent result and the current setting includes non-constant, not necessarily normalized L, the paper should identify which steps of [10] use stochasticity and how they adapt. As written, the dependence on [10] is too coarse to verify the hypotheses, especially in view of the stochasticity issue raised above.","section":"Section 7, Theorem 7.2"}],"minor_comments":[{"comment":"In the displayed formula for W_n(ω)^*W_n(ω), the Kronecker delta should be δ_{j_{k+1} i_k} rather than δ_{i_{k+1} j_k}, and the final rank-one projector should be |j_1⟩⟨j_1| with the ordering W_n=L(ω_n)...L(ω_1). This does not affect the final value γ_1=−h/2, but the indexing should be corrected.","section":"Section 8"},{"comment":"The notation 'Ech(k|∧2 W_n|/tr(W_n^*W_n))' uses Ech without definition; presumably it denotes expectation with respect to the probability P_ch, but this should be stated explicitly.","section":"Section 6, Proposition 6.8"},{"comment":"The phrase 'By abuse of language we consider V_i: Ω→M_k as a random variable V_i(ω)=ω_i' is confusing because V_i was not introduced; the coordinates of ω=(ω_1,ω_2,...) are already matrices, so W_n is simply the product L(ω_n)...L(ω_1). Please align the notation.","section":"Section 5"},{"comment":"The clause 'whenever γ_1=−∞' in item (a) is unclear: if γ_1=−∞, the difference γ_2−γ_1 is not defined, so the statement that γ_2−γ_1<0 needs separate interpretation or a limiting formulation.","section":"Section 7, Theorem 7.3(a)"},{"comment":"There are minor typographical issues, e.g., 'dm(v)' in the abstract should be 'dμ(v)', and 'funtion' in Definition 2.2 should be 'function'.","section":"Abstract and Section 2"}],"recommendation":"reject","confidential_remarks":"The paper's advertised genericity result is not supported: Theorem 7.3 is false as stated because a Baire-generic L is not stochastic, and Lemma 9.14 contains a covering argument that is invalid. The Markov-chain computation is a nice sanity check but does not compensate for the failure of the main generic Lyapunov claim. The manuscript also relies very heavily on the authors' own [12] and on [10] for key definitions and proofs; if the authors resubmit a substantially revised version, the editor should ask for a fully self-contained proof of the genericity statements inside the stochastic subset."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The Markov-chain example that gives γ1 = -h/2 is a concrete, instructive computation, and the question of whether the purification condition is generic is worth asking. But the paper's two main claims — generic purification and generic Lyapunov exponents — are not proven as they stand, and the second is actually false as stated.\n\nWhat the paper does well: the example is the strongest part. Modulo an indexing typo in W_n^*W_n, the calculation is essentially correct: the nonzero singular value is sqrt of the product of transition probabilities, and the ergodic limit gives γ1 = -h/2. That is a nice connection between channel entropy and a dynamical exponent, and it is new relative to the prior work the authors build on. The idea of adapting the Benoist–Fraas–Pautrat–Pellegrini framework to a variable L is also natural and worth pursuing.\n\nThe soft spots are not minor. Lemma 9.14 is supposed to reduce the uncountable intersection over all projections to a countable dense set. The proof says that for each π in a countable dense K2 there is a ball B(π,ε(π)) on which a fixed L satisfies the purification condition for nearby projections, and then claims the union of these balls covers P2. That covering step does not follow from density: the radii ε(π) may shrink too fast, leaving gaps. This is a genuine hole in the proof of the main genericity theorem.\n\nThe stochasticity problem is even more serious. Section 7 starts by assuming L is a Φ-Erg stochastic map, which is exactly what makes P(O_n)=∫ tr(W_n ρ W_n^*) dμ^n a probability. Theorem 7.3 then drops that assumption and just says “L is generic for μ”. But the genericity results of Section 9 are Baire statements in B(M_k), all bounded continuous L, and the stochastic condition is not generic. For k=1 and μ=(δ_0+δ_1)/2, the stochastic set is closed and nowhere dense. So a Baire-generic L will not be stochastic, P is not a probability, the martingale Proposition 6.5 and the “Pν-a.s.” statements lose their meaning, and the ergodic theorem from [10] does not apply. This is not a small fix; the statement of Theorem 7.3 is false as written. Restricting to stochastic L would require reworking the purification genericity on that subset, which the paper does not do.\n\nThe example itself satisfies stochasticity, so its conclusion is safe. But the two headline genericity results need major revision. This is a paper for specialists in the ergodic theory of quantum channels, and it deserves a serious referee — not a desk reject — but the referee should expect to send it back with substantial demands.","headline":"A nice explicit example and a plausible genericity question, but both main theorems have real gaps: Lemma 9.14's covering argument is invalid, and Theorem 7.3 silently drops the stochasticity assumption that makes the Lyapunov exponents well-defined.","tokens_in":15573,"tokens_out":8083,"would_cite":false,"duration_ms":79257,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37H15","37A30","81P45"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that, for a fixed measure, generic bounded continuous channel maps L satisfy the purification condition, so quantum channels of the form φ_L(ρ)=∫L(v)ρL(v)† dμ(v) have well-defined Lyapunov exponents whenever ∫|L|²log|L|…","keywords":["quantum channels","Lyapunov exponents","purification condition","generic properties","channel entropy","Markov chains","random matrix products","Phi-Erg property"],"falsifier":"Take the two-state Markov example with transition matrix $P=\\begin{pmatrix}p_{11}&p_{12}\\\\p_{21}&p_{22}\\end{pmatrix}$ and invariant vector $\\pi$, and compute numerically the finite-time growth rate $\\frac1n\\log\\|W_n(\\omega)^*W_n(\\omega)\\|^{1/2}$ for long $n$. If this does not converge to $-\\frac12(\\pi_1 h_1+\\pi_2 h_2)$, where $h_j=-\\sum_i p_{ij}\\log p_{ij}$, the entropy formula fails; separately, constructing a bounded continuous $L$ that is purification-generic but fails $\\phi_L^*(\\mathrm{Id})=\\mathrm{Id}$ would show that the stochastic hypothesis cannot be dropped from Theorem 7.3.","tokens_in":14527,"feed_emoji":"⚛️","tokens_out":7227,"duration_ms":74841,"temperature":0.7,"pith_summary":"This paper studies quantum channels of the form $\\phi_L(\\rho)=\\int L(v)\\rho L(v)^\\dagger\\,d\\mu(v)$, where $L$ is a measurable matrix-valued function and $\\mu$ a measure on matrices. Its central claim is that for a fixed $\\mu$, most choices of $L$—in the Baire-generic sense over bounded continuous functions with the uniform topology—are good enough to make the associated stochastic processes behave ergodically: irreducibility, the $\\Phi$-Erg property, and the purification condition all hold at once. Under those hypotheses and a mild integrability condition $\\int |L(v)|^2\\log|L(v)|\\,d\\mu(v)<\\infty$, the paper establishes that the Lyapunov exponents $\\infty>\\gamma_1\\ge\\gamma_2\\ge\\dots\\ge\\gamma_k\\ge-\\infty$ of the multiplicative process $W_n=L(v_n)\\cdots L(v_1)$ are well defined. The paper then computes the top exponent for a channel built from a stationary Markov chain and obtains $\\gamma_1=-\\tfrac12 h$, where $h$ is the Markov entropy, with $\\gamma_2=-\\infty$. The point of the paper is to connect an information-theoretic entropy defined through a Ruelle-operator formalism to the genuinely dynamical quantity of a Lyapunov exponent, and to extend Lyapunov theory from the constant case $L=\\mathrm{Id}$ to generic variable channels.","feed_headline":"Generic quantum channels yield well-defined Lyapunov exponents","feed_subtitle":"New proof: purification holds generically, and the top exponent reproduces Markov entropy as γ₁ = −h/2.","key_machinery":"The central object is the random product $W_n(\\omega)=L(\\omega_n)\\cdots L(\\omega_1)$ and its wedge powers $\\bigwedge^p W_n$; the Lyapunov exponents are the almost-sure growth rates of the singular values, obtained through the identity $|\\bigwedge^p W_n|=a_1(W_n)\\cdots a_p(W_n)$. The argument runs on three structural conditions: irreducibility, which gives a Perron-Frobenius theorem for positive maps; the $\\Phi$-Erg property, which guarantees a unique minimal invariant subspace and hence uniqueness of the invariant probability for the Markov kernel; and the purification condition, which forces the tail variable $Y_\\infty=\\lim W_n^*W_n/\\operatorname{tr}(W_n^*W_n)$ to be a rank-one projection almost surely. The purification condition yields exponential decay of the expected second exterior power, $\\int |\\bigwedge^2(L(v_n)\\cdots L(v_1))|\\,d\\mu^{\\otimes n}\\le C\\beta^n$ with $\\beta<1$, and this decay drives both the spectral gap $\\gamma_2-\\gamma_1<0$ and the Baire-category proof that purification is generic.","core_discovery":"The paper's discovery is twofold. First, Proposition 9.1 states that for any measure $\\mu$ supported on more than one point, the set of bounded continuous functions $L$ for which the pair $(L,\\mu)$ satisfies the purification condition is generic in $B(M_k)$: it contains a dense $G_\\delta$ set obtained by intersecting open dense sets over countably many projections. Combined with the earlier genericity of irreducibility and of the $\\Phi$-Erg property, this means that the hypotheses of Theorem 7.2—irreducibility, $\\Phi$-Erg, purification, and $\\int |L(v)|^2\\log|L(v)|\\,d\\mu(v)<\\infty$—are satisfied by generic $L$. Theorem 7.3 then asserts that for generic $L$ the Lyapunov exponents are well defined, that $\\gamma_2-\\gamma_1<0$ (a spectral gap), and that the radial limit $\\lim_{n\\to\\infty}\\frac1n(\\log|W_n(x)|-\\log\\|W_n\\|)=0$ holds almost surely. The Markov-chain example, with $L=I$ and $\\mu=\\sum_{ij}\\delta_{V_{ij}}$, $V_{ij}=\\sqrt{p_{ij}}\\,|i\\rangle\\langle j|$ for a column-stochastic matrix $P=(p_{ij})$, gives the exact formula $\\gamma_1=\\frac12\\sum_{i,j}\\pi_j p_{ij}\\log p_{ij}=-\\frac12 h$, where $h$ is the Shannon entropy of the stationary Markov measure; because $W_n^*W_n$ is always rank one, $\\gamma_2=-\\infty$.","pith_inferences":["The exact relation $\\gamma_1=-\\tfrac12 h$ suggests testing a broader conjecture: for Gibbs channels in the same thermodynamic formalism, the largest Lyapunov exponent may equal $-\\tfrac12$ times the channel entropy, with the factor $1/2$ reflecting the Hilbert-Schmidt norm; finite-dimensional iterated-function-system examples would settle whether this is structural or special to the Markov constru","The paper constructs the generic purification set inside all bounded continuous functions but does not intersect it with the stochastic normalization $\\phi_L^*(\\mathrm{Id})=\\mathrm{Id}$; a natural extension is to check whether purification remains generic in the uniform topology on stochastic channels, since the probability interpretation of the Lyapunov process needs that constraint.","The exponential decay of $\\int|\\bigwedge^2 W_n|$ under purification may control the speed of convergence of quantum trajectories to the unique invariant measure of the Markov kernel, giving quantitative mixing rates for generic channels.","A testable extension is to perturb the Markov example by adding a second measure component while keeping $L$ variable and bounded, and to see whether the identity $\\gamma_1=-h/2$ survives or acquires correction terms from the non-rank-one part."],"forward_implications":["If the paper is right, then for a fixed measure $\\mu$, the three hypotheses of irreducibility, $\\Phi$-Erg, and purification hold simultaneously for a Baire-generic set of bounded continuous maps $L$, so the ergodic temporal-mean theorem applies generically.","Generic channels of this form have a well-defined full Lyapunov spectrum with a spectral gap whenever the top exponent is finite, and in the Markov example the spectrum degenerates with $\\gamma_2=-\\infty$.","The entropy of the Markov-chain channel is a dynamical observable: $h=-2\\gamma_1$, so the Ruelle-operator entropy defined in the previous work can be recovered from the growth rate of random products.","The purification genericity result means the Lyapunov theory is not confined to the constant case $L=\\mathrm{Id}$ but applies to a residual set of variable channel maps.","Because the Markov example's second singular value is exactly zero, the method also predicts that channels built from Dirac measures supported on rank-one Kraus operators produce degenerate spectra beyond the top exponent."],"supporting_citations":[{"why":"Supplies the martingale argument for $Y_n=W_n^*W_n/\\operatorname{tr}(W_n^*W_n)$, the existence of $Y_\\infty$, and the Lyapunov-exponent formalism that the authors adapt to general $L$.","marker":"[10]"},{"why":"The authors' earlier work supplies the entropy definition, the genericity of irreducibility and the $\\Phi$-Erg property, and the Markov-chain example whose entropy matches the Shannon value.","marker":"[12]"},{"why":"Gives the identity $|\\bigwedge^p W_n|=a_1(W_n)\\cdots a_p(W_n)$ connecting wedge-product norms to singular values, which is how the Lyapunov exponents are read off.","marker":"[11]"},{"why":"Provides the Perron-Frobenius theory for positive maps on trace ideals used to equate irreducibility with the spectral-radius and fixed-point properties.","marker":"[25]"},{"why":"Provides the spectral theory of positive maps on C*-algebras used in the equivalent formulations of irreducibility.","marker":"[15]"},{"why":"Supplies the standard formula for the entropy of the stationary Markov measure used in the final computation of $\\gamma_1=-h/2$.","marker":"[24]"}],"fun_headline_variants":["Quantum channels: generically well-defined Lyapunov exponents","Purification is generic, yielding Lyapunov exponents","Top Lyapunov exponent equals −half Markov entropy","Generic channels: Lyapunov exponents and entropy formula","Entropy formula for Lyapunov exponents in quantum channels"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes the channel is stochastic, so that $\\phi_L^*(\\mathrm{Id})=\\mathrm{Id}$ and the expression $P(O_n)=\\int_{O_n}\\operatorname{tr}(W_n\\rho W_n^*)\\,d\\mu^{\\otimes n}$ is a genuine probability, but the generic set of maps $L$ is constructed over all bounded continuous functions without showing it remains generic among the stochastic channels.","fun_headline_variants_meta":{"raw":{"variants":["Quantum channels: generically well-defined Lyapunov exponents","Purification is generic, yielding Lyapunov exponents","Top Lyapunov exponent equals −half Markov entropy","Generic channels: Lyapunov exponents and entropy formula","Entropy formula for Lyapunov exponents in quantum channels"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000366,"raw_usage":{"total_tokens":2130,"prompt_tokens":1268,"completion_tokens":862,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":884,"completion_tokens_details":{"reasoning_tokens":784}},"tokens_in":884,"tokens_out":862,"duration_ms":7811,"temperature":1.0,"reasoning_tokens":784,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:35:27.441754+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the two-state Markov example with transition matrix $P=\\begin{pmatrix}p_{11}&p_{12}\\\\p_{21}&p_{22}\\end{pmatrix}$ and invariant vector $\\pi$, and compute numerically the finite-time growth rate $\\frac1n\\log\\|W_n(\\omega)^*W_n(\\omega)\\|^{1/2}$ for long $n$. If this does not converge to $-\\frac12(\\pi_1 h_1+\\pi_2 h_2)$, where $h_j=-\\sum_i p_{ij}\\log p_{ij}$, the entropy formula fails; separately, constructing a bounded continuous $L$ that is purification-generic but fails $\\phi_L^*(\\mathrm{Id})=\\mathrm{Id}$ would show that the stochastic hypothesis cannot be dropped from Theorem 7.3.","supporting_citations":[{"cited_title":"Benoist, M","cited_arxiv_id":null,"evidence_quote":"Supplies the martingale argument for $Y_n=W_n^*W_n/\\operatorname{tr}(W_n^*W_n)$, the existence of $Y_\\infty$, and the Lyapunov-exponent formalism that the authors adapt to general $L$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The authors' earlier work supplies the entropy definition, the genericity of irreducibility and the $\\Phi$-Erg property, and the Markov-chain example whose entropy matches the Shannon value."},{"cited_title":"Bougerol and J","cited_arxiv_id":null,"evidence_quote":"Gives the identity $|\\bigwedge^p W_n|=a_1(W_n)\\cdots a_p(W_n)$ connecting wedge-product norms to singular values, which is how the Lyapunov exponents are read off."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the spectral theory of positive maps on C*-algebras used in the equivalent formulations of irreducibility."},{"cited_title":"Pollicott and M","cited_arxiv_id":null,"evidence_quote":"Supplies the standard formula for the entropy of the stationary Markov measure used in the final computation of $\\gamma_1=-h/2$."}],"review_version":1}