The Riemann Hypothesis via a Lyapunov Dynamic Cascade in the Explicit Formula

THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE IN THE EXPLICIT FORMULA PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Abstract. We study the horizontal derivative of the completed zeta-function through admissible coefficients R∈S0 (real, even about x=1 2 , nonnegative, with a quadratic zero there). Writing g(x, t) = log |ξ(x+it)|2, we define the horizontal energy ER(t) = ZR R(x)|∂xg(x, t)|2dx, ER,T =ZR ER(t)ϖT(t)dt. Using the Guinand–Weil explicit formula, weighted Plancherel in x , and blockwise estimates for the Gamma, prime, and zero components under the Gaussian window, we obtain for each admissible Rthe bound ER,T ≤C(R)1 + log3(3+T)(T > 0), and uniformly so over compact families of kernels. The zero–block input is a windowed zero–sum lemma depending only on the unit– band zero count N(u; 1) ≪log(2 + |u|) and Poisson–type damping; no spacing hypothesis is required. A local analysis shows that if ρ=β+iγ is a zero with β=1 2 and R(β)>0 , then ER(t)≍ |t−γ|−1 , forcing ER,T = +∞ for every T > 0 . This divergence is incompatible with the global EF bound. Thus no off–line zero can exist and every nontrivial zero of ζ(s) lies on the critical line ℜs=1 2 . The argument is entirely classical and “measure–not–modify”: ζ and ξ are never altered, all regulators act only on external tests, and all T –dependence appears explicitly through a harmless factor 1 + log3(3+T). Date: December 18, 2025. 2020 Mathematics Subject Classification. 11M26, 11M06, 11M45; 35P05, 47A07, 47B25; 68Q17. Key words and phrases. Riemann Hypothesis, explicit formula, resonance kernel, Friedrichs operator, quadratic forms, Lyapunov–dynamic cascade, Gaussian windowing, windowed zero–sum. 1 2 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) 1. Introduction This paper presents a complete, unconditional proof of the Riemann Hypothesis: every nontrivial zero of the Riemann zeta-function ζ ( s )lies on the critical line ℜs = 1 2 . The approach refines an earlier zero–flux framework by removing its sole obstruction, namely a global L2 assumption that implicitly required the very integrability being tested. Here the argument is recast entirely on the test side of the explicit formula: we attach a Lyapunov-type energy to admissible kernels in S0 , derive an explicit-formula bound for this energy for each fixed window scale T > 0(with all T -dependence made explicit), and oppose this to a robust local cusp forced by any off-line zero. A key point, and the most delicate one analytically, is the zero–block control. In particular, we do not assert (and do not need) any pointwise decay in the frequency variable ν that holds uniformly in t : at t = γ the Poisson–type damping in the zero block disappears (the Poisson multipliers are 1), so such a claim is impossible. Instead, Appendix E supplies a windowed mean–square frequency estimate with explicit large– |ν| decay and an integrable low–frequency envelope as ν→ 0, which is exactly what is used in the global explicit–formula bound (see §5.7, Step 5). Theorem A (Riemann, 1859).Every nontrivial zero of ζ ( s )lies on the critical line ℜs=1 2. Theorem A is the classical target statement. Our main structural contribution is a Lyapunov formulation, which recasts Theorem A as an explicit-formula energy statement for the observable g(x, t) := log |ξ(x+it)|2.(1.1) Theorem B (Lyapunov–explicit-formula equivalence).Let S0 denote the class of admissible kernels: real R∈S ( R ), even about x = 1 2 , nonnegative, with R ( 1 2 ) = R′ ( 1 2 ) = 0 and R′′ ( 1 2 ) > 0, and such that a fixed Schwartz square root √R∈S ( R )with ( √R ) 2 = R is chosen once and for all (see Appendix A, Standing class and notation). For R∈S0 put ER(t) := ZR R(x)|∂xg(x, t)|2dx, (1.2) ER,T := ZR ER(t)ϖT(t)dt, (1.3) where ϖTis the mass–one Gaussian window ϖT(t) := (√π T)−1e−t2/T2.(1.4) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 3 For each fixed t∈R the map x7→ g ( x, t )is smooth away from the real parts of the zeros of ξ , so that ∂xg ( x, t )exists for a.e. x and the quantities ER ( t )and ER,T are well-defined (as extended-real values) for all R∈S0 and all T > 0. Since R ( x ) ≥ 0for all x∈R and ϖT ( t ) ≥ 0 for all t∈R, one has ER(t)≥0for all tand ER,T ≥0 (T > 0).(1.5) Consider the following two assertions: (a) (Global EF energy bound) For every compact family K⊂S0 there exists a constant C ( K ) <∞ such that for all R∈K and all T > 0one has ER,T ≤C(K)1 + log3(3+T).(1.6) In particular, for each admissible R∈S0 the windowed energy ER,T is finite for every T > 0, with any dependence on T appearing explicitly through the factor 1 + log3(3+T). (b) (Absence of a persistent off-line cusp) There is no zero ρ = β + iγ of ζ with β = 1 2 and no admissible R∈S0 with R ( β ) > 0for which the local profile ER(t)≍ |t−γ|−1(t→γ)(1.7) yields a nonintegrable, window-persistent contribution ZR ER(t)ϖT(t)dt = +∞for all T > 0.(1.8) Then the Riemann Hypothesis is equivalent to the conjunction of (a) and (b): RH holds ⇐⇒ (a) and (b) both hold. (1.9) The precise analytic formulation and proof of this equivalence are given in Section 4. In particular, the Lyapunov–explicit-formula equivalence expressed by Theorem B shows that proving the global EF energy bound (a) for each fixed T > 0and establishing the incompatibility of any persistent offline cusp (b) suffices to deduce Theorem A. Throughout, the windowed energy ER,T is understood in the Tonelli sense as the extended-real double integral ER,T =ZZR2 R(x)|∂xg(x, t)|2ϖT(t)dx dt, (1.10) with no truncation of neighbourhoods of zero ordinates and no modification of ζ or ξ . Since R≥ 0and ϖT≥ 0, the integrand in the Tonelli integral is nonnegative and ER,T ∈ [0 , + ∞ ]. The explicit-formula 4 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) analysis in Sections 5.7 and 7 bounds this full nonnegative double integral directly, via the linear explicit formula and blockwise estimates, while the local cusp analysis in §5.5–§5.6 shows that any off-line zero with R ( β ) > 0forces the same double integral to be + ∞ for every T > 0. The contradiction arises precisely from this clash, and not from any modification of ζ or ξ nor from excising neighbourhoods of zero ordinates: any auxiliary frequency–side mollifications or low–frequency normalisations are external test–side devices used only to justify interchanges and to expose integrable envelopes, and are absorbed into the constants (see Appendix E and Step 5 in §5.7). The remainder of the introduction outlines the framework and situates the argument within the standard analytic theory of the zeta–function due to Riemann, Hadamard, Selberg [5], Weil [7], Titchmarsh [1], and Iwaniec–Kowalski [10]. All tools are classical: explicit–formula decompositions, Stirling estimates on vertical strips, Schur–type bounds, and basic Lebesgue/Fubini measure theory. No spectral ansatz, no Hilbert–Pólya hypothesis, and no unproved spacing assumptions are employed. 1.1. Horizontal energy and windowed averages. Let g(x, t) := log |ξ(x+it)|2,(1.11) and let R∈S0 be an admissible coefficient: R is real, nonnegative, even about x = 1 2 , with R ( 1 2 ) = R′ ( 1 2 ) = 0 and R′′ ( 1 2 ) > 0. For each fixed t∈R the function x7→ g ( x, t )is smooth away from the finite set of abscissae of zeros of ξ at height t , and locally integrable on R ; thus ∂xg ( x, t )exists for a.e. x and all expressions involving ∂xg below are understood in this a.e. sense. The associated horizontal energy ER(t) := ZR R(x)|∂xg(x, t)|2dx (1.12) captures the local horizontal geometry of ξ . We average this energy in tagainst the mass-one Gaussian window ϖT(t) := (√π T)−1e−t2/T2,ZR ϖT(t)dt = 1,(1.13) and write ER,T := ZR ER(t)ϖT(t)dt. (1.14) By admissibility R ( x ) ≥ 0for all x and by construction ϖT ( t ) ≥ 0for all t,soER(t)≥0for all t(possibly with value +∞) and ER,T ≥0for every T > 0. The window is an external averaging device: it is inserted THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 5 inside the explicit-formula pairings and never appears in any contour integral defining ξor ζ. At the linear level, the explicit formula decomposes ∂xg into three frequency-side blocks: a Gamma block, a Dirichlet–Euler (prime) block, and a zero–sum block. These are treated distributionally, with the t -window applied to the frequency coefficients. In particular, the zero block is controlled by a windowed mean–square frequency estimate (Appendix E), with explicit large– |ν| decay and an integrable low– frequency envelope; no pointwise ν –decay claim uniform in t is asserted or used. 1.2. Global EF control via compact kernel paths. The global side of the argument is an explicit-formula “bank” inequality. For each admissible R∈S0 , Proposition 5.6 (see Section 5.7, together with Appendices D and E) shows that the windowed energy ER,T can be expressed in terms of the Gamma, prime, and zero blocks in the Guinand–Weil explicit formula and that there exists a finite constant C(R)such that ER,T ≤C(R)1 + log3(3+T)(T > 0).(1.15) The constant C ( R )depends only on finitely many Schwartz seminorms of R , and all T -dependence appears explicitly through the factor  1 + log3 (3 + T )  . The proof uses Stirling’s formula on vertical lines for the Gamma block, symbol-type estimates and rapid decay on the prime side, and a windowed zero–sum lemma for the zero block, based only on the unit-band zero count N ( u ; 1) ≪log (2 + |u| )and Poisson-type damping. Crucially, the zero–sum lemma is formulated and proved in a form suitable for the subsequent ν –integration in the weighted Plancherel picture: it yields explicit decay as |ν| → ∞ together with a harmless, integrable envelope as ν→ 0(see Step 5 in §5.7 and Lemma A.12), avoiding any impossible pointwise ν–decay claim. The bound is stable under compact deformations of the kernel. If K⊂S0 is compact in the Schwartz topology, the same arguments yield a constant C(K)<∞such that ER,T ≤C(K)1 + log3(3+T)for all R∈Kand all T > 0. (1.16) In practice, K is taken to be the image of a short, pinned path τ7→ Rτ in S0 , constructed by Gaussian mollification and jet pinning at x = 1 2 . The path is continuous in the Schwartz topology, remains admissible for all τ in a fixed interval [0 , τ∗ ], and preserves positivity at any off-line abscissa β . This “finite-time cascade” is purely on the test side and 6 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) serves only to organise the compactness and continuity arguments for the EF bound. 1.3. Local behaviour near a zero. The local pillar of the argument is a neighbourhood analysis of ξ′/ξ near a zero. If ρ = β + iγ is a zero of multiplicity m≥ 1, the standard factorisation ξ ( s )=( s−ρ ) mh ( s ) with hanalytic and nonzero at ρshows that ∂xg(x, t)=2ℜξ′ ξ(x+it)=2m(x−β) (x−β)2+ (t−γ)2+regular terms. (1.17) If β = 1 2 , the quadratic vanishing of R cancels this singularity and ER ( t ) remains locally integrable. If β=1 2and R(β)>0, then ER(t)≍ |t−γ|−1(t→γ),(1.18) a non-integrable cusp. Since ϖT(γ)>0for every T > 0, this forces ER,T = +∞for all T > 0(1.19) whenever such an off-line zero exists. The cusp profile is stable under small perturbations of R in the Schwartz topology: if Rτ stays positive at β , then the same |t−γ|−1 divergence persists for ERτ ( t )near t = γ . This τ -persistence along compact kernel paths is what ultimately contradicts the global explicitformula bound. 1.4. Methods and scope. The proof is entirely classical and “measure–not–modify”: the zeta-function is never altered, and no artificial zero-free regions are imposed. Instead, we vary only the external test kernels R∈S0 and exploit the explicit formula at the level of distributions. On the functional-analytic side, the quadratic form qR [ h ] = RR|h′|2 is shown to be closed and to generate a nonnegative self-adjoint Friedrichs operator on L2 ( R ); see Kato [21] and Reed–Simon [19, 20]. On the analytic-number-theory side, the Guinand–Weil explicit formula is used in the form of Titchmarsh [1] and Iwaniec–Kowalski [10], with admissible even Schwartz tests. Vertical-line Stirling, unit-band zero counts, and standard properties of ζ′/ζ on fixed strips suffice for all bounds. Appendix G records the measure-theoretic lemmas (Tonelli, Fubini, dominated convergence, truncation at zero ordinates) needed to justify all interchanges, and Appendix E records the windowed mean–square zero–block estimate in the exact form used in §5.7. Independent cross-checks—a smoothed Riemann–von Mangoldt formula and a comparison with Li’s criterion in the spirit of Bombieri- THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 7 Lagarias [11] confirm that our normalisations and signs match the classical literature. These checks are non-evidentiary: they play no role in the logical chain from Theorem B to Theorem A. 1.5. Main result and roadmap. Combining the global explicitformula bound over compact kernel paths with the local cusp forced by any off-line zero yields an immediate contradiction. The precise structural statement is formulated in Theorem B (Section 4), which shows that the Riemann Hypothesis is equivalent to the coexistence of • a unique stationary cut at x = 1 2 in the ERU flux geometry, and • a global Lyapunov-type bound for the explicit-formula energy ER,T , valid for every T > 0with explicit (1 + log3 (3 + T ))- dependence and stable under compact kernel deformations, whose proof relies on windowed mean–square control of the zero block (with integrable low–frequency envelope) rather than any pointwise ν–decay claim. In particular, Theorem B implies Theorem A. The analytic components needed to prove Theorem B are developed in the sections that follow, with Appendices A–G providing the functional-analytic, Fourier, explicit-formula, and measure-theoretic infrastructure. 2. The Clay statement and what must be shown Goal. The Riemann Hypothesis (RH) asserts that every nontrivial zero ρ of the Riemann zeta-function ζ ( s )satisfies ℜ ( ρ ) = 1 2 . The purpose of this section is to specify what we mean by a Clay–compliant proof: a proof carried out entirely in the classical analytic framework of ζ and ξ , using only analytic continuation, the functional equation, the Euler product, and the Guinand–Weil explicit formula under the 2 π –Fourier convention (cf. Titchmarsh [1], Weil [7], Iwaniec–Kowalski [10]). No deformation, smoothing, evolution, or modification of ζ or its zero set is permitted. We also fix the class of admissible operations: distributional pairings against even Schwartz tests, mass–one Gaussian time windows ϖT , admissible spatial kernels R∈S0 (as in Theorem B), and the order of limits T→ ∞ followed by α↓ 0for any x –mollification scale when such limits are required. Admissible transformations may act only on the test side of explicit–formula pairings, never on ζ itself. All such operations are standard in analytic number theory and appear throughout the literature on mean values, smoothing, and Tauberian theory; see also Landau [9] and Jutila [12]. Operations that would constitute a reformulation are stated explicitly below and are disallowed. 8 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) 2.1. The Clay statement. Definition 2.1 (Clay–RH).All nontrivial zeros ρ of ζ ( s )satisfy ℜ ( ρ ) = 1 2 . A Clay–compliant proof establishes this within the classical analytic framework of ζ and ξ , using only analytic continuation, the functional equation, the Euler product, and explicit–formula identities of Guinand–Weil type. In particular, no modified version of ζ is introduced, no external dynamics is imposed on ζ or ξ , and the zero set is never altered. 2.2. Admissible operations and acceptance criteria. Every operation used in this paper is standard, classical, and Clay–compliant. All steps are performed in analytic regimes where they are rigorously valid; all limit processes are explicitly justified; and no global L2 assumption in the time variable is ever made. Admissible Operation (A1): Explicit–formula pairings. Pairings of −ζ′/ζ ( σ + it )with admissible φ∈S ( R )are taken in the sense of distributions in t , as in the Guinand–Weil explicit formula under the 2 π –Fourier normalisation (cf. Weil [7], Iwaniec–Kowalski [10]). When needed, we also use the analogous explicit formula for ξ′/ξ . All such pairings are either absolutely convergent or justified as tempered distribution pairings against φ; no manipulation ever alters ζor ξ. Admissible Operation (A2): Gaussian time windows. For T > 0, the mass–one Gaussian ϖT(t) := (√π T)−1e−t2/T2(2.1) may be inserted inside these pairings in the t –variable. All identities involving ϖT are first established for each fixed T > 0; when a limit T→ ∞ is required, it is justified by Lemma B.5 and the vertical-line envelopes for ζ′/ζ . For T≥ 1one has the uniform bound ϖT ( t ) ≤ 1 /√π for all t∈R , which provides an L1 –dominating function independent of T in this range. No Parseval or Plancherel identity in t is used at any point, and any T –dependence in the bounds appears explicitly through factors such as (1 + logC(3+T)) for fixed integers C≥1. Admissible Operation (A3): Spatial Schwartz weights and mollified kernels. Even Schwartz functions in the real part x are allowed, notably the pinned Gaussian mollifications R(α) of admissible kernels R∈S0 constructed in Appendix F. These appear only inside x –integrals or distributional explicit–formula pairings and are removed via α↓ 0after uniform bounds are proved. No global claim depends on the mollified data; all final statements concern only the original R. Admissible Operation (A4): Finite–time Lyapunov cascade on tests. A key organisational device in the present proof is a finite–time Lyapunov THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 9 cascade ( Rτ ) τ∈[0,τ∗] acting only on admissible kernels. Here Rτ is a short, pinned path in S0 , continuous in the Schwartz topology, obtained by Gaussian mollification and jet pinning at x = 1 2 (Appendix F), and chosen so that Rτ ( β ) > 0persists along the path whenever R0 ( β ) > 0at an off–line abscissa β . No differential equation is imposed on ζ or ξ ; the parameter τ is simply a path parameter in the test space, and we never pass to a limit τ→ ∞ . This finite–time cascade is Clay–compliant because it evolves only the test kernel; the zeta function and its zeros remain entirely unchanged. It serves purely as an analytic bookkeeping device on the test side of the explicit formula, organising compactness and continuity arguments for the EF energy bounds. Admissible Operation (A5): Contour shifts. Vertical shifts of the line ℜ ( s ) = σ with standard indentations at s = 1 and at nontrivial zeros are permitted when justified by classical decay and absolutely convergent series (cf. Titchmarsh [1]). No nonstandard contours or exotic weights are introduced. Admissible Operation (A6): Interchange of limits and differentiation. The interchanges T→ ∞ , α↓ 0, and ∂σ are allowed only when dominated convergence (or a closely related theorem) applies (Lemma B.5 and the envelopes in Appendix B). All such interchanges are justified explicitly and are Clay–compliant. Unless stated otherwise, regulator limits are taken in the order T→ ∞ then α↓ 0, with τ ranging over fixed compact intervals [0, τ∗]. Admissible Operation (A7): Classical equivalents as checks only. Comparisons with the Riemann–von Mangoldt formula or Li’s coefficients (in the sense of Bombieri–Lagarias [11]) are used only as consistency checks. No unproved equivalent of RH is invoked as an assumption. Definition 2.2 (Admissible test functions).A function φ∈S ( R )is admissible if it is even, its Fourier transform bφ is real–valued and rapidly decaying, and bφ≥ 0when needed. A family {φα}α>0 is an approximate identity if φα→δ in S′ ( R )as α↓ 0and if all explicit–formula pairings converge uniformly in compact σ– intervals under this regularisation. Definition 2.3 (Admissible time windows).For T > 0, the Gaussian ϖT is an admissible time window. An identity is admissibly windowed if it is proved for all T > 0and the limit T→ ∞ (when required) exists and is justified by Lemma B.5. Unless stated otherwise, limits are taken in the order T→ ∞ then α↓0. 16 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Fourier transform of a Gaussian under the 2 π –convention. No global positivity of c Rαis used later. Heat–semigroup parametrisation. Let gα ( X ) = e−αX2 . Under the 2π–Fourier convention, bgα(ξ) = pπ/α e−π2ξ2/α, et∂2 xf=Gt∗f, (3.19) with c Gt(ξ)=e−4π2tξ2. Setting t=1 4αgives gα∗f=pπ/α et∂2 xf, (3.20) and the identity Rα=−∂αgα(3.21) holds pointwise. This representation is purely analytic, obtained by comparing Fourier multipliers; no evolution of ζ is introduced at any stage. Mellin transform in the local coordinate. For ℜs>−2, Z∞ 0 Xs−1X2e−αX2dX =1 2α−(s+2)/2Γ s+2 2,(3.22) obtained by the substitution u = αX2 . This identity is useful for normalisation checks in explicit–formula computations and for tracking constants. Order of limits and Clay–compliance. Limits in the regulators are always taken in the order T→ ∞, α ↓0,(3.23) justified by dominated convergence and vertical-line bounds for ζ and ζ′ (Lemma B.5; see also Appendix B and Titchmarsh [1], Ivić [2]). In particular, for each fixed α > 0the windowed explicit–formula pairings are absolutely convergent and their T→ ∞ limits are controlled by Lemma B.5; after this, the limit α↓ 0is taken using the uniform S –bounds on Rα . This preserves the zero set of ζ and ensures full Clay–compliance in the sense of Definition 2.1. Only after these limits are taken do any global conclusions about the zeros of ζ enter the argument. 3.4. Full definition of HRα as an α –explicit closed form. Fix α > 0and consider the canonical admissible kernel Rα(x)=(x−1 2)2e−α(x−1 2)2, x ∈R,(3.24) an even, nonnegative Schwartz function with the required quadratic vanishing at x = 1 2 . Throughout we work on L2 ( R ), writing f′ for the distributional derivative in x. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 17 Closed form and associated operator. Define the nonnegative sesquilinear form qRα[f, g] := ZR Rα(x)f′(x)g′(x)dx, (3.25) with form domain DqRα:= nf∈L2(R) : pRαf′∈L2(R)o.(3.26) Since Rα∈L∞ ( R ) ∩S ( R )is nonnegative and C∞ c ( R ) ⊂ DqRα , the form qRα is densely defined on L2 ( R ). The map f7→ √Rαf′ is closed from L2 ( R )(with domain H1 ( R )) into L2 ( R ), and hence qRα is a closed, nonnegative quadratic form (see Kato [21], Reed–Simon [19]). Writing qRα [ f ] := qRα [ f, f ], the first representation theorem (see Kato [21], Reed–Simon [19,20]) provides a unique self–adjoint operator HRα≥ 0 such that ⟨HRαf, g⟩=qRα[f, g] (f∈ D(HRα), g ∈ D(qRα)).(3.27) On the core C∞ c(R), HRαf=−Rαf′′(in the distributional sense),(3.28) with no boundary condition imposed at x = 1 2 despite the quadratic degeneracy Rα ( 1 2 ) = R′ α ( 1 2 ) = 0. The vanishing is entirely absorbed into the coefficient of this divergence–form operator. Optional form sums (recorded for robustness only). If w ( x ) = e−βx2 with β > 0, define s[f, g] := ZR w(x)f′(x)g′(x)dx, (3.29) a nonnegative closed form on {f∈L2 ( R ) : √w f′∈L2 ( R ) } . For λ∈R set qRα,λ[f, g] := qRα[f, g]+λs[f, g].(3.30) If λ≥ 0, then qRα,λ is clearly closed and nonnegative, with associated operator HRα,λ ≥ 0. If λ < 0but sufficiently small, s is qRα –form bounded with relative bound 0(KLMN theorem; see Kato [21]), so qRα,λ remains closed and semibounded and again defines a self–adjoint operator HRα,λ . This perturbative robustness is never used in the RH argument; we record it only to indicate that the framework survives small form perturbations of the kernel. 18 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Strong divergence form. If Bα := Rα + λw is C1 (for example when λ≥0), then on H2 loc(R)the operator acts as HRα,λf=−d dxBα(x)f′(x),(3.31) with the L2 –realisation determined uniquely by the closed form qRα,λ . Green’s identity is the usual one for divergence–form operators and coincides with Proposition 5.2, equation (5.46) , upon replacing R by Bα. Basic bounds and monotonicity. For every f∈ DqRα,λ, 0≤qRα,λ[f] = ZRRα(x)+λw(x)|f′(x)|2dx ≤∥Rα∥L∞+|λ|∥w∥L∞∥f′∥2 L2. (3.32) For λ1≤λ2we have qRα,λ1[f]≤qRα,λ2[f]for all f∈ D(qRα,λ2).(3.33) For the canonical family Rα ( x ) = ( x−1 2 ) 2e−α(x−1 2)2 and any fixed f∈ D(qR0,0)we have Rα(x)↑(x−1 2)2(α↓0, x ∈R),(3.34) so qRα,0 [ f ] ↑qR0,0 [ f ]as α↓ 0whenever f∈ D ( qR0,0 ), where R0 ( x ) := (x−1 2)2. Removal of the spatial regulator. As α↓0, Rα(x)↑(x−1 2)2for every x∈R.(3.35) Thus the family of forms {qRα,λ}α>0 increases pointwise to qR0,λ on D ( qR0,λ ). By Kato’s monotone convergence theorem for quadratic forms (see Reed–Simon [20, Thm. VIII.3.11]), (HRα,λ +I)−1s.r. −−−→ α↓0(HR0,λ +I)−1,(3.36) where “s.r.” denotes strong resolvent convergence. In the RH proof we always take λ = 0 and remove regulators in the Clay–compliant order: first T→ ∞ (time window), then α↓ 0(spatial smoothing), as in Section 2.2 and Lemma B.5. Clay–compliance and intended use. All appearances of Rα arise solely inside quadratic forms applied to the observable g ( x, t ) = log |ξ ( x + it ) |2 or in explicit–formula pairings with admissible tests. They never modify ζ , impose any external PDE or dynamics on ζ , or alter the zero set. After sending T→ ∞ and α↓ 0, only the properties of the classical zeta function remain. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 19 In particular, whenever g ( ·, t )lies in D ( qRα )(for example after localisation in x and truncation in t as in Appendix G), we have the energy identity qRα[g(·, t)] = ZR Rα(x)|∂xg(x, t)|2dx. (3.37) We use this merely as an analytic bookkeeping device for the horizontal energy. The finite–time Lyapunov cascade later acts only on the test kernel R ; the operator HRα plays no dynamical role and serves solely to express the horizontal energy in a closed–form, Clay–compliant manner. 3.5. Positivity and Plancherel for HRα (and RH α –invariance). Throughout we use the 2π–Fourier convention b f(ξ) = ZR f(x)e−2πixξ dx, (3.38) b f′(ξ) = (2πiξ)b f(ξ). We first record the canonical model kernel Rα(x) = (x−1 2)2e−α(x−1 2)2, α > 0,(3.39) and the auxiliary Gaussian weight w(x) = e−βx2, β > 0.(3.40) Both are real, even, nonnegative, rapidly decaying, and appear exclusively inside L2 –pairings in the x –variable. No step alters ζ , replaces it by a smoothed version, or introduces any evolution of ζ . The canonical choice Rα is used for explicit computations; all explicit– formula and Lyapunov estimates in the sequel are formulated for general admissible kernels R∈S0and their α–mollifications. Define the quadratic forms qRα[f] = ZR Rα(x)|f′(x)|2dx, (3.41) qRα,λ[f] = qRα[f]+λZR w(x)|f′(x)|2dx, λ ≥0. The natural domains are D(qRα) = f∈L2(R) : pRαf′∈L2(R),(3.42) D(qRα,λ) = f∈L2(R) : pRα+λw f′∈L2(R), λ ≥0, where f′ is understood in the sense of distributions. Since Rα vanishes quadratically at x = 1 2 , we have only the inclusion H1 ( R ) ⊂ D ( qRα ) (not equality). Let HRα denote the unique nonnegative self–adjoint 20 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) operator associated with qRα via the Friedrichs construction, as in §3.4; similarly HRα,λ corresponds to qRα,λ for λ≥0. (1) Positivity, semiboundedness, and KLMN stability. For all f∈ D(qRα), qRα[f] = ∥pRαf′∥2 2≥0,(3.43) and for all λ≥0and f∈ D(qRα,λ), qRα,λ[f] = ZR (Rα+λw)(x)|f′(x)|2dx ≥0.(3.44) Thus each qRα,λ is densely defined, closed, and semibounded from below, and for every λ≥ 0there exists a unique nonnegative self–adjoint operator HRα,λ with qRα,λ as its Friedrichs form (Kato [21, Ch. VI], Reed–Simon [20, Ch. VIII]). In the RH argument we use only λ = 0; the perturbative family λ > 0is recorded solely to indicate robustness under small positive form–sums. No later step uses any additional property of HRα,λ beyond nonnegativity and self–adjointness. (2) Plancherel representation in x . For f∈S ( R ), inserting the Fourier representation of f′ into the x –integral and applying Fubini (justified by the Schwartz decay of Rαand f) yields qRα[f] = ZR Rα(x)|f′(x)|2dx =ZZR2c Rα(ξ−η) (2πη)(2πξ)b f(η)b f(ξ)dη dξ. (3.45) Similarly, ZR w(x)|f′(x)|2dx =ZZR2bw(ξ−η) (2πη)(2πξ)b f(η)b f(ξ)dη dξ. (3.46) Hence qRα,λ[f] = ZZR2c Rα+λbw(ξ−η) (2πη)(2πξ)b f(η)b f(ξ)dη dξ (3.47) for all f∈S ( R ), with extension to f∈ D ( qRα,λ )by density. This is a purely x –side representation of qRα,λ . We never apply Plancherel in the t –variable, nor do we assume any global L2 condition in t . All windowed explicit–formula estimates insert the Gaussian in t at the linear ξ′/ξ level and then use Cauchy–Schwarz in (σ, t)(cf. Proposition 5.6). (3) Uniform continuity bounds in H1 . Young’s convolution inequality together with Plancherel in the x–variable gives, for all f∈S(R), 0≤qRα[f]≤ ∥c Rα∥L1(R)∥f′∥2 2,(3.48) and qRα,λ[f]≤∥c Rα∥1+λ∥bw∥1∥f′∥2 2.(3.49) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 21 No uniform control of ∥c Rα∥1 as α↓ 0is required anywhere in the proof. The limiting operator is accessed via monotone convergence of forms (see below), not via uniform L1bounds. (4) Canonical kernel in frequency. For the model Rα one computes explicitly c Rα(ξ)=e−πiξ √π 2α3/21−2π2ξ2 αe−π2ξ2/α.(3.50) Global positivity of c Rα is not assumed: the factor in parentheses shows that c Rα changes sign for large |ξ| . In the global EF bounds we use only the pointwise positivity of qRα and the upper bounds above; no spectral positivity of c Rαis required. (5) Monotone convergence and strong–resolvent limits. As α↓ 0we have pointwise monotone convergence Rα(x)↑(x−1 2)2=: R0(x)for each x∈R.(3.51) Consequently, for every f∈ D(qR0), qRα[f] = ZR Rα(x)|f′(x)|2dx ↑ZR R0(x)|f′(x)|2dx =qR0[f].(3.52) By Kato’s monotone convergence theorem for closed forms (Reed– Simon [20, Thm. VIII.3.11]), we obtain strong resolvent convergence (HRα,λ +I)−1s.r. −−−→ α↓0(HR0,λ +I)−1, λ ≥0,(3.53) where “s.r.” denotes convergence in the strong resolvent sense. In particular, the spatial regulator α is removed only after the T –dependent bounds for the windowed energies have been established (Proposition 5.6), in accordance with Clay–compliance and the order of limits specified in §2.2. (6) Clay–compliance. All appearances of Rα or w occur solely inside L2 –pairings in x or in explicit–formula coefficients. At no point is ζ replaced by a smoothed, truncated, or otherwise modified version, and no PDE or flow is imposed on ζ or ξ . All limits are taken in the prescribed order T→ ∞ first, α ↓0second,(3.54) with interchanges justified by Lemma B.5 and dominated convergence on vertical lines. Proposition 2.1 then implies that all identities obtained after removing regulators concern the classical ζ and its unmodified zero set. 22 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) (7) Consequence for RH: α–invariance of the cusp behaviour. For Eα(t) = ZR Rα(x)|∂xg(x, t)|2dx, g(x, t) = log |ξ(x+it)|2,(3.55) the neighbourhood–divergence phenomenon near an off–line zero is independent of α . The following lemma makes this uniformity precise. Its proof is purely local in ( σ, t )and uses only the factorisation of ξ near a zero; in particular, it does not use any explicit–formula bounds or T–dependent estimates. Lemma 3.2 (Uniform α –invariance of the RH flux criterion).Let ρ=β+iγ be a zero of ζof multiplicity m≥1. (i) If β = 1 2 , then there exist α0> 0, η > 0and constants c1, c2, C0> 0(independent of α∈(0, α0]) such that c1 m2 |t−γ|−C0≤Eα(t)≤c2 m2 |t−γ|+C0,0<|t−γ| ≤ η, (3.56) and in particular Eα(γ)=+∞for every α∈(0, α0]. (ii) If β=1 2, then Eα(t)<∞for all t∈Rand all α > 0. Proof. We work locally near ρ. Write ξ(s)=(s−ρ)mh(s),(3.57) where his analytic and nonvanishing in a neighbourhood of ρ. Then ξ′ ξ(s) = m s−ρ+h′ h(s),(3.58) and hence, for g(x, t) = log |ξ(x+it)|2, ∂xg(σ, t) = 2 ℜξ′ ξ(σ+it)= 2 ℜm σ−β+i(t−γ)+2 ℜh′ h(σ+it), (3.59) where σ∈R. Case (i): β=1 2.Set δ:= |β−1 2|>0and choose Iβ:= [β−δ 2, β +δ 2].(3.60) Then |σ−1 2|≥δ/ 2for all σ∈Iβ , and therefore for all α∈ (0 , 1] and σ∈Iβ, cδ≤Rα(σ) = (σ−1 2)2e−α(σ−1 2)2≤Cδ(3.61) for suitable constants 0 < cδ≤Cδ<∞ depending only on δ (not on α ). In particular, the restriction of Rα to Iβ is bounded above and below by positive constants uniformly in α∈(0,1]. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 23 Restricting the last display for ∂xg to σ∈Iβ and writing ∆ := t−γ , we have ∂xg(σ, t) = 2mσ−β (σ−β)2+ ∆2+O(1),(3.62) where the O (1) term is uniform in ( σ, t )as long as |σ−β| ≤ δ/ 2 and |t−γ| ≤ 1, since h′/h is analytic and bounded on a compact neighbourhood of ρ and the gamma factors in the definition of ξ are holomorphic and bounded on vertical lines in any fixed strip |σ−β| ≤ δ/2. Hence, for 0 <| ∆ | ≤ η (with η > 0chosen small enough) and σ∈Iβ , ∂xg(σ, t)2=4m2(σ−β)2 (σ−β)2+ ∆22+O1 (σ−β)2+ ∆2+O(1),(3.63) with constants independent of α . Integrating over Iβ against Rα ( σ ) dσ and using the uniform bounds on Rαon Iβgives ZIβ Rα(σ)(σ−β)2 ((σ−β)2+ ∆2)2dσ ≍δZδ/2 −δ/2 u2 (u2+ ∆2)2du ≍1 |∆|,(3.64) and ZIβ Rα(σ)dσ (σ−β)2+ ∆2≪δlog1 + δ |∆|≪δlog 1 |∆|,(3.65) for 0 <| ∆ |≤η , with implied constants independent of α . The contributions to Eα ( t )from R\Iβ are uniformly bounded in t and α (since ξ′/ξ is holomorphic and bounded on compacta and Rα is Schwartz), and the logarithmic term is lower order compared with | ∆ |−1 as | ∆ | → 0. Collecting these estimates yields constants c1, c2, C0> 0, independent of α∈ (0 , α0 ](for some α0≤ 1) and 0 <|t−γ| ≤ η , such that c1 m2 |t−γ|−C0≤Eα(t)≤c2 m2 |t−γ|+C0.(3.66) The inequality Eα ( γ )=+ ∞ follows since the integral of |t−γ|−1 in any neighbourhood of t=γdiverges. Case (ii): β=1 2.In this case the factorisation of ξnear ρgives ∂xg(σ, t) = 2 ℜm σ−1 2+i(t−γ)+O(1),(3.67) where the error term is again bounded on compact sets in ( σ, t ). Thus, near σ=1 2and fixed t, ∂xg(σ, t)2≪1 (σ−1 2)2+ (t−γ)2+ 1,(3.68) 24 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) and the integrand in Eα(t)satisfies Rα(σ)|∂xg(σ, t)|2≪(σ−1 2)21 (σ−1 2)2+ (t−γ)2+ 1.(3.69) The first term is locally integrable in σ near σ = 1 2 , and the second term is integrable because Rα is Schwartz. Therefore Eα ( t ) <∞ for all t∈Rand all α > 0, proving (ii). □ Thus the cusp detection mechanism for off–line zeros is independent of the spatial smoothing scale α : whenever a zero ρ = β + iγ with β = 1 2 exists, the associated horizontal energy Eα ( t )has a locally nonintegrable |t−γ|−1 –type singularity for every α∈ (0 , α0 ], while zeros on the critical line produce no such divergence. Combined with the monotone–convergence limit HRα→HR0 and the Clay–compliant removal of regulators, all global conclusions in the sequel are ultimately taken for the unregularised kernel R0 ( x )=( x−1 2 ) 2 and the classical zeta function. 3.6. Cumulative energy/flux embedding (ERU) and flux functional. Quantifier banner. Fix R∈S0 (real, even about x = 1 2 , R ( 1 2 ) = R′ ( 1 2 ) = 0, R′′ ( 1 2 ) > 0, R≥ 0and strictly positive off 1 2 ), and for α > 0let R(α) = ϕα∗R be an admissible spatial mollification in the sense of Definition 2.2. Fix also the mass–one Gaussian time window ϖT(t) = (√πT)−1e−t2/T2, T > 0.(3.70) All implicit constants in this subsection depend only on finitely many S –seminorms of R (uniformly over compact K⊂S0 ) and are independent of Tand α. We work with the observables f(x, t) := log |ζ(x+it)|2, g(x, t) := log |ξ(x+it)|2,(3.71) where ξ(s) = 1 2s(s−1)π−s/2Γs 2ζ(s).(3.72) The nontrivial zeros of ξ and ζ coincide, and the functional equation gives the symmetry g(x, t) = g(1−x, t)⇒∂xg1 2, t= 0 for all t∈R(cf. Titchmarsh [1], Ivić [2]). (3.73) We prefer g for symmetry; all statements below remain valid with f after adding/removing the smooth gamma and polynomial calibrants. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 25 Cumulative energy and local flux. Fix α > 0. For each t∈R define the cumulative x–energy ΦR(α)(x, t) := Zx −∞ R(α)(y)|∂yg(y, t)|2dy, (3.74) whenever the integral converges on ( −∞, x ]. Since R(α)∈L1∩L∞ and ∂xg ( ·, t ) ∈L2 loc ( R )for each fixed t (cf. Lemma B.5 and Appendix G), the density y7→ R(α) ( y ) |∂yg ( y, t ) |2 belongs to L1 loc ( R ), and hence the map x7→ Φ R(α) ( x, t )is finite, absolutely continuous, and nondecreasing on any compact interval disjoint from the real part of a zero of ζ . The precise divergence near zeros is described, for the canonical kernel, in Sections 5.5 and 5.6; the same local argument applies to any R∈S0 with R(β)>0. For x0∈R and ε > 0define the (mollified) local flux across the vertical cut {x=x0}by F(α) ε(x0, t) := 1 2εZx0+ε x0−ε R(α)(x)|∂xg(x, t)|2dx, (3.75) and, when the limit exists, the pointwise flux F(α)(x0, t) := lim ε↓0F(α) ε(x0, t).(3.76) By the fundamental theorem of calculus (in x ) applied to the locally integrable density R(α) ( x ) |∂xg ( x, t ) |2 and by the Lebesgue differentiation theorem, ∂xΦR(α)(x, t) = R(α)(x)|∂xg(x, t)|2for a.e. x∈R,(3.77) so F(α) ( x0, t ), when it exists, represents the x –derivative of Φ R(α) ( ·, t ) at x0. Lemma 3.3 (Zero flux ⇐⇒ stationary cut).Fix α > 0and t∈R . For any x0∈Rthe following are equivalent: (1) F(α)(x0, t) = 0; (2) ∂x Φ R(α) ( x, t ) = 0 in a neighbourhood of x0 , in the distributional sense; (3) R(α)(x)|∂xg(x, t)|2= 0 a.e. in a neighbourhood of x0. In particular, if R(α) ( x0 ) > 0, then F(α) ( x0, t ) = 0 if and only if ∂xg(x0, t) = 0. Proof. The equivalence of (ii) and (iii) follows directly from (3.77) : if ∂x Φ R(α) = 0 in the sense of distributions on an open interval J∋x0 , then R(α)(x)|∂xg(x, t)|2= 0 a.e. on J, and conversely. 32 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) and Appendix E). In particular, under RH we have FR,T <∞ for all admissible Rand all T > 0. (b) (iii) ⇒ (i). Assume (iii) and suppose, for contradiction, that there is an off–line zero ρ = β + iγ . Choose R∈S0 with R ( β ) > 0, and let {Rτ}0≤τ≤τ∗ be a kernel path with R0 = R and Rτ ( β ) ≥c0 ( β ) > 0for all τ . By (ii) the cusp at t = γ persists along the path and forces FT ( τ ) = + ∞ for every τ∈ [0 , τ∗ ] and every T > 0, contradicting the EF–bank bound (5.174) with K = {Rτ : 0 ≤τ≤τ∗} . Hence no such ρ can exist and all nontrivial zeros lie on ℜs=1 2. (c) (ii) characterises the alternative regime in which RH fails: every off–line zero produces a |t−γ|−1 cusp and a uniform, τ –persistent obstruction to the EF–bank bound (5.174) for any kernel path that stays positive at its abscissa β . In particular, (ii) and (iii) are mutually exclusive. Consequently, (i) and (iii) are equivalent, and failure of RH is equivalent to the cusp obstruction in (ii). In particular, the EF–bank & zero–flux thesis encoded in Theorem 1 is equivalent to the Lyapunov–explicit– formula formulation in main Theorem B, and hence to the Clay–RH statement in main Theorem A. Remark 4.1 (Interpretation and zero–flux geometry).From a geometric viewpoint, FR,T is a Lyapunov–type energy defined entirely on the test side of the Guinand–Weil explicit formula. The kernel path {Rτ} plays the role of a finite–time cascade inside S0 : it explores a compact region of admissible tests without ever altering ζor its zeros. Under RH, the ERU embedding of §3.6 shows that the time–averaged flux F(α) T ( x0 )has a unique stationary cut at x0 = 1 2 (Proposition 3.1), and the EF–bank estimate (5.174) provides a uniform–in– R Lyapunov bound on compact kernel families (with explicit (1 + log3 (3 + T ))– dependence in T ) along every pinned path. If an off–line zero exists, the local analysis of ζ′/ζ produces a |t−γ|−1 cusp in the horizontal derivative, stable under all admissible kernels and all smoothing scales (Lemma 3.2); this cusp propagates into the zero block ZR and forces FR,T = + ∞ for every kernel path that remains positive at β , contradicting the EF–bank regime (5.174). The proof of Theorem 1, and hence of main Theorems A and B, uses only classical tools: the explicit formula in the sense of Weil and Iwaniec–Kowalski [7, 10], Stirling’s approximation on vertical strips (Titchmarsh [1], Ivić [2]), Schur’s inequality and the unit–band zero count N ( u ; 1) ≪log (2 + |u| )(Montgomery–Vaughan [4]), and standard Hilbert–space form theory for HR (Kato [21], Reed–Simon [19,20]). No THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 33 spectral ansatz, no Hilbert–Pólya hypothesis, and no unproved spacing assumptions are used; the Lyapunov apparatus acts only on admissible test kernels and remains fully Clay–compliant throughout. 5. Technical sections (details) This section assembles the rigorous analytic backbone of the proof. It provides all domain definitions, operator–theoretic facts, local asymptotics, explicit–formula decompositions, limit justifications, and frequency– side bounds needed for the Lyapunov–explicit–formula / cusp equivalence in Main Theorem B and Theorem 1. All arguments here are derived from the classical completed zeta ξ, the observable g(x, t) = log |ξ(x+it)|2,(5.1) and admissible test functions only. For each fixed t∈R , the map x7→ g ( x, t )is smooth away from the finitely many real parts of zeros at height t and locally integrable on R , so that ∂xg ( ·, t )is well-defined almost everywhere. No modification of ζ or its zero set is ever made, and all regulators are introduced only inside L2–pairings and removed in admissible limits justified in Section 2. Every statement below is formulated in an analytic regime where it is rigorously valid and all integrals are either absolutely convergent or understood as Tonelli integrals with nonnegative integrand, so that no hidden assumption enters the finite–time Lyapunov/cusp contradiction. Order of presentation and logical dependency. (1) Notation and admissible objects (§5.1): fixes Fourier conventions, the completed zeta observable, the admissible kernel classes, and the admissible time windows. (2) Positivity, Plancherel, and frequency bookkeeping (§3.5): records L2 –positivity and frequency–side control for the quadratic forms qRα , and develops the bookkeeping needed for the explicit– formula Lyapunov energy. Weighted Plancherel in x is used only pointwise in t where ER ( t ) <∞ , as made explicit in Remark A.2. (3) Spectral uniqueness of the canonical kernel (§5.2): derives the Gaussian–quadratic profile Rα from a Sturm–Liouville variational problem and records its extremal properties. This is used only for explicit estimates; Clay–level conclusions are uniform in R∈S0. 34 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) (4) Self–adjointness and Green identity (§5.3): shows that HR is the Friedrichs extension of − ( Rh′ ) ′ and records the weighted Green identity needed for later integration–by–parts arguments. (5) Analytic–continuation filter and local zero asymptotics (§5.4): provides the local model for ∂xg near any nontrivial zero ρ , with uniform constants on a bidisc, and notes harmonicity of g off the zero set. (6) Neighbourhood–divergence lemma (§5.5): establishes the universal |t−γ|−1 blow–up rate for R –weighted horizontal energy in a neighbourhood of an off–line zero. In its uniform form over compact kernel families K⊂S0 with infR∈KR ( β ) > 0, this yields the local cusp pillar used in the Lyapunov/cusp contradiction. (7) Explicit–formula bounds for the Lyapunov energy (§5.7): decomposes the explicit formula into Gamma, Dirichlet–Euler, and zero blocks (for each fixed admissible kernel), treats each block linearly before squaring, and applies Stirling asymptotics together with Schur’s lemma and the unit–band zero count N ( u ; 1) ≪log (2 + |u| ). The zero block is controlled via the windowed zero–sum lemma of Appendix E, which uses only unit–band zero counts and Poisson damping and produces a (1 + |ν| ) −p decay with logarithmic loss (1 + log3 (3 + T )) for some p > 1. This yields explicit–formula bounds for the Lyapunov functional FR,T attached to R and T , and, after promotion from a single kernel to compact families K⊂S0 , gives the EF–bank pillar FR,T ≤C(K)1 + log3(3+T)for all R∈K, T > 0,(5.2) with constants depending only on finitely many S –seminorms of R. (8) Lyapunov functional, pinned path, and contradiction (§5.9): constructs, for any admissible R0∈S0 with R0 ( β ) > 0at an off–line abscissa β , a short pinned path τ∈ [0 , τ∗ ] 7→ Rτ∈S0 that is continuous in the Schwartz topology and satisfies Rτ ( β ) ≥c0> 0 for all τ . The image K = {Rτ : 0 ≤τ≤τ∗} is compact, so the local cusp pillar from §5.5 forces RERτ ( t ) ϖT ( t ) dt = + ∞ for all τ, T, while the EF–bank pillar from §5.7 yields FRτ,T ≤C(K)1 + log3(3+T)(∀τ∈[0, τ∗], T > 0).(5.3) This finite–time compact–path contradiction rules out off–line zeros and completes the Lyapunov/cusp equivalence in Main Theorem B and Theorem 1. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 35 (9) Measure–theoretic audit (§5.10): handles the discrete set of zero ordinates, clarifies t –null exceptional sets, and fixes the interpretation of pointwise versus a.e. statements under the Gaussian window. (10) Numerical sanity checks (§5.11): optional, non-evidentiary demonstrations of the model asymptotics and the stability of the T – averaged energies. Throughout, every identity and bound is derived from g(x, t) = log |ξ(x+it)|2,(5.4) admissible Schwartz weights in x , and Gaussian windows in t . No deformation or modification of ζ or its Euler–Gamma factors is ever made; regulators act solely on the test side and are removed in admissible limits before any RH conclusion. This “measure–not–modify” principle guarantees full Clay compliance in the sense of Section 2. Dependence on Tand kernel families. All analytic bounds established below hold for every T > 0, with any T –dependence made explicit through factors such as  1 + log3 (3 + T )  . The implicit constants depend only on finitely many S –seminorms of a given admissible kernel R and are independent of T . When we later work with a compact subset K⊂S0 , these seminorms are uniformly bounded on K , so the same constants C ( K )apply uniformly for all R∈K . This uniformity is required for the global explicit–formula bounds in Proposition 5.6 and matches the formulation of the EF–bank and Lyapunov equivalences in Main Theorem B and Theorem 1. Standing quantifier banner. Fix R∈S0 and the Gaussian family {ϖT}T>0, ϖT(t) = (√π T)−1e−t2/T2.(5.5) All constants depend only on R through finitely many S –seminorms and are independent of T . When we later work with a compact subset K⊂ S0 , these seminorms are uniformly bounded on K , so the corresponding constants C(K)apply uniformly for all R∈K. 5.1. Notation, conventions, and admissible objects. This subsection fixes the analytic conventions and admissible objects used in all subsequent arguments. The goal is to ensure that every pairing, explicit–formula identity, operator statement, and limit passage is rigorously posed with explicit T –dependence, and that dependence on the spatial kernel R always occurs through finitely many Schwartz seminorms (which are uniformly bounded on compact families K⊂S0 ). 36 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) No hidden uniformity in T is assumed beyond displayed factors such as 1 + log3(3+T). Fourier transform and Plancherel normalisation. We use the unitary 2π–Fourier convention b f(ξ) := ZR f(x)e−2πixξ dx, f(x) = ZRb f(ξ)e2πixξ dξ, (5.6) so Plancherel holds isometrically: ∥f∥2 2=∥b f∥2 2.(5.7) Completed zeta function and observable. Let ξ(s) = 1 2s(s−1) π−s/2Γs 2ζ(s),(5.8) which is entire of order 1and satisfies ξ ( s ) = ξ (1 −s )(cf. Titchmarsh [1, Chs. II–IV], Ivić [2, §6]). We measure the real–valued observable g(x, t) = log |ξ(x+it)|2,(5.9) which obeys the symmetry g(x, t)=g(1 −x, t) =⇒∂xg(1 2, t) = 0 whenever ξ(1 2+it)= 0.(5.10) Away from the zero set of ξ , the function ( x, t ) 7→ ∂xg ( x, t )is realanalytic and, for each fixed t , locally harmonic in x . Differentiating (5.9) gives the basic identity ∂xg(x, t) = 2 ℜξ′ ξ(x+it),(5.11) which is the starting point for all explicit–formula representations of ∂xg . In particular, ∂xg ( ·, t ) ∈L2 loc ( R )for each fixed t , due to the vertical-line bounds on ξ′/ξ (Lemma B.5). Local model near a zero. If ρ = β + iγ is a zero of multiplicity m , then ξ(s)=(s−ρ)mh(s), h(ρ)= 0,(5.12) so on a bidisc around ρwe obtain ∂xg(x, t) = 2m(x−β) (x−β)2+ (t−γ)2+∂xlog |h(x+it)|2.(5.13) The first term governs the universal cusp used in the neighbourhood–divergence lemma of §5.5; the second term is holomorphic in ( x, t )and is uniformly bounded on compact subsets. Thus all later asymptotics near ρ are fully controlled and Clay-compliant. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 37 Admissible spatial kernels. All spatial localisations use kernels R∈S(R), R real, even, nonnegative.(5.14) No global Fourier–positivity b R≥ 0is ever assumed. Since R is Schwartz, R∈L1∩L∞ , and all frequency-side pairings involving R and f′ are absolutely convergent whenever f′∈L2. Subclass for explicit–formula analysis. For the explicit–formula and Lyapunov analysis we use the subclass S0:= nR∈S(R) : Ris real and even about 1 2, R(1 2)=R′(1 2) = 0, R′′(1 2)>0, and we fix √R∈S(R)with (√R)2=R o. The quadratic vanishing at x = 1 2 cancels the leading on–line contribution from the Gamma block in the explicit formula (§5.7) while leaving off–line zeros fully exposed. Canonical regulators and recentering. A canonical family in S0is Rα(x) = (x−1 2)2e−α(x−1 2)2, α > 0.(5.15) As α↓ 0, Rα→ ( x−1 2 ) 2 pointwise and in S′ ( R ). These regulators are used only inside L2 pairings and explicit–formula coefficients; they are removed through the Clay–compliant order of limits (5.17) . All Clay–level conclusions concern arbitrary R∈S0 ; Rα serves only as a model. Admissible time windows. All t –integrals are taken against the mass–one Gaussians ϖT(t) = (√π T)−1e−t2/T2, T > 0.(5.16) These satisfy ∥ϖT∥∞ = 1 / ( √πT )and RRϖT ( t ) dt = 1. The T –dependence in all estimates therefore appears explicitly and the Gaussian imposes no hidden L2assumption in t. Other nonnegative, even Schwartz windows of unit mass would also be admissible, but we use the Gaussian family {ϖT}T>0 since the windowed zero–sum lemma in Appendix E produces explicit (1 + log3 (3 + T )) bounds for this choice. Measure–theoretic conventions. The zero ordinates {γ : ξ ( 1 2 + iγ ) = 0 } form a discrete, hence Lebesgue–null, subset of R . Dominated convergence, Fubini–Tonelli, and limit statements of the form t→γ are always interpreted modulo this null set: integrands may be altered on a set of t of measure zero without affecting Gaussian–weighted integrals. All exceptional sets are controlled explicitly in §5.10. 38 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Terminology: convexity, closed forms, and arrows. Convexity of quadratic forms is used in the standard sesquilinear sense (see Kato [21, §VI.1]); closedness and Friedrichs extensions follow Reed–Simon [19, Thm. VIII.15, Thm. VIII.3.11, Cor. VIII.3.12]. Monotone arrows a↓a0 , a↑a0 and one–sided limits x→a± are interpreted in their usual analytic sense. Regulator limits and order of operations. All regulators are removed in the order T→ ∞ (remove time window), α ↓0(remove spatial regulator). (5.17) The limit T→ ∞ is taken using explicit windowed bounds  1 + log3 (3 + T )  for the Lyapunov energies, together with dominated convergence on vertical lines (Lemma B.5) and the measure lemmas in §5.10. Only after these bounds are established do we pass to the limit α↓ 0, using monotone convergence of closed forms and strong resolvent convergence for HRα (Appendix A). This limit scheme is built into every explicit–formula and Lyapunov argument. Role in later sections. The symmetry (5.10) furnishes a stationary cut at x = 1 2 for the ERU embedding of §3.6. The identity (5.11) allows explicit–formula control of ∂xg and hence of the horizontal energy. The local model (5.13) produces the universal cusp used in §5.5. The admissible kernel class (5.14) (i.e. R∈S0 ) and the limit order (5.17) are used throughout the proof spine and are embedded into the finite–time Lyapunov framework of §4. 5.2. Spectral uniqueness of Rα via Sturm–Liouville. This subsection justifies the use of the Gaussian–polynomial profiles Rα(x) := x−1 22e−α(x−1 2)2, α > 0,(5.18) as canonical admissible kernels in S0 (cf. (5.15) ). Each Rα is real, even in the recentered coordinate y=x−1 2, nonnegative, rapidly decaying, and vanishes quadratically at x = 1 2 . We record two structural features: (a) a constrained Rayleigh–Ritz (Sturm–Liouville) principle which singles out Rα as the natural minimiser among even kernels with R(1 2) = 0; (b) a strictly positive band for the shift–corrected Fourier profile eπiξ c Rα ( ξ ), which underlies the band–positivity arguments in Lemma 5.1. These are the only properties of Rα used downstream, and they serve purely as analytic conveniences; Clay–level conclusions remain uniform THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 39 in R∈S0 , and any kernel sufficiently close to Rα in the Schwartz topology may be used in its place. Variational selection (harmonic oscillator with R (0) = 0 constraint). Work in the recentered variable y = x−1 2 and write R ( y ). Fix ω > 0 and consider the quadratic functional Eω[R] := ZR|R′(y)|2+ω2y2|R(y)|2dy, (5.19) on H1(R)∩L2(R), restricted to the closed subspace V:= R∈H1(R)∩L2(R):R(−y) = R(y)for all y, R(0) = 0.(5.20) The Euler–Lagrange operator is the harmonic oscillator Lω:= −d2 dy2+ω2y2,(5.21) whose eigenfunctions are ϕ(ω) n(y) = cnHn(√ω y)e−ωy2/2, λn(ω) = ω(2n+1), n ≥0,(5.22) with Hn the Hermite polynomials; even eigenfunctions correspond to n even. Proposition 5.1 (Constrained Rayleigh minimiser).Fix ω > 0. Among all R∈ V with ∥R∥2 = 1, the unique Rayleigh minimiser of Eω lies in span{ϕ(ω) 0, ϕ(ω) 2}and, up to scale, R⋆(y) = y2e−ωy2/2.(5.23) Equivalently, with α = ω/ 2and y = x−1 2 , the minimiser is R⋆ ( x ) = Rα(x)as defined in (5.18). Proof. On the even subspace, Lω has eigenvalues λ2n ( ω ) = ω (4 n + 1) with eigenfunctions ϕ(ω) 2n . The constraint R (0) = 0 imposes one nontrivial linear condition on this even subspace. Since ϕ(ω) 2n (0)  = 0 for all n , the constraint removes one dimension, and the Rayleigh minimiser must lie in span{ϕ(ω) 0, ϕ(ω) 2}. Using H0(z)=1and H2(z) = 4z2−2, one computes ϕ(ω) 2(y)+2ϕ(ω) 0(y)∝y2e−ωy2/2.(5.24) Strict convexity of Eω on the affine constraint set {R∈ V : ∥R∥2 = 1 } yields uniqueness of the minimiser up to phase. Existence of a minimiser follows from standard compactness of the harmonic oscillator form domain in L2(R)and lower semicontinuity of Eω.□ 40 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Thus, among even R with R (0) = 0, the Gaussian–quadratic profile is the simplest variationally distinguished choice. This is the kernel we adopt for explicit estimates; the Clay–compliant arguments are, however, uniform in R∈S0 and depend only on the structural properties recorded below. Fourier profile and low–frequency positivity band. For the centred kernel y2e−αy2, the 2π–Fourier transform (5.6) yields \ y2e−αy2(ξ) = √π α5/2α 2−π2ξ2e−π2ξ2/α.(5.25) Recentring from y back to x−1 2 multiplies the Fourier transform by e−2πi(1 2)ξ=e−πiξ, so eπiξ c Rα(ξ) = √π α5/2α 2−π2ξ2e−π2ξ2/α ∈R.(5.26) Hence the shift–corrected Fourier profile eπiξ c Rα ( ξ )is strictly positive on the symmetric band |ξ|< δα, δα:= √α √2π.(5.27) This is exactly the positivity used together with Lemma 5.1 to obtain frequency–diagonal control in the low–frequency range in §5.7. Scaling laws. In the centred coordinate, y2e−αy2=α−1(√α y)2e−(√α y)2,(5.28) and the Fourier transform obeys the standard scaling law c Rα(ξ)=α−3/2e−πiξ \ y2e−y2ξ/√α.(5.29) Thus the low–frequency positivity window in (5.27) scales like √α , while the amplitude scales like α−2 . In particular, the band |ξ|≤δα is always nontrivial for α > 0and shrinks to { 0 } as α↓ 0, consistent with the regulator role of Rα. Stability under admissible perturbations. In §5.7 only two structural features of Rαare used: •quadratic vanishing at x=1 2; • the strict low–frequency positivity band (5.27) for the shift– corrected profile eπiξ c Rα(ξ). Both properties are open in the Schwartz topology. Lemma 5.1 (Robustness of quadratic vanishing and low–frequency positivity).Fix α > 0and let δα be as in (5.27) . There exist εα, ηα> 0 THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 41 such that if S∈S ( R )is real, even in y = x−1 2 , satisfies S ( 1 2 )=0, and max 0≤j≤2sup |y|≤1|∂j y(S−Rα)(1 2+y)|< εα,∥S−Rα∥L1(R)< ηα,(5.30) then: (i) S has a zero of exact order 2at x = 1 2 , i.e. S ( 1 2 ) = S′ ( 1 2 )=0 and S′′(1 2)>0; (ii) the shift–corrected profile satisfies ℜeπiξ b S(ξ)≥1 2 √π α5/2α 2−π2ξ2e−π2ξ2/α for |ξ|≤δα.(5.31) Proof. (i) In the coordinate y = x−1 2 , Rα ( y ) = y2e−αy2 satisfies Rα (0) = R′ α (0) = 0 and R′′ α (0) = 2. The bounds in (5.30) imply S (0) = S′ (0) = 0 (by evenness and smallness) and |S′′ (0) − 2 |< εα , so S′′ (0) > 1if εα< 1. (ii) The map f7→ b fis continuous L1→L∞under (5.6), hence sup ξ∈R|b S(ξ)−c Rα(ξ)|≤∥S−Rα∥L1(R)≤ηα.(5.32) Choosing ηα smaller than half the minimum of eπiξ c Rα ( ξ )on [ −δα, δα ] gives (5.31). □ Remark 5.1 (Downstream usage and relation to §3.2).Lemma 5.1 shows that in §5.7 and §5.5, Rα can be replaced by any nearby S satisfying (5.30) without affecting the argument. Both the Sturm–Liouville construction here and the moment–constrained variational characterisation of §3.2 single out kernels with the same two structural properties: quadratic vanishing at x = 1 2 and a shift–corrected low–frequency positivity band. No further structure of Rα is used in the Clay–level proof of the Lyapunov–explicit–formula equivalence (Main Theorem B) and the thesis theorem of §4. 5.3. Self–adjointness, semi–coercivity, and compactness for HR . We formalise here the measurement operator acting in the spatial variable x . Throughout this subsection, t∈R is fixed and all statements concern only the x –variable. In the proof spine, HR denotes the nonnegative self–adjoint Friedrichs operator associated with the quadratic form qR (called “HC” in earlier drafts; see the Operator Dictionary). This subsection is auxiliary: it provides a canonical functional–analytic realisation of HR , but all RH–level conclusions in the Lyapunov–explicit– formula equivalence (Main Theorem B) and in the thesis Theorem 1 remain valid for every R∈S0. 48 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) • (Divergence–form perturbations). If W∈L∞ ( R )satisfies |W(x)| ≤ ϑR(x)a.e. for some ϑ∈[0,1), define δq[h] := ZR W(x)|h′(x)|2dx. (5.62) Then |δq [ h ] | ≤ ϑqR [ h ], so qR + δq is closed and semibounded by KLMN. This covers all small bounded perturbations of the measurement weight within qR (e.g. replacing R by (1 + η ) R with ∥η∥∞<1). Remark 5.3 (Short–range potentials).Since qR controls only the weighted energy RRR ( x ) |h′ ( x ) |2dx , general L1 loc potentials need not be qR –form–bounded. When such terms appear, we work with the regulator Rα,ε from above, apply KLMN relative to the coercive form qRα,ε on compact supports, and then send ε↓ 0using Lemma 5.6. No uniformity in ε is required in the RH analysis. Compactness: what holds and what does not. Because R is noncompact and Rdecays at infinity, compactness must be understood locally. Proposition 5.4 (Absence of global L2 –compactness).The embedding ( D ( qR ) ,∥·∥qR ) ,→L2 ( R )is not compact. Indeed, for hn ( x ) = ϕ ( x−n ) with fixed ϕ∈C∞ c ( R ),( hn )is bounded in ∥·∥qR and qR [ hn ] → 0, but (hn)has no L2–convergent subsequence. Proposition 5.5 (Compactness under confinement).For ω > 0define qR,ω[h] := qR[h]+ω2ZR x2|h(x)|2dx. (5.63) Then the operator associated with qR,ω has compact resolvent on L2 ( R ). The graph norm controls both a weighted derivative and a quadratically growing potential; Rellich–Kondrachov on bounded intervals and tightness at infinity yield compact embedding D ( qR,ω ) ,→L2 ( R )(cf. Reed–Simon [20, Ch. XIII]). Remark 5.4 (Weighted target spaces).Compactness can alternatively be recovered by working in L2 ( ⟨x⟩−kdx )for k > 1, which penalises translation at infinity. The RH framework does not require resolvent compactness in L2 : all EF, Plancherel, and Lyapunov arguments use only closedness, self–adjointness, and the exact identity ⟨HRh, h⟩ = qR[h]. Relevance to the RH framework. • KLMN stability ensures robustness of all operator manipulations under the perturbations actually used downstream (bounded potentials, small perturbations of Ron the test side). THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 49 • Compact resolvent is never needed; all RH–critical steps rely only on closability, self–adjointness, and the weighted form identity ⟨HRh, h⟩ = qR [ h ]together with the explicit–formula analysis in §5.7. 5.4. Analytic–continuation filter and off–line zero asymptotics. Quantifier banner. Fix the admissible Gaussian family {ϖT}T>0 with ϖT ( t ) = ( √πT ) −1e−t2/T2 , and work with spatial kernels R∈S ( R )as in Section 5.1. All constants below depend only on the size of a fixed bidisc and on finitely many S –seminorms of the functions involved, and are independent of T. For s=x+it with ξ(s)= 0, recall g(x, t) := log |ξ(x+it)|2, ∂xg(x, t)=2ℜξ′ ξ(x+it).(5.64) Statements “at t = γ ” are interpreted as limits t→γ under the Gaussian window ϖT(cf. Section 2.2 and Lemma B.5). Lemma 5.7 (Local factorisation and filtered decomposition).Let ρ = β + iγ be a nontrivial zero of ξ (hence of ζ ) of multiplicity m≥ 1. Then there exist ε0, δ0> 0and an analytic, nonvanishing function h on the bidisc U:= {(x, t)∈R2:|x−β| ≤ ε0,|t−γ|≤δ0}(5.65) such that ξ(s)=(s−ρ)mh(s), s =x+it ∈ U.(5.66) Hence ∂xg(x, t) = 2m(x−β) (x−β)2+ (t−γ)2+b(x, t),(x, t)∈ U,(5.67) where b ( x, t ) = ∂xlog |h ( x + it ) |2 . Moreover, there exist constants B, L > 0(depending only on Uand h) such that |b(x, t)| ≤ B, |b(x, t)−b(β, γ)| ≤ L|x−β|+|t−γ|,(x, t)∈ U. (5.68) Proof. The factorisation (5.66) is the Weierstrass local representation of an entire function at a zero of multiplicity m : there is an analytic h with h ( ρ )  = 0 such that ξ ( s ) = ( s−ρ ) mh ( s )in a neighbourhood of ρ . Since log |ξ ( s ) |2 = 2 ℜlog ξ ( s ), differentiating in x gives ∂xg = 2 ℜ ( ξ′/ξ ) wherever ξ= 0, and inserting (5.66) yields (5.67). As h is analytic and nonvanishing on U , after possibly shrinking U we may assume it is simply connected, so log h is analytic on U . Then b ( x, t ) = ∂xlog |h ( x + it ) |2 is real analytic on U . On a slightly smaller bidisc U′⋐U , Cauchy estimates give uniform bounds on first derivatives of log h , hence on b and its first derivatives. This 50 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) implies both the uniform bound |b ( x, t ) |≤B and the Lipschitz estimate |b ( x, t ) −b ( β, γ ) | ≤ L ( |x−β| + |t−γ| )on U′ , which after renaming U′ as Ugives (5.68). □ Corollary 5.1 (Universal singular slope; on–line/off–line dichotomy). In the setting of Lemma 5.7: (1) For fixed twith t→γand xnear β, ∂xg(x, t) = 2m(x−β) (x−β)2+ (t−γ)2+O(1),(5.69) with the O(1) uniform on U. In particular, at t=γ, ∂xg(x, γ) = 2m x−β+O(1) (x→β).(5.70) (2) If β = 1 2 (off–line), then for any admissible kernel R∈S ( R ) with R ( β ) > 0, the integrand R ( x ) |∂xg ( x, γ ) |2 has a nonintegrable u−2 singularity at u = x−β . If β = 1 2 and R∈S0 (so R ( 1 2 ) = 0 with a quadratic zero), the factor ( x−1 2 ) 2 in R cancels this principal pole in the on–line contribution to the explicit–formula energy in Section 5.7. Proof. Inserting the bounds (5.68) for b into (5.67) immediately gives the stated O(1) terms and the uniformity on U. For (2), set u=x−β. From the first part we have ∂xg(x, γ) = 2m u+O(1) (u→0),(5.71) so |∂xg(x, γ)|2=4m2 u2+O(1) (u→0),(5.72) with implicit constants depending only on U and h . If R is continuous at β with R ( β ) > 0, there exists ε > 0and c > 0such that R ( β + u ) ≥c for |u|≤ε. Then Z|u|≤ε R(β+u)|∂xg(β+u, γ)|2du ≥cZ|u|≤ε 4m2 u2du −C= +∞,(5.73) because R|u|≤εu−2du = + ∞ . Thus R ( x ) |∂xg ( x, γ ) |2 is not locally integrable at x=β. When β = 1 2 and R∈S0 , write R ( x ) = ( x−1 2 ) 2e R ( x )with e R smooth and bounded near x=1 2. With u=x−1 2we have R(x)|∂xg(x, γ)|2=u2e R(x)4m2 u2+O(1)= 4m2e R(x) + O(u2), (5.74) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 51 which is locally integrable near u = 0. Hence the on–line singularity is exactly cancelled by the quadratic vanishing of Rat x=1 2.□ Lemma 5.8 (Gaussian window admissibility on vertical lines).Fix ϵ∈ (0 ,1 2 )and σ∈ [ 1 2−ϵ, 1 2 + ϵ ]. There exists Cϵ> 0such that for all t∈R,ξ′ ξ(σ+it)≤Cϵ1 + log(2 + |t|).(5.75) Consequently, for every admissible R∈S(R), ER(t) := ZR R(x)|∂xg(x, t)|2dx ≪R,ϵ 1 + log2(2+|t|),(5.76) and hence, for every T > 0, ZR ER(t)ϖT(t)dt ≪R,ϵ ZR1 + log2(2+|t|)ϖT(t)dt, (5.77) uniformly in T . The right–hand side is finite for each T , so all Gaussian– windowed Fubini/dominated–convergence arguments in t are admissible (cf. Lemma B.5). Proof. The vertical–strip estimate for ξ′/ξ follows from the functional equation for ξ , Stirling’s formula for Γ ′/ Γ, and the classical bounds for ζ and ζ′/ζ (cf. Titchmarsh [1, Ch. III–IV], Ivić [2, §6]). In particular, for every fixed ϵ∈(0,1 2)there is Cϵ>0such that ξ′ ξ(σ+it)≤Cϵ1 + log(2 + |t|)(5.78) for all t∈R and all σ∈ [ 1 2−ϵ, 1 2 + ϵ ]. Since ∂xg ( x, t ) = 2 ℜ ( ξ′/ξ )( x + it ), we then have |∂xg(x, t)| ≪ϵ1 + log(2 + |t|), x ∈[1 2−ϵ, 1 2+ϵ].(5.79) Let R∈S ( R )be admissible. As R is rapidly decaying and bounded, and the vertical–line bounds above hold on every compact subinterval of (0,1) containing the support where Ris appreciable, we obtain ER(t) = ZR R(x)|∂xg(x, t)|2dx ≪R,ϵ 1 + log2(2+|t|),(5.80) where the implied constant depends only on finitely many Schwartz seminorms of Rand on ϵ. This is (5.76). Multiplying by the nonnegative Gaussian window ϖT ( t )and integrating over tgives ZR ER(t)ϖT(t)dt ≪R,ϵ ZR1 + log2(2+|t|)ϖT(t)dt. (5.81) 52 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) For each fixed T > 0, the Gaussian decay of ϖT and the polynomial growth of log2 (2 + |t| )ensure that the right–hand side is finite. This uniform majorant justifies all subsequent uses of Fubini and dominated convergence in tunder Gaussian windowing (cf. Lemma B.5). □ Remark 5.5 (Symmetry at the critical line and exceptional ordinates). By the functional equation, g ( x, t ) = g (1 −x, t ). Hence ∂xg ( 1 2, t ) = 0 whenever ξ ( 1 2 + it )  = 0. The exceptional set {t : ξ ( 1 2 + it ) = 0 } is discrete and therefore null with respect to the measures ϖT ( t ) dt . Time– averaged statements at x = 1 2 are thus unaffected (cf. the ERU/flux formalism in Section 3.6). Summary and downstream use. Lemma 5.7 isolates the universal singular part of ∂xg near a zero, with a controlled C1 remainder. Corollary 5.1 shows that any off–line zero forces a nonintegrable R –weighted cusp whenever R ( β ) > 0, while on–line zeros are neutralised by the quadratic vanishing in admissible kernels R∈S0 . Lemma 5.8 guarantees that Gaussian windowing is compatible with all limit operations in t and provides a uniform majorant for ER ( t ). These inputs are used verbatim in the Neighbourhood–Divergence Lemma (Section 5.5) and in the windowed explicit–formula energy analysis (Section 5.7), which together feed into the Lyapunov/cusp framework of Theorem 1). Clay–compliance note. All weights R and ϖT appear solely as admissible test functions in L2 pairings; no modification of ζ or ξ occurs. Limits in T are taken only after establishing the window–uniform vertical–line bounds above (and Lemma B.5). Thus all conclusions of this subsection refer to the unaltered analytic behaviour of the classical zeta function and are fully Clay–compliant. 5.5. Neighbourhood–divergence lemma (cylindrical flux): statement and setup. Quantifier banner. Fix an admissible spatial kernel R∈S ( R )(real, even, nonnegative) and the normalised Gaussian window family {ϖT}T>0 with ϖT ( t ) = ( √πT ) −1e−t2/T2 . All constants below depend only on finitely many S –seminorms of R and on the C1 bounds for the remainder term b in the local expansion of ξ′/ξ (see Section 5.4), and are independent of T . The analysis is purely local in the t–variable. Throughout this subsection we work with the completed zeta function ξ(s) = 1 2s(s−1) π−s/2Γs 2ζ(s), g(x, t) := log |ξ(x+it)|2.(5.82) Then g ( x, t ) = g (1 −x, t )and g is real–analytic away from the zero set of ξ . If R ( β ) > 0at some point β∈R , continuity of R yields cR> 0 THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 53 and εR>0such that R(x)≥cR>0for all x∈[β−εR, β +εR].(5.83) Cylindrical flux. For x0∈R , ε > 0, and t∈R , define the localised (or cylindrical) flux across the vertical cut {x=x0}by FR,ε(x0, t) := Zx0+ε x0−ε R(x)|∂xg(x, t)|2dx. (5.84) With the time window ϖTwe define the windowed cylindrical flux FR,ε,T (x0) := ZRFR,ε(x0, t)ϖT(t)dt. (5.85) Recall also the global weighted energy ER(t) := ZR R(x)|∂xg(x, t)|2dx, FR,ε(x0, t)≤ER(t).(5.86) Thus FR,ε is a local probe of ER(t)near x0. Local zero model. Let ρ = β + iγ be a zero of ξ of multiplicity m≥ 1. By Section 5.4 there exists a bidisc neighbourhood U of ( β, γ )and an analytic, nonvanishing function hon Usuch that ξ(s)=(s−ρ)mh(s), s =x+it ∈ U.(5.87) Therefore ∂xg(x, t) = 2m(x−β) (x−β)2+ (t−γ)2+b(x, t),(5.88) where b ( x, t ) = ∂xlog |h ( x + it ) |2 is C1 (hence bounded and Lipschitz) on U. Lemma 5.9 (Neighbourhood divergence of cylindrical flux at an off–line zero).Let ρ = β + iγ be a zero of ξ with β = 1 2 and multiplicity m≥ 1. Fix an admissible kernel R∈S ( R )with R ( β ) > 0. Then there exist ε0, δ0> 0and constants c1, c2, C > 0—depending only on m , the values of R in a small neighbourhood of β , and the local C1 bounds for b —such that for all 0< ε ≤ε0and all twith 0<|t−γ|< δ0, c1 |t−γ|−C≤ FR,ε(β, t)≤c2 |t−γ|+C. (5.89) In particular,Z|t−γ|<δ FR,ε(β, t)dt = +∞(0 < δ ≤δ0),(5.90) and for every T > 0, FR,ε,T (β)=+∞,(5.91) 54 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) since ϖT is continuous with ϖT ( γ ) > 0. Consequently the global weighted energy diverges: ER(γ)=+∞,(5.92) and for t→γ, ER(t)≥ FR,ε(β, t)≍ |t−γ|−1.(5.93) Remark 5.6 (Sharp leading constant and stability).Write u = x−β and a = t−γ . Using (5.88) , R ( β ) > 0, and the boundedness of b , we have FR,ε(β, t) = Zε −ε R(β+u) 2mu u2+a2+b(β+u, t) 2 du. (5.94) Freezing R ( β + u ) = R ( β ) + O ( u )on |u| ≤ ε and expanding the square, the dominant singular contribution comes from 4m2R(β)Zε −ε u2 (u2+a2)2du =2πm2R(β) |a|+OR,ε(1),(5.95) since Zε −ε u2 (u2+a2)2du =1 |a|arctanε |a|−ε ε2+a2=π 2|a|+Oε(1) (a→0). (5.96) By choosing ε sufficiently small, the constants c1 and c2 in (5.89) may be taken arbitrarily close to 2 πm2R ( β ). If finitely many other zeros lie in |t−γ| ≤ η , their contributions are bounded and can be absorbed into the O(1) term via Cauchy–Schwarz. Corollary 5.2 (Uniform cusp bounds on compact kernel families).Let ρ = β + iγ be a zero of ξ with β = 1 2 and multiplicity m≥ 1. Let K⊂S(R)be compact and assume inf R∈KR(β)≥c0>0.(5.97) Then there exist ε0, δ0> 0and constants c1, c2, C > 0(depending only on m , c0 , and finitely many S –seminorms of R as R ranges over K ) such that for every R∈K , every 0 < ε ≤ε0 , and every t with 0<|t−γ|< δ0, c1 |t−γ|−C≤ FR,ε(β, t)≤c2 |t−γ|+C. (5.98) In particular, for all R∈Kand all T > 0we have FR,ε,T (β)=+∞, ER(γ)=+∞,(5.99) and ER(t)≍ |t−γ|−1as t→γ, with constants uniform in R∈K. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 55 Proof. Continuity of R7→ R ( β )in the Schwartz topology and compactness of K give a uniform positive lower bound c0 for R ( β )on K . The constants in Lemma 5.7 and Lemma 5.8 depend only on the size of the bidisc and finitely many seminorms of R ; these seminorms are uniformly bounded on K . The proof of Lemma 5.9 therefore carries over with constants uniform in R∈K.□ Clay–compliance note. All quantities in (5.84) – (5.86) are L2 pairings of the classical observable g = log |ξ|2 against admissible kernels R (and ϖT , when present). No modification of ζ or ξ occurs, and no external dynamics is imposed. The divergences (5.90)–(5.92) and Corollary 5.2 record intrinsic behaviour of the unaltered ξ in any neighbourhood of an off–line zero, uniformly over compact families of admissible kernels with R(β)>0. Roadmap for the proof. In §5.6 we insert the expansion (5.88) into (5.84) , freeze R ( x ) = R ( β ) + O ( |x−β| )on [ β−ε, β + ε ], and use the exact integral identity Zε −ε u2 (u2+a2)2du =1 |a|arctanε |a|−ε ε2+a2=π 2|a|+Oε(1) (a→0). (5.100) The bounded b –terms and the linear variation of R contribute O (1), yielding the two–sided estimate (5.89) . In the global argument this local divergence is then combined with the window–uniform explicit–formula energy bound from Section 5.7, producing the implication (ii) ⇒ (i) in Theorem 1. 5.6. Leading–order asymptotics and kernel reduction. Let ρ = β+iγ be as in Lemma 5.9, with the standard analytic factorisation ξ(s)=(s−ρ)mh(s), h analytic on a bidisc about ρ, h(ρ)= 0. (5.101) Writing s = x + it and recalling g ( x, t ) = log |ξ ( x + it ) |2 , we obtain the exact decomposition g(x, t)=mlog (x−β)2+ (t−γ)2+ log |h(x+it)|2,(5.102) ∂xg(x, t) = 2m(x−β) (x−β)2+ (t−γ)2+b(x, t), b(x, t) := ∂xlog |h(x+it)|2. (5.103) By analyticity of h, there exists a bidisc U:= (x, t)∈R2:|x−β| ≤ ε0,|t−γ| ≤ δ0(5.104) 56 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) on which b∈C1(U). Hence there exist constants B, L > 0such that |b(x, t)| ≤ B, |b(x, t)−b(β, γ)|≤L|x−β|+|t−γ|,(x, t)∈ U. (5.105) All O ( · )–constants below depend only on m , on finitely many S –seminorms of R restricted to [ β−ε0, β + ε0 ], and on the C1 –bounds of b on U , and are uniform for 0<|t−γ|< δ0. Reduction of the cylindrical flux to the universal kernel. Fix ε∈ (0 , ε0 ]. Set a:= t−γ= 0, u := x−β, (5.106) and let R∈S ( R )be admissible with R ( β ) > 0. The cylindrical flux at x0=βis FR,ε(β, t) = Zε −ε R(β+u)|∂xg(β+u, t)|2du. (5.107) Substituting (5.103) and expanding yields FR,ε(β, t)=4m2Zε −ε R(β+u)u2 (u2+a2)2du (principal term) + 4mZε −ε R(β+u)u b(β+u, t) u2+a2du (cross term) +Zε −ε R(β+u)|b(β+u, t)|2du. (remainder) We now track each contribution as a=t−γ→0. (1) Principal term. Since R∈S, a Taylor expansion at βgives R(β+u) = R(β)+R′(β)u+O(u2) (|u| ≤ ε0).(5.108) The R′ ( β ) u contribution vanishes upon integration against the even kernel u2 ( u2 + a2 ) −2 , because the integrand is then odd in u on [ −ε, ε ]. Hence 4m2Zε −ε R(β+u)u2 (u2+a2)2du = 4m2R(β)Iε(a)+OR,ε(1),(5.109) where the universal model integral is Iε(a) := Zε −ε u2 (u2+a2)2du =1 |a|arctan ε |a|−ε ε2+a2.(5.110) A direct expansion as a→0yields Iε(a) = π 2|a|−1 ε+Oε(a2),(5.111) so the singular growth is exactly π 2|a|. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 57 (2) Cross term. Using (5.105) and the Taylor expansion of R, R(β+u)b(β+u, t)=R(β)b(β, γ)+O(|u|+|a|),(5.112) uniformly for |u| ≤ ε , |a|< δ0 . The constant term integrates to zero, since Zε −ε u u2+a2du = 0.(5.113) For the remainder, write e b ( u, a ) := b ( β + u, γ + a ) −b ( β, γ ); then |e b(u, a)| ≪ |u|+|a|, and therefore Zε −ε ue b(u, a) u2+a2du ≪Zε 0 u(|u|+|a|) u2+a2du ≪1 + |a|log(ε/|a|).(5.114) Thus the cross term contributes OR,ε,h (1) as a→ 0and is negligible compared to the principal |a|−1singularity. (3) Remainder term. By (5.105) and boundedness of R on [ β−ε, β + ε ], we have Zε −ε R(β+u)|b(β+u, t)|2du ≪R,ε,h 1,(5.115) uniformly for 0<|a|< δ0. Asymptotics and leading constant. Combining (principal term) with (5.109)–(5.111) gives, for some ε∗∈(0, ε0]and all 0<|t−γ|< δ∗, FR,ε∗(β, t) = 4m2R(β)Iε∗(t−γ)+Oε∗,R,h(1) = 2πm2R(β) |t−γ|+Oε∗,R,h(1). (5.116) Since R ( β ) > 0, continuity of R allows us to choose ε∗> 0so small that inf |u|≤ε∗ R(β+u)≥1 2R(β),sup |u|≤ε∗ R(β+u)≤2R(β).(5.117) With such a choice, we may take explicit constants c1:= 2πm2inf |u|≤ε∗ R(β+u), c2:= 2πm2sup |u|≤ε∗ R(β+u), C := sup |t−γ|<δ∗|Oε∗,R,h(1)| (5.118) to obtain the two–sided estimate c1 |t−γ|−C≤ FR,ε∗(β, t)≤c2 |t−γ|+C. (5.119) Shrinking ε∗↓ 0forces c1, c2→ 2 πm2R ( β ), giving the sharp leading constant. 64 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) with implicit constants depending only on finitely many seminorms of R. The explicit formula decomposes BR(ν, t)=GR(ν, t) + PR(ν, t) + ZR(ν, t),(5.162) where ZR ( ν, t )is the zero–block coefficient in the frequency representation, as analysed in Appendix E, including the fixed low–frequency ϑ–normalisation built into the definition of ZRthere. For frequency smoothing we use the standard convolutional envelope BR,η(ν, t) := (BR(·, t)∗Kη)(ν) = ZR BR(ν′, t)Kη(ν−ν′)dν′,(5.163) where {Kη}η>0 is a fixed family of even, nonnegative Schwartz kernels (approximate identities) used to justify frequency–side truncations and robustness. In particular, the zero–block estimate below already supplies an integrable low–frequency envelope in ν , so no additional ν –cutoff is required for the convergence of the ν–integrals. The windowed zero–sum lemma (Theorem 2 in Appendix E) asserts that for all ν∈R, ZR|ZR(ν, t)|2ϖT(t)dt ≤CZ(R) (1+|ν|)21+log3(3+T)1+log2 2+|ν|−1, (5.164) where CZ ( R )depends only on finitely many S –seminorms of R and is independent of T (and we interpret log (2 + |ν|−1 )as 0at ν = 0, as in Appendix E). The additional factor 1 + log2 (2 + |ν|−1 )is a harmless, integrable low–frequency envelope; it replaces any false claim of pointwise ν –decay uniformly in t (which is impossible at t = γ ). The logarithmic exponents arise from the unit–band zero density, bilinear zero–zero pairing in the mean square, and the gamma–tail under the smoothing window; see Appendix E for the precise implementation. Integrating (5.164) in ν and using Tonelli’s theorem (nonnegative integrand) gives ZRZR|ZR(ν, t)|2ϖT(t)dt dν ≤CZ(R)1+log3(3+T)ZR 1 + log2(2+|ν|−1) (1+|ν|)2dν. (5.165) The ν –integral on the right is finite (absolute) since (1 + |ν| ) −2 is integrable at infinity and R1 0log2 (1 /ν ) dν < ∞ at the origin (equivalently, by Lemma A.12). Absorbing its value into the constant yields BZ[R;T]≤C′ Z(R)1 + log3(3+T),(5.166) where BZ [ R ; T ]denotes the zero contribution to the windowed energy in the frequency picture (cf. (5.162) and the bookkeeping after (5.161) ), THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 65 and C′ Z ( R )again depends only on finitely many Schwartz seminorms of Rand is independent of T. Moreover, if R ranges in a compact K⊂S0 , the constants CZ ( R ) and C′ Z ( R )may be chosen uniformly in R∈K ; we then write CZ ( K ), C′ Z ( K )for the corresponding envelopes. Finally, the same bounds hold with ZR replaced by the smoothed profile ( ZR ( ·, t ) ∗Kη )( ν ), by Young’s inequality (since ∥Kη∥L1= 1) and Tonelli. Step 6: Assembly and T –profile. Combining (5.138) with the bounds (5.147) and (5.166) , together with the Dirichlet–Euler block estimate from Appendix F.2, gives ZR ER(t)ϖT(t)dt ≤12BΓ[R;T]+BZ[R;T]+BP[R;T](5.167) ≤12CΓ(R)1 + log2(3+T)+CDE(R) +C′ Z(R)1 + log3(3+T) ≤C∗(R)1 + log3(3+T), with C∗ ( R )depending only on finitely many S –seminorms of R and independent of T . Renaming C ( R ) := C∗ ( R )yields (5.125) , completing the proof of Proposition 5.6. Remark 5.7 (Tonelli viewpoint and full-domain control).Throughout, the windowed energy ZR ER(t)ϖT(t)dt (5.168) is understood in the Tonelli sense as the extended-real double integral of the nonnegative integrand R(x)|∂xg(x, t)|2ϖT(t): ZR ER(t)ϖT(t)dt =ZZR2 R(x)|∂xg(x, t)|2ϖT(t)dx dt. (5.169) The explicit–formula decomposition in Steps 1–5 controls this full double integral directly, via the linear explicit formula and blockwise bounds, and does not rely on any pointwise finiteness of ER ( t )or on the removal of exceptional ordinates. In particular, possible cusp times t = γ at off–line zeros—where ER ( t )may be infinite—do not affect either side of (5.125) ; the Lyapunov–cusp contradiction in §5.5–§5.6 and §4 arises precisely from opposing this global Tonelli bound to the local |t−γ|−1 divergence forced by any off–line zero. Lemma 5.10 (A.e. EF–Plancherel compatibility).In the setting of Proposition 5.6 there exists a null set NR⊂R such that for every 66 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) t /∈NRone has ER(t) = ZR R(x)|ft(x)|2dx =ZR|BR(ν, t)|2dν, (5.170) with both sides finite, and the weighted Plancherel identity of Lemma D.1 applies. The exceptional set NR may be chosen to contain all ordinates of off–line zeros, and has Lebesgue measure zero; hence for every mass–one Schwartz window ωT, ZR ER(t)ωT(t)dt =ZR\NR ER(t)ωT(t)dt, (5.171) ZRZR|BR(ν, t)|2ωT(t)dt dν =ZR\NRZR|BR(ν, t)|2ωT(t)dt dν. (5.172) In particular, isolated cusp times t = γ at off–line zeros do not affect any of the windowed EF identities. Proof. Proof. From (5.125) with a fixed T > 0and ϖT> 0a.e. we obtain that ER ( t ) <∞ for a.e. t∈R ; declare NR to be the complement of this full–measure set, enlarged if necessary to include all ordinates of off–line zeros. For t /∈NR the function √R ft lies in L2 ( R )and belongs to the form domain of qR , so the weighted Plancherel identity of Lemma D.1 applies to give (5.170) . Since NR has Lebesgue measure zero, it does not contribute to any windowed integral in t , and (5.171) – (5.172) follow by restricting the t–integration to R\NR.□ Remarks. • Cancellation at the central line and the log–cube. The quadratic vanishing of R at x = 1 2 (and its even symmetry about 1 2 ) is exploited in the detailed frequency–side analysis of the zero block in Appendix E: it removes the principal on–line pole and enables the standard Γ–zero cancellation in ξ′/ξ at the critical line. At the level of the bound (5.125) this manifests as a polylogarithmic growth in T , with three independent logarithmic factors (zero density in unit bands, bilinear zero–zero pairing, and the window/gamma tail), giving the overall profile ZR ER(t)ϖT(t)dt =OR1 + log3(3+T).(5.173) There is no known unconditional mechanism to reduce this to O(log2T). • Fourier conventions and no b R≥ 0requirement. We retain the 2 π –Fourier normalisation in x from §5.1, while the Gaussian window uses the plain oscillation (5.154) in t . At THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 67 no point is global Fourier–positivity b R≥ 0assumed; x –space positivity of R and standard decay of b R suffice, together with the explicit–formula structure and the weighted Plancherel identities in Appendix D. Promotion to compact kernel families (EF–bank). For the finite– time Lyapunov framework of §4 we require a uniform control over compact families of kernels, with an explicit  1 + log3 (3 + T )  growth profile. This is an immediate consequence of Proposition 5.6 and the finiteness of the relevant seminorms on compact sets. Corollary 5.3 (EF–bank for compact kernel families).Let K⊂S0 be compact in the Schwartz topology. Then there exists a constant C(K)<∞such that for all T > 0, sup R∈KZR ER(t)ϖT(t)dt ≤C(K)1 + log3(3+T).(5.174) Proof. In the proof of Proposition 5.6 the constants CΓ ( R ), CDE ( R ), C′ Z ( R ), and hence C ( R ), depend only on finitely many S –seminorms of R (and of b R ). A compact set K⊂S0 is bounded in each of these seminorms, so supR∈KC ( R ) <∞ . Taking C ( K ) := supR∈KC ( R )gives (5.174). □ Clay–compliance and role in the Lyapunov cascade. All appearances of R and ϖT occur solely as admissible Schwartz tests in explicit– formula pairings; neither ζ nor ξ is ever modified or evolved. The Gaussian window is inserted at the linear explicit–formula stage and only then is a mean–square taken, so no global L2 hypothesis in t is used. The constants in (5.125) and (5.174) are completely explicit up to finitely many Schwartz seminorms of R , and the T –dependence is fully accounted for by the factor 1 + log3(3+T). For a compact kernel path K = {Rτ : τ∈ [0 , τ∗ ] } as in §5.9, Corollary 5.3 furnishes a global envelope for the windowed energies ZR ERτ(t)ϖT(t)dt =OK1 + log3(3+T)(0 ≤τ≤τ∗),(5.175) with at most polylogarithmic growth in T . Combined with the local |t−γ|−1 cusp forced by any off–line zero (§5.5, §5.6), which makes the windowed integrals diverge for every T > 0as soon as R ( β ) > 0 at an off–line zero, this global envelope is incompatible with the cusp regime (ii) in Theorem 1, and thus forms one half of the Lyapunov–cusp contradiction in the Lyapunov dynamic cascade framework. 68 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) 5.8. Limiting procedures and regularity envelopes. We record here the limiting statements used in passing to the regularity limits required for (5.17). All steps are quantitative and rest solely on: (i) the T –controlled explicit–formula envelope of Proposition 5.6, with explicit profile 1 + log3(3+T), (ii) basic L2 –continuity of convolution and multiplier operators in the frequency variable, (iii) dominated convergence under the Gaussian window. Quantifier banner. Fix an admissible kernel R∈S0 once and for all. For t∈Rdefine the weighted frame coefficients F(t, ν) := ⟨ft,Φν⟩L2 σ, ft(σ) := ∂xg(σ, t),Φν(σ) := pR(σ)e−2πiνσ, (5.176) where g ( x, t ) = log |ξ ( x + it ) |2 and the L2 space is taken in the σ variable. The weighted Parseval identity (cf. Lemma D.1) gives, for a.e. t∈R, ZR|F(t, ν)|2dν =ZR R(σ)|ft(σ)|2dσ =ER(t),(5.177) where ER ( t )is the horizontal energy defined in (5.189) . All bounds below are uniform in the auxiliary parameters: the time window T > 0, a frequency multiplier scale λ∈ (0 , 1], and the convolution kernels Kη appearing in (5.163) . Implicit constants depend only on finitely many Schwartz seminorms of R and are therefore uniform over compact families K⊂S0. Admissible multipliers and kernels. We employ two standard families acting in the frequency variable ν: (M) Cutoff multipliers Mλ with |Mλ ( ν ) | ≤ 1for all ν and all λ∈(0,1], and Mλ(ν)−→ 1for each fixed ν(λ↓0).(5.178) (K) Approximate identities Kη∈L1 ( R )with ∥Kη∥1≤CK for all η∈(0,1], and Kη−→ δ0in S′(R) (η↓0).(5.179) Define the regularised frequency coefficients Fλ,η(t, ν) := (Kη∗(MλF(t, ·)))(ν).(5.180) Lemma 5.11 (Frequency envelopes preserve L2 ).For all λ, η ∈ (0 , 1] and a.e. t∈R, ZR|Fλ,η(t, ν)|2dν ≤ ∥Kη∥2 1ZR|Mλ(ν)|2|F(t, ν)|2dν ≤C2 KER(t). (5.181) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 69 Proof. Young’s inequality gives ∥Kη∗G∥2≤ ∥Kη∥1∥G∥2 for G∈L2 ( R ). Apply this with G ( ν ) = Mλ ( ν ) F ( t, ν ), use |Mλ| ≤ 1, and the identity (5.177). □ Lemma 5.12 (Windowed envelope).Let ϖT be the Gaussian window used in Section 5.7. Then for all λ, η ∈(0,1] and all T > 0, ZRZR|Fλ,η(t, ν)|2dνϖT(t)dt ≤C2 KZR ER(t)ϖT(t)dt ≤C2 KC(R)1+log3(3+T), (5.182) where C(R)is the constant in Proposition 5.6. Proof. Integrate the inequality of Lemma 5.11 against ϖT and use Proposition 5.6. □ Lemma 5.13 (Passage to limits).Fix T > 0. If Mλ→ 1pointwise with |Mλ| ≤ 1, and Kη→δ0in S′, then for a.e. t∈R, lim λ↓0, η↓0ZR|Fλ,η(t, ν)−F(t, ν)|2dν = 0.(5.183) Moreover, lim λ↓0, η↓0ZRZR|Fλ,η(t, ν)−F(t, ν)|2dνϖT(t)dt = 0.(5.184) Proof. For each fixed t , the map G7→ MλG is a contraction on L2 ν , and G7→ Kη∗G is continuous on L2 ν with norm ≤CK . Hence Fλ,η ( t, · ) → F(t, ·)in L2 νfor a.e. t. For the weighted statement, observe that |Fλ,η(t, ν)−F(t, ν)|2≤2|Fλ,η(t, ν)|2+|F(t, ν)|2,(5.185) and Lemma 5.11, together with (5.177) and Proposition 5.6, provides, for each fixed T > 0, an integrable majorant for the inner ν –integral under ϖT ( t ) dt , uniform in λ, η ∈ (0 , 1]. Dominated convergence in ( t, ν ) then applies. □ Proposition 5.7 (Regularity limits).For admissible multipliers Mλ and smoothing kernels Kηas above, and for every fixed T > 0, lim λ↓0, η↓0ZR ZR|Fλ,η(t, ν)|2ϖT(t)dtdν =ZR ZR|F(t, ν)|2ϖT(t)dtdν. (5.186) In particular, any internal frequency–side regularisation by ( Mλ, Kη ) may be inserted and removed inside the EF–controlled windowed energy representation without changing the limit. 70 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Proof. By Lemma 5.13, for each fixed T > 0we have ZRZR|Fλ,η(t, ν)−F(t, ν)|2dνϖT(t)dt −→ 0(5.187) as λ, η ↓ 0. The integrand is nonnegative, so Tonelli’s theorem applies and justifies the interchange of the ν –integral with the limit. Lemma 5.12 ensures that all intermediate expressions are finite and uniformly bounded in λ, η for the chosen T.□ Remark 5.8 (On admissible windows).Only three properties of ϖT are used: (i) ϖT≥0and ZR ϖT(t)dt = 1; (ii) ϖTis even in t; (iii) the Fourier transform of ϖTis bounded uniformly in T. Any such window family (Fejér, Poisson, compactly supported smooth approximate units) would yield identical regularity statements with the same  1 + log3 (3 + T )  profile in the EF envelope, provided the explicit–formula bound of Proposition 5.6 is available for that window. In this paper we fix the Gaussian family to keep the contour analysis and Appendix E as transparent as possible. Summary. Proposition 5.7, together with the EF control of Section 5.7, provides the regularity envelopes needed to justify the limiting scheme (5.17) : frequency regularisations ( Mλ, Kη )may be inserted and removed inside windowed EF energies without affecting the Clay–level conclusions. The dependence of all constants on R is through finitely many Schwartz seminorms, so the same regularity envelopes hold uniformly over compact families K⊂S0. Clay–compliance. All weights in ( σ, ν, t )are Schwartz test functions. The time window ϖT is removed only after establishing the T –controlled bound of Proposition 5.6; no uniformity beyond the explicit  1+ log3 (3+ T )  factor is claimed. No pointwise–in– t or global L2 hypothesis is assumed; the EF–bound is an averaged L2 statement only, and all regularity limits are taken under this windowed, Clay–compliant control. 5.9. Lyapunov functional and contradiction at off–line zeros. Quantifier banner. Fix an admissible kernel R∈S0 that is strictly positive away from the critical line: R∈S0, R(x)>0for all x=1 2.(5.188) (For example, Rα ( x ) = ( x−1 2 ) 2e−α(x−1 2)2 satisfies these conditions.) All constants below depend only on finitely many S –seminorms of THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 71 R and are independent of the Gaussian time scale T > 0; no growth information in T is needed here beyond finiteness of the EF envelope for each fixed T. We study the weighted horizontal energy ER(t) = ZR R(x)|∂xg(x, t)|2dx, g(x, t) = log |ξ(x+it)|2.(5.189) With ϖT(t) = 1 √πT e−t2/T2, set wT(t) := e−t2/T2=√πT ϖT(t).(5.190) Basic properties and extended–real viewpoint. Since ER ( t ) = qR [ g ( ·, t )] ≥ 0, the function ER takes values in the extended nonnegative reals [0 ,∞ ]. For t with ξ ( 1 2 + it )  = 0, analyticity of ξ off its zero set and vertical– line bounds for ξ′/ξ (cf. Lemma 5.8) imply ∂xg ( ·, t ) ∈L2 loc ( R )and Gaussian–weighted integrability at infinity, hence ER ( t ) <∞ for such t . When t = γ is the ordinate of a zero ρ = β + iγ , the local model of Section 5.4 gives ∂xg(x, γ) = 2m x−β+O(1),(5.191) with m≥ 1the multiplicity. If R ( β ) > 0, the resulting u−2 singularity at u=x−βforces Z|x−β|≤ε R(x)|∂xg(x, γ)|2dx = +∞(5.192) for every ε > 0, so ER ( γ ) = + ∞ . Thus ER ( t )is finite for a.e. t , but has a nonintegrable spike at each zero ordinate where R(β)>0. Lemma 5.14 (Cylindrical lower bound; nonintegrable spike).Let ρ = β + iγ be a zero of ξ of multiplicity m≥ 1and suppose R ( β ) > 0. Then there exist ε∗∈ (0 , 1), δ∗∈ (0 , 1), and constants c1, c2, C > 0such that c1|t−γ|−1−C≤ FR,ε∗(β, t)≤c2|t−γ|−1+C, 0<|t−γ|< δ∗, (5.193) where the local (cylindrical) flux is FR,ε∗(x0, t) := Zx0+ε∗ x0−ε∗ R(x)|∂xg(x, t)|2dx. (5.194) Consequently FR,ε∗(β, ·)/∈L1 loc at t=γ, and for every T > 0, ZRFR,ε∗(β, t)ϖT(t)dt = +∞.(5.195) 72 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) In particular, ER(γ)=+∞,ZR ER(t)ϖT(t)dt = +∞for every T > 0.(5.196) Proof. This is precisely the neighbourhood–divergence statement established in Sections 5.5 and 5.6. The two–sided estimate (5.193) comes from inserting the local expansion ∂xg(x, t) = 2m(x−β) (x−β)2+ (t−γ)2+b(x, t),(5.197) freezing R ( x ) = R ( β ) + O ( |x−β| )on a small interval around β , and computing the principal kernel Zε −ε u2 (u2+a2)2du =π 2|a|+Oε(1) (a=t−γ→0),(5.198) with R ( β ) > 0and b controlled in C1 . The lower bound in (5.193) shows that FR,ε∗ ( β, · )has a nonintegrable |t−γ|−1 cusp at t = γ . Since ϖT ( γ ) > 0for every T > 0, the Gaussian–weighted integral in (5.195) diverges. Finally, FR,ε∗(β, t)≤ER(t)implies (5.196). □ Windowed EF–bound (recall). For each R∈S0, Proposition 5.6 gives ZR ER(t)ϖT(t)dt ≤C(R)1 + log3(3+T)(T > 0),(5.199) with C(R)independent of T. Equivalently, for wT(t) = e−t2/T 2, ZR ER(t)wT(t)dt ≤C(R)1+log3(3+T)ZR wT(t)dt =C(R)1+log3(3+T)√π T. (5.200) In particular, for every fixed T > 0the windowed integral RRER ( t ) ϖT ( t ) dt is finite. Remark 5.9 (Exceptional times and windowing).By Lemma G.3, the EF/Plancherel identity and the blockwise spectral bounds of Section 5.7 hold for almost every t∈R , and all windowed quantities RER ( t ) ϖT ( t ) dt ignore the measure–zero exceptional set on which ER ( t )=+ ∞ . In particular, the cusp times t = γ arising from off– line zeros with R ( β ) > 0belong to such a null set and do not enter the EF assembly at any stage. Thus the local spike of Lemma 5.14 and the global EF envelope (5.199) are logically compatible in their hypotheses, and it is their numerical incompatibility that drives the Lyapunov contradiction below. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 73 Remark 5.10 (Tonelli interpretation of the Lyapunov functional).For later use we stress that the windowed Lyapunov functional LR,T := ZR ER(t)ϖT(t)dt (5.201) is always interpreted as the Tonelli integral of the nonnegative integrand R(x)|∂xg(x, t)|2ϖT(t)over R2: LR,T =ZZR2 R(x)|∂xg(x, t)|2ϖT(t)dx dt. (5.202) No truncation in t or exclusion of neighbourhoods of zero ordinates is ever performed. In particular, the exceptional times at which ER ( t ) = + ∞ (such as t = γ for an off–line zero with R ( β ) > 0) lie in a set of Lebesgue measure zero and do not affect the value of any windowed integral. The windowed EF–bound (5.199) therefore applies to the same Tonelli integral that Lemma 5.14 forces to diverge under an off–line zero. The contradiction in Proposition 5.8 thus arises from incompatible bounds on the same extended-real double integral, not from any form of regularisation or a.e. restriction. Definition 5.1 (Static Lyapunov functional).For R∈S0 and T > 0 set LR,T := ZR ER(t)ϖT(t)dt ∈[0,∞].(5.203) Proposition 5.8 (Static Lyapunov contradiction: exclusion of off–line zeros).Fix R∈S0 that is strictly positive on R\{1 2} (for example, Rα ( x ) = ( x−1 2 ) 2e−α(x−1 2)2 ). If a nontrivial zero ρ = β + iγ with β = 1 2 exists, then LR,T = + ∞ for every T > 0. On the other hand, the windowed EF–bound (5.199) implies LR,T <∞ for every T > 0. This contradiction forces β = 1 2 for all nontrivial zeros, i.e. the Riemann Hypothesis holds. Proof. Since R ( x ) > 0for all x = 1 2 , any off–line zero ρ = β + iγ necessarily satisfies R(β)>0, and Lemma 5.14 yields LR,T =ZR ER(t)ϖT(t)dt = +∞(5.204) for every T > 0. This contradicts the EF–bound (5.199) , which asserts LR,T ≤C ( R )  1 + log3 (3 + T ) <∞ for all T > 0. Thus no off–line zero can exist, and every nontrivial zero lies on ℜs=1 2.□ Remark 5.11 (Order of regulators; Clay compliance).All weights R and windows ϖT are admissible Schwartz tests, inserted only inside L2 pairings. No modification or evolution of ζ or ξ is ever made. When 80 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) This matches the analytic expansion in Section 5.6 and visually confirms that the leading behaviour of the cylindrical flux is governed entirely by the universal kernel, with all dependence on R and on the remainder b(x, t)absorbed into the bounded Oε(1) term. Practical caveats. Reliable numerical evaluation of ξ and ξ′/ξ requires high–precision arithmetic, stable evaluation of Γ ′/ Γ, and careful truncation in any Riemann–Siegel or explicit–formula implementation. In particular: •restrict to moderate heights |t|for graphical displays; • avoid sampling extremely near zero ordinates γ when forming time averages, since the true cusp is nonintegrable; • use the model profile ∂xgmodel to illustrate the singular part, and treat plots built from the full ∂xgas heuristic only. None of the rigorous estimates or bounds in this paper depends on any numerical calculation. Summary. The numerical illustrations mirror the analytic picture: off– line zeros generate a 1 /|t−γ| cusp in the cylindrical flux, on–line zeros are neutralised by the quadratic vanishing of R at x = 1 2 , and Gaussian windowing produces averaged energies that behave in a manner consistent with the polylogarithmic T –profile supplied by the EF bound. These displays are optional and intuition–only; the proof of Theorem B, and hence of Theorem A, is completely analytic and self–contained. 6. Independent cross–checks (do not change the proof) This section records several classical consistency checks. None of them enters the proof of Theorem B; they merely verify that the Lyapunov/flux framework, the windowed explicit–formula decomposition, and the admissible kernel family R∈S0 sit harmoniously inside standard analytic number theory. We fix the Gaussian window ϖT(t) = 1 √π T e−t2/T 2, wT(t) := e−t2/T 2=√π T ϖT(t),(6.1) and recall that all regularisation limits are taken in the order T→ ∞ then α↓0(6.2) for the canonical family Rα(x) = x−1 22e−α(x−1 2)2,(6.3) as prescribed in Section 5.10 and used throughout the explicit–formula and Lyapunov analysis. In the present subsection the Gaussian window THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 81 itself does not appear; we work instead with compactly supported (up to Schwartz tails) cutoffs in the height variable t to recover the classical zero–counting asymptotics. 6.1. Riemann–von Mangoldt zero counting. We first check that the explicit–formula identities underlying our global energy functional reproduce, under smoothing/desmoothing, the classical Riemann–von Mangoldt formula N(T) := #{ρ: 0 <ℑρ≤T, 0<ℜρ < 1}=T 2πlog T 2πe +O(log T), (6.4) where the sum runs over nontrivial zeros ρ of ζ , counted with multiplicity. This is a consistency check only; it is not used in the proof of Theorem B. Smoothed counting window. Let ϕ∈S ( R )be even, nonnegative, with RRϕ = 1 and b ϕ≥ 0. For T > 1and ∆ ∈ (0 , 1], define the rescaled bump and smoothed cutoff ϕ∆(t) := ∆−1ϕ(t/∆), ψT,∆:= 1[0,T]∗ϕ∆.(6.5) Then ψT,∆∈S(R),0≤ψT,∆≤1, and 1[0,T]≤ψT,∆≤1[−∆, T+∆].(6.6) Let N∆(T) := X ρ ψT,∆(ℑρ),(6.7) where the sum runs over nontrivial zeros ρ = β + iγ , counted with multiplicity. Then N(T)≤N∆(T)≤N(T+∆)+O(1),(6.8) where the O (1) term accounts for endpoints and for the symmetry γ↔ −γacross the real axis. Zero block via the σ –integration. Within the explicit–formula decomposition of Section 5.7, pairing the zero sum Z(σ, t) := ⟨X⟩ρ 1 σ+it −ρ(6.9) against R∈S0 in the σ –variable produces an associated Schwartz kernel κR∈S(R)in t, defined by ZR[h] := ZZRZ(σ, t)R(σ)h(t)dσ dt =X ρ (κR∗h)(ℑρ), h ∈S(R). (6.10) 82 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) For the canonical family Rαone can normalise so that ZR κRα(t)dt = 1,(6.11) and the kernels satisfy the approximate–identity convergence κRα S′ −−−→ α↓0δ0.(6.12) Thus κRα acts as a (real, even) Schwartz approximate identity on the t–side, uniformly on compact α–ranges. For any Schwartz window h∈S ( R ), the zero block tested in t can be written as ZRα[h] := ZZRZ(σ, t)Rα(σ)h(t)dσ dt =X ρ (κRα∗h)(ℑρ).(6.13) In particular, with h=ψT,∆and using (6.11)–(6.12), ZRα[ψT,∆] = X ρ ψT,∆(ℑρ)+OR,ϕ,∆(1) = N∆(T)+OR,ϕ,∆(1),(6.14) uniformly for T≥ 2and ∆ ∈ (0 , 1]. The O (1) term comes from the uniform approximation κRα∗ψT,∆ = ψT,∆ + OR,ϕ,∆ (1), combined with the standard bound N(u; 1) ≪log(2 + |u|)on unit–band zero counts. Gamma/rational block: main term. For G(σ, t) := 1 s+1 s−1−1 2log π+1 2 Γ′ Γs 2, s =σ+it, (6.15) Stirling’s formula on vertical strips and dominated convergence allow us, after integration in σ against Rα , to replace σ by 1 2 up to a bounded error. This yields GRα[ψT,∆] := ZZRG(σ, t)Rα(σ)ψT,∆(t)dσ dt =1 2πZR ψT,∆(t) log |t| 2πdt+O(1). (6.16) A standard computation using (6.6) shows that, uniformly in ∆ ∈ (0 , 1] and in αon compact subsets of (0,∞), GRα[ψT,∆] = T 2πlog T 2πe +O(log T), T ≥2.(6.17) Prime (Dirichlet–Euler) block: lower order. For P(σ, t):=−X n≥2 Λ(n) nσ+it ,(6.18) testing against Rαin σyields coefficients aα(n) := −Λ(n) ΦRα(log n),(6.19) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 83 where Φ Rα is the Mellin shadow of Rα and belongs to S ( R )(cf. Section 5.7). In particular, for each M≥2, aα(n)≪M,Rα Λ(n) (1 + log n)M.(6.20) Since d ψT,∆(u) = sin(πuT) πu b ϕ(∆u),(6.21) with b ϕ rapidly decaying, a standard Dirichlet–polynomial estimate and partial summation give PRα[ψT,∆] := ZZRP(σ, t)Rα(σ)ψT,∆(t)dσ dt =OR,ϕ,∆(1), T ≥2. (6.22) Assembly and desmoothing. The (linearly) tested explicit formula for the Rα–weighted zero block reads ZRα[ψT,∆] = GRα[ψT,∆]+PRα[ψT,∆],(6.23) so combining (6.14), (6.17), and (6.22) gives N∆(T) = T 2πlog T 2πe +O(log T),(6.24) uniformly for T≥ 2and ∆ ∈ (0 , 1], and uniformly in α on compact subsets of (0 ,∞ ). Passing to the limit α↓ 0(using the S′ –convergence (6.12)) does not change the asymptotics. Finally, the bracketing (6.8) implies N(T) = T 2πlog T 2πe +O(log T),(6.25) which is precisely the classical Riemann–von Mangoldt formula (6.4). Remarks. (1) The smoothing/desmoothing uses only admissible test functions in σ and t and the same vertical–line bounds for ξ′/ξ already exploited in Section 5.7; no new hypothesis or distributional input on zeros enters the argument. (2) Replacing 1 [0,T] by 1 [T,T+H] with 1 ≤H≤T and repeating the same steps yields the usual smoothed short–interval zero density. (3) Clay–compliance is automatic: Rα , ϕ∆ , and ψT,∆ are all test functions on the explicit–formula side; the zeta function and its zero set are never modified. The limits T→ ∞ and α↓ 0are taken in the fixed order of Section 5.10 under uniform majorants, so all conclusions concern the unaltered ζ and its classical zero set. 84 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) 6.2. Li’s coefficients and positivity structure. Li’s criterion states that for n≥1, λn:= 1 (n−1)! dn dsnsn−1log ξ(s)s=1 =X ρ1−1−1 ρn,(6.26) where the sum runs over the nontrivial zeros ρ of ζ , counted with multiplicity, and that λn≥0∀n⇐⇒ RH.(6.27) We do not use Li’s criterion anywhere in the proof of Theorem B. The purpose of this subsection is purely interpretive: to explain how the sign architecture of our quadratic Lyapunov energy aligns naturally with the positivity structure of Li’s coefficients, while remaining logically independent from it. Quadratic positive weighting of zeros. For any admissible R∈S0 and T > 0, recall ER,T := ZR ER(t)ϖT(t)dt, ER(t) :=ZR R(x)|∂xg(x, t)|2dx, (6.28) where g ( x, t ) = log |ξ ( x + it ) |2 and ϖT ( t ) = ( √π T ) −1e−t2/T2 . Section 5.7 shows, via the windowed explicit formula applied linearly in ( x, t )and then paired in xagainst R, that ER,T can be decomposed as ER,T =X ρ WR(ρ;T)+MR(T),(6.29) where: • each WR ( ρ ; T ) ≥ 0may be taken to be the contribution of the energy integrand localised to a small neighbourhood of the ordinate γ = ℑρ in the t –variable (using a nonnegative partition of unity in t); and • the remainder MR ( T )collects the (archimedean and prime) Gamma/Dirichlet–Euler blocks together with the energy away from the zero ordinates, and satisfies |MR(T)| ≪R1 + log3(3+T),(6.30) in line with the global EF envelope of Section 5.7. Thus ER,T is a quadratic, nonnegative functional of the zero set: it is built from L2 energies of the universal 1 / ( x−β )slope profile of ∂xg , localised by Rand averaged in t. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 85 Alignment and distinction from Li’s criterion. Li’s coefficients λn form alinear functional on the multiset of zeros, with weights cn(ρ)=1−1−1 ρn.(6.31) Under RH one has ℜcn ( ρ ) ≥ 0for each ρ , hence λn = Pρcn ( ρ )has a positivity structure that characterises RH. By contrast, our quantities ER,T arise from quadratic measurements: ER,T =ZRZR R(x)|∂xg(x, t)|2dx ϖT(t)dt, (6.32) which induce nonnegative local weights WR ( ρ ; T ) ≥ 0in the sense of (6.29) after localisation in t . The philosophical alignment is that each zero contributes a nonnegative amount to the relevant functional under the regime compatible with RH (Li’s coefficients in the linear setting, ER,T in the quadratic Lyapunov setting). The methodological distinction is essential: (local cusp) ER(t)≍ |t−γ|−1 near any off–line zero ρ=β+iγ, (Sections 5.5 and 5.6), (global EF envelope) ZR ER(t)ϖT(t)dt ≤C(R)1 + log3(3+T)for all T > 0, (Section 5.7). No assumption or conclusion about Li’s coefficients is ever invoked. Thus Li’s criterion and our Lyapunov framework are compatible but logically disjoint: they both produce nonnegative zero–weights under RH, but the Lyapunov dynamic cascade proof is closed entirely within the explicit–formula/Lyapunov architecture. A canonical positive family. For the canonical Gaussian–quadratic kernels Rα(x) = (x−1 2)2e−α(x−1 2)2, α > 0,(6.33) the local representation of Section 5.4 may be written schematically as ∂xg(x, t) = X ρ m(ρ)2(x−β) (x−β)2+ (t−γ)2+b(x, t),(6.34) where m ( ρ )is the multiplicity and b is a bounded C1 remainder. Pairing in x against Rα and averaging in t against ϖT produces, after grouping the contributions near each ordinate γ using a nonnegative partition of unity in t, a representation of the form ERα,T =X ρ m(ρ)2(Kα∗ϖT)(ℑρ)+Rα(T),(6.35) 86 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) where Kα∈S ( R )is even and nonnegative, and Rα ( T )absorbs the cross–terms between distinct zeros together with the contribution of the remainder b(x, t)and satisfies |Rα(T)| ≪Rα1 + log3(3+T).(6.36) Each zero thus contributes a nonnegative amount m ( ρ ) 2 ( Kα∗ϖT )( ℑρ ), and as α↓ 0the kernels Kα form an approximate identity on the t –axis. In particular, ( Kα∗ϖT )behaves as a smoothed counting kernel in the sense of Section 6.1. These observations are purely interpretive: they are consistent with, but not used in, the proof spine. The rigorous bounds and the contradiction mechanism rely only on the neighbourhood divergence and on the uniform EF envelope. Interpretive summary. The Lyapunov/energy framework generates a flexible family of nonnegative quadratic functionals on the zero set through the weights WR ( ρ ; T ) ≥ 0in (6.29) . This positivity structure is compatible with the positivity of Li’s coefficients under RH, in the sense that both assign nonnegative contributions to individual zeros in their respective regimes. However, the logical route to Theorem B is entirely independent of Li’s criterion: the proof proceeds solely via the cascade local cusp at any off–line zero +global windowed EF bound =⇒contradiction. (6.37) within the Clay–compliant Lyapunov/explicit–formula framework developed in Sections 4 and 5. 7. Sensitivity, robustness, and kernel families Quantifier banner. Throughout this section we adopt the Fourier conventions of Sections 5 and 5.1 and the mass–one Gaussian ϖT(t) = 1 √π T e−t2/T 2, wT(t) = √π T ϖT(t).(7.1) All implicit constants depend only on finitely many Schwartz seminorms of the spatial kernel R and (when present) of the time window ω , and are independent of T ; any T –dependence in the bounds appears explicitly through a factor of the form  1 + log3 (3 + T )  . Whenever we speak of uniformity in R , it is with respect to bounded subsets of S ( R ). This ensures that all explicit–formula (EF) bounds and all neighbourhood– divergence (NDL) statements used in the proof spine remain valid THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 87 under perturbations of R and of the time windows, and along the Lyapunov–dynamic cascade τ7→ Rτ. 7.1. Admissible kernels and structural hypotheses. We work with two nested classes of spatial kernels: (A0)Minimal admissible class S0:= nR∈S(R) : R≥0, R(x)=R(1−x), R(1 2) = R′(1 2)=0, R′′(1 2)>0o. (7.2) Thus R is real, nonnegative, even about x = 1 2 , and vanishes quadratically at the critical line. This quadratic zero is precisely the structural feature that cancels the on–line pole in the EF analysis of Section 5.7 and Appendix E. (A+ 0)Positive–definite subclass. S+ 0:= R∈S0:b R(ξ)≥0for all ξ∈R,(7.3) where b R denotes the 2 π –Fourier transform in the x –variable, as fixed in §5.1. The Gaussian–polynomial kernels Rα(x) = (x−1 2)2e−α(x−1 2)2, α > 0,(7.4) belong to S0 . They have strictly positive low–frequency Fourier mass, but c Rα changes sign for large |ξ| , so typically Rα/∈S+ 0 . The proof spine was developed first for these model kernels and then lifted to general R∈S0 ; here we make the robustness of all EF and NDL statements explicit. Remark 7.1 (Fourier positivity not required).Fourier positivity of R is never used in the global EF–bound or in the NDL. Whenever frequency–diagonal positivity is needed, it is supplied by auxiliary convolution kernels K with b K≥ 0(cf. Section 3.5). All constants in the EF and NDL estimates depend only on finitely many seminorms of R and on the quadratic vanishing R ( 1 2 ) = R′ ( 1 2 )=0; no hypothesis of the form b R≥ 0is required. 7.2. Stability of the neighbourhood–divergence lemma (NDL). Recall that for a zero ρ = β + iγ of multiplicity m≥ 1, the cylindrical flux is FR,ε(β, t) := Zβ+ε β−ε R(x)|∂xg(x, t)|2dx, (7.5) 88 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) where g ( x, t ) = log |ξ ( x + it ) |2 . For the model kernels Rα , Lemma 5.9 and Section 5.6 show that if β=1 2then FRα,ε(β, t)≍ |t−γ|−1(t→γ),(7.6) with explicit constants depending only on m and on finitely many seminorms of Rα . We now extend this uniformly to every R∈S0. Proposition 7.1 (Robust NDL for general admissible kernels). Let R∈S0 . If ρ = β + iγ is a nontrivial zero of multiplicity m≥ 1with β = 1 2 , then for all sufficiently small ε > 0 there exist constants c1, c2, C > 0, depending only on m , finitely many seminorms of R , and the local C1 bounds of the analytic remainder bin Section 5.4, such that c1 R(β) |t−γ|−C≤FR,ε(β, t)≤c2 R(β) |t−γ|+C, 0<|t−γ|< δ0. (7.7) Hence FR,ε(β, ·)/∈L1 loc at t=γ, and ER(γ)=+∞. Sketch. By Lemma 5.7, ∂xg(x, t) = 2m(x−β) (x−β)2+ (t−γ)2+b(x, t),(7.8) with b∈C1 on a fixed bidisc around ( β, γ ). Write u = x−β and a=t−γ. Taylor expanding R(β+u)gives R(β+u)=R(β) + O(|u|)(7.9) for |u| ≤ ε , with constants depending on finitely many seminorms of R. The principal integral Zε −ε u2 (u2+a2)2du =π 2|a|+Oε(1) (7.10) yields the dominant contribution 2 πm2R ( β ) |a|−1 ; cross terms involving b and the linear error in R are O (1) by the C1 bounds on R and b and the fact that u/ ( u2 + a2 )is odd. This produces the two–sided estimate (7.7). □ Remark 7.2 (Dependence on the distance to the critical line). Since R∈S0 satisfies R ( 1 2 ) = R′ ( 1 2 )=0and R′′ ( 1 2 ) > 0, Taylor’s theorem shows R(β)≍ |β−1 2|2for βnear 1 2.(7.11) Thus the prefactor R ( β )in (7.7) has an explicit quadratic dependence on the horizontal distance of the zero from the critical line. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 89 Remark 7.3 (Stability under finite local clusters).If finitely many other zeros lie within |t−γ′| ≪ |t−γ| , their singular kernels contribute only bounded cross–terms to FR,ε ( β, t ), which are absorbed into the constant C by Cauchy–Schwarz. The |t−γ|−1 slope is universal and survives under finite clustering. 7.3. Stability of the explicit–formula energy bound (EF– bound). Proposition 5.6 established, for each fixed R∈S0 , the T–controlled EF–bound ZR ER(t)ϖT(t)dt ≤C(R)1 + log3(3+T),(7.12) with C ( R )depending only on finitely many seminorms of R and independent of T . We now show that the same type of bound holds uniformly for all admissible Schwartz windows, with constants depending only on finitely many seminorms of ( R, ω ). Proposition 7.2 (Robust EF–bound for kernels and windows). Let R∈S0 and let ω∈S ( R )be nonnegative with RRω = 1. Set ωT ( t ) = T−1ω ( t/T ). Then there exists C ( R, ω ) <∞ , depending only on finitely many seminorms of Rand ω, such that ZR ER(t)ωT(t)dt ≤C(R, ω)1 + log3(3+T)for all T > 0. (7.13) Sketch. Repeat the EF decomposition of Section 5.7 with ωT in place of ϖT , inserting the window at the linear stage and then taking a windowed mean square. (1) Gamma/rational block. On the strip |σ−1 2|≤δ , Stirling’s formula gives |Γ′/Γ((σ+it)/2)| ≪δ1 + log(2 + |t|).(7.14) Arguing exactly as in Section 5.7 but with ωT in place of ϖT yields BΓ[R;T] := ZZR R(σ)|G(σ, t)|2ωT(t)dσ dt ≤CΓ(R, ω)1+log2(3+T), (7.15) where CΓ ( R, ω )depends only on finitely many seminorms of (R, ω). The moment bound ZR log2(2+|t|)ωT(t)dt ≪ω1 + log2(3+T)(7.16) follows from the scaling t=Tu and the Schwartz decay of ω. 96 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) modified, regularised, evolved, or replaced by any surrogate object. We also fix the ambient measure–theoretic conventions, the admissible order of regulators, and the precise points at which classical bounds are invoked. Quantifier banner and ambient conventions. Fix the mass–one Gaussian window ϖT(t) := 1 √π T e−t2/T 2so wT(t) = e−t2/T 2=√π T ϖT(t),(8.1) and any admissible spatial kernel R∈S0=R∈S(R) : R≥0, R(1−x)=R(x), R(1 2) = R′(1 2) = 0, R′′(1 2)>0, (8.2) as in Sections 5.1 and 5.3. All implicit constants in this section depend only on finitely many S –seminorms of R (and, when relevant, of alternative windows ω ) and are independent of T and of any auxiliary smoothing parameter α . Any explicit T –dependence appears only through the polylogarithmic factor  1 + log3 (3 + T )  inherited from the windowed EF envelopes (Propositions 5.6 and 7.2). All x –integrals are taken with respect to R ( x ) dx , all t –integrals with respect to ϖT ( t ) dt (or dt when explicitly stated). Tonelli, Fubini, and dominated convergence are used only once explicit dominating envelopes have been produced—primarily: (i) vertical–line bounds for ξ′/ξ (see Lemmas B.5 and 5.8), and (ii) the windowed explicit–formula bounds (Proposition 5.6 and their robust variant Proposition 7.2), which provide control of the form ZR ER(t)ωT(t)dt ≪R,ω 1 + log3(3+T) (T > 0).(8.3) Statements “for almost every t ” refer to Lebesgue measure dt , hence also to ϖT ( t ) dt by absolute continuity. Evaluations “at t = γ ” are interpreted via truncation |t−γ|> η followed by η↓0, or via time–window limits under ϖT(cf. Section 5.10). The observable is always g(x, t) := log |ξ(x+it)|2, ∂xg(x, t) = 2 ℜξ′ ξ(x+it),(8.4) with functional–equation symmetry g ( x, t ) = g (1 −x, t )(cf. Sections 5.1 and 5.4). THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 97 8.1. Inventory of analytic devices (roles and limits). Time windows. The family {ϖT}T>0⊂S ( R )is used only to form windowed averages of nonnegative quantities, e.g. LR,T := ZR ER(t)ϖT(t)dt, ER(t) = ZR R(x)|∂xg(x, t)|2dx. (8.5) Windows never act on ζ or ξ as arguments; they merely average in t . No structural identity depends on any special property of Gaussians beyond being nonnegative, mass–one, and Schwartz. More generally, any nonnegative mass–one Schwartz window ωT may be used instead of ϖT ; the EF envelope is uniform in T up to the explicit factor  1 + log3 (3 + T )  for all such families (Proposition 7.2, Section 7.5). Spatial kernels. Kernels R∈S0appear only: (1) inside L2pairings in x; and (2) in the quadratic form qR [ h ] = RRR ( x ) |h′ ( x ) |2dx, see Section 5.3. They never enter as arguments of ζ or ξ , and never act as mollifiers of the zeta function. The quadratic vanishing at x = 1 2 removes the on–line pole of ∂xg ; evenness enforces the functional–equation symmetry; Schwartz regularity ensures all explicit–formula blocks (Gamma, Dirichlet–Euler, zero block) satisfy the required bounds with at most polylogarithmic growth in T . Robustness across the full class S0 (and compact kernel families) is established in Section 7. Operators HR (Friedrichs realisation) and the ERU symbol. The symbol HR (or “HC” in earlier drafts) denotes the nonnegative self–adjoint operator associated with qR: HRh=−(Rh′)′on its Friedrichs domain in L2(R),(8.6) cf. Proposition 5.2. This operator is never used to evolve ζ or ξ ; it functions only as a bookkeeping device for the weighted horizontal energy of the observable g ( x, t ). The mnemonic “ERU” designates the measurement triad (kernel R , time window, quadratic energy) and encodes no dynamics, spectral hypothesis, or replacement for the zero set. Every identity involving HR reduces to explicit L2 pairings and standard explicit–formula manipulations, all of which concern the unaltered zeta function. 98 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) 8.2. Allowed manipulations and prohibited practices. Permitted (used): – Classical analytic continuation of ζ and ξ and the functional equation for ξ; Stirling’s formula for Γ′/Γon vertical strips. – Pairings of ∂xg = 2 ℜ ( ξ′/ξ )with admissible spatial kernels R and time windows ϖT (or more general ωT ); Tonelli/Fubini applied with nonnegativity or under explicit dominating envelopes. – The Guinand–Weil explicit formula with Schwartz test functions (zero, prime, and archimedean blocks), together with the standard unit–band zero count N ( u ; 1) ≪log (2 + |u| ); cf. Sections E and 5.7. – Closed quadratic forms, the Friedrichs extension, KLMN perturbation theory, and monotone/strong resolvent convergence for qR; cf. Sections 5.3 and 7. Prohibited (not used): – Any deformation, substitution, or redefinition of ζ or ξ ; any spectral proxy whose eigenvalues “model” zeros. – Any unproved hypotheses: pair–correlation, spacing conjectures, density estimates, GRH, or any conditional input. – Any unnormalised limit interchange lacking a priori T – controlled bounds and explicit domination (the log–cube profile 1 + log3(3+T)is always explicit). – Dependence on a special kernel (e.g. a particular Gaussian); any knife–edge constants that fail under S–perturbations. 8.3. Regulator architecture and order of limits. We employ two benign regulator families: mass–one Gaussian windows {ϖT}T>0 and, for form–theoretic localisation only, the model kernels Rα(x) = (x−1 2)2e−α(x−1 2)2, α > 0.(8.7) All regulator limits are taken in the fixed order (i) Fix R∈S0and let T→ ∞ (or use the T–controlled identities). (ii) Then, if invoked, let α↓0. (8.8) The EF–bound is  1 + log3 (3 + T )  –controlled in T for all nonnegative mass–one Schwartz windows (Proposition 7.2); the limit α↓ 0follows from Plancherel/dominated convergence THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 99 and monotone/strong resolvent convergence of closed forms (Lemma 5.6, Proposition 7.3). No regulator ever alters the zero set of ξ. 8.4. Micro–lemmas certifying compliance (standalone verifications). Lemma 8.1 (Window removal does not alter zeros or signs). Let f≥ 0be measurable on R and let ϖT be mass–one Gaussians. If f∈L1 ( R ), then ZR f ( t ) ϖT ( t ) dt →ZR f ( t ) dt as T→ ∞ . If f ( t ) ≡ + ∞ on a set of positive Lebesgue measure (for example, in a neighbourhood where f ( t ) ≍ |t−γ|−1 ), then ZR f(t)ϖT(t)dt = +∞for every T > 0. Proof. Approximate–identity convergence for ϖT gives the first claim. For the second, if f = + ∞ a.e. on a set of positive measure, then Rf ϖT = + ∞ for every T > 0by monotone convergence. □ Lemma 8.2 (Kernel removal preserves pairings).If Rα→R in Sand hsatisfies R|h′|2∈L1(R), then ZR Rα(x)|h′(x)|2dx −→ ZR R(x)|h′(x)|2dx. (8.9) Hence ERα(t)→ER(t)for a.e. t, and ZR ERα(t)ϖT(t)dt −→ ZR ER(t)ϖT(t)dt (8.10) for each fixed T > 0. Proof. Since Rα→R in L1∩L∞ loc and R|h′|2∈L1 ( R ), we have Rα|h′|2→R|h′|2pointwise a.e. with |Rα(x)||h′(x)|2≤(|R(x)|+ 1) |h′(x)|2,(8.11) for all sufficiently large α , and the right–hand side is integrable. Dominated convergence applies. □ Lemma 8.3 (Vertical–line envelope).For σ∈[1 2−ϵ, 1 2+ϵ], ξ′ ξ(σ+it)≪ϵ1 + log(2 + |t|).(8.12) Thus, for xin this strip, R(x)|∂xg(x, t)|2≪R,ϵ 1 + log2(2+|t|),(8.13) 100 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) giving a T –controlled dominating envelope for all t –integrals against ϖT(and more general ωT), with ZR1 + log2(2+|t|)ωT(t)dt ≪ω1 + log2(3+T).(8.14) (See Lemma 5.8 and the Gamma/rational block of Section 5.7.) Lemma 8.4 (No special–kernel dependence).For every R∈S0 , the neighbourhood–divergence asymptotics FR,ε ( β, t ) ≍ |t−γ|−1 and the EF–bound ZR ER(t)ωT(t)dt ≪R,ω 1 + log3(3+T) (T > 0) (8.15) hold with constants depending only on finitely many seminorms of Rand ω. See Proposition 7.1 and Proposition 7.2. 8.5. Final contradiction re–expressed in compliance form. Proposition 8.1 (Contradiction at fixed window scale).Fix R∈S0 and any T > 0. If an off–line zero ρ = β + iγ exists, then the neighbourhood–divergence lemma (Sections 5.5 and 5.6) implies ZR ER(t)ϖT(t)dt = +∞,(8.16) while the explicit–formula bound (Section 5.7, Proposition 5.6 with ω=ϖ) implies ZR ER(t)ϖT(t)dt ≤C(R)1 + log3(3+T)<∞.(8.17) Contradiction. Regulator order checkpoint. The contradiction holds for each fixed T > 0. Whenever T→ ∞ is invoked, it occurs only after establishing T –controlled bounds with the explicit profile  1 + log3 (3 + T )  (Lemma 8.3, Section 5.7). If Rα is used for localisation, then α↓ 0is taken last (Lemma 5.6, Proposition 7.3). 8.6. Threat model (referee attack points) and responses. (1) “You modified ζ with weights.” Response: No. Weights appear only in external L2 pairings; ζ and ξ enter solely via ∂xg= 2ℜ(ξ′/ξ). (2) “The Gaussian is special.” Response: Any nonnegative mass–one Schwartz window works (Proposition 7.2, Section 7.5); kernel robustness is established in Section 7. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 101 (3) “Hidden non-uniformity in T .” Response: Vertical–line envelopes supply T –controlled dominators with an explicit  1 + log3 (3 + T )  profile in all EF bounds (Lemma 8.3, Sections 5.7 and 7.3). (4) “You assume b R≥ 0.” Response: Not required. The EF analysis uses only R, b R∈S and the quadratic cancellation R(1 2)=R′(1 2) = 0; cf. Remark 7.1. (5) “Compactness or spectral proxies are smuggled in.” Response: No compactness in L2 ( R )is used (Proposition 5.4); no Hilbert–Pólya surrogate appears. (6) “Illicit limit interchanges.” Response: All interchanges occur under explicit dominators and T –controlled bounds with the log–cube profile (Lemma 8.3, Sections 5.7 and 7.3); regulator limits are ordered and justified (Lemma 5.6, Proposition 7.3). 9. Conclusion We conclude by assembling the logical spine of the argument and stating the result in its definitive analytic form. All objects are classical: g(x, t) = log |ξ(x+it)|2, ER(t) = ZR R(x)|∂xg(x, t)|2dx, (9.1) with R∈S0 an admissible kernel and ϖT a nonnegative mass– one Schwartz window. The proof rests on two complementary analytic pillars: – Local pillar (Neighbourhood–Divergence Lemma). If ρ = β + iγ is an off–line zero of ξ of multiplicity m≥ 1 and R(β)>0, then for every sufficiently small ε > 0, FR,ε(β, t) := Zβ+ε β−ε R(x)|∂xg(x, t)|2dx ≍ |t−γ|−1(t→γ),(9.2) with explicit constants depending only on m , finitely many S –seminorms of R , and local C1 –bounds on the analytic remainder (cf. Sections 5.5 and 5.6). In particular, for every fixed T > 0, ZR FR,ε(β, t)ϖT(t)dt = +∞.(9.3) – Global pillar (Windowed explicit–formula bound). The Guinand–Weil explicit formula, applied linearly with 102 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Schwartz tests and only then squared, yields a uniformly T–controlled Lyapunov–type envelope ZR ER(t)ϖT(t)dt ≤C(R)1 + log3(3+T)for all T > 0,(9.4) where C ( R )depends only on finitely many S –seminorms of R and is independent of T (see Section 5.7, Proposition 5.6 and the robust variant Proposition 7.2). The delicate component is the zero block, which is controlled in the windowed mean–square sense by the windowed zero–sum lemma (Theorem 2 in Appendix E, packaged for later use as Lemma A.12 in Appendix A); crucially, no pointwise decay in |ν|uniformly in tis asserted or needed. These two pillars cannot simultaneously hold if an off–line zero exists: the local pillar forces divergence of a windowed energy that the global pillar bounds (for the same fixed pair ( R, T )). This finite–time, time–windowed contradiction is the analytic core of the equivalence established in Theorem B, and of its finite–time Lyapunov–cascade implementation in Theorem 1. In particular, the implementation theorem implies the target statement Theorem A. Corollary 9.1 (RH via Lyapunov/explicit–formula contradiction).Let R∈S0 be admissible and let ϖT be any nonnegative mass–one Schwartz window. Then for all T > 0, ZR ER(t)ϖT(t)dt ≤C(R)1 + log3(3+T).(9.5) If there existed a zero ρ = β + iγ with β = 1 2 and R ( β ) > 0, the neighbourhood divergence would imply ZR ER(t)ϖT(t)dt = +∞(9.6) for the same fixed pair ( R, T ), contradicting the global T –controlled bound. Hence every nontrivial zero ρ of ζ satisfies ℜρ = 1 2 . Thus Theorem A holds. Condensed proof. Fix R∈S0 and T > 0. The explicit–formula bound gives ZR ER(t)ϖT(t)dt ≤C(R)1 + log3(3+T)<∞.(9.7) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 103 If an off–line zero ρ = β + iγ existed with R ( β ) > 0, then Sections 5.5 and 5.6 yield FR,ε(β, t)≍ |t−γ|−1, and hence ZR FR,ε(β, t)ϖT(t)dt = +∞.(9.8) Since ER ( t ) ≥FR,ε ( β, t ), this forces ZR ER ( t ) ϖT ( t ) dt = + ∞ , contradicting the EF bound. Thus no off–line zero exists and Theorem A holds. □ Analytic inputs. The argument uses only classical tools: analytic continuation and the functional equation for ξ ; Stirling’s formula for Γ ′/ Γon vertical strips; the Guinand–Weil explicit formula with Schwartz tests (archimedean, prime, and zero blocks); and the unconditional unit–band zero count N ( u ; 1) ≪log (2 + |u| ). On the zero side, the only required control is the windowed mean–square estimate of Appendix E (Theorem 2), which yields large– |ν| decay and an integrable low–frequency envelope as ν→ 0(and explicit (1 + log3 (3 + T )) growth in the window scale). No deformation of ζ or ξ occurs, and all kernels R∈S0 and windows ϖT serve solely as admissible test weights in L2 pairings, removed in the disciplined order certified in Sections 5.10, 7 and 8. The contradiction is obtained at each fixed T > 0; the time window is a measurement device, not a limiting hypothesis. Scientific posture. For each admissible R and each fixed T > 0 exactly one of the following regimes can hold: (global EF regime) :ZR ER(t)ϖT(t)dt < ∞,(9.9) or (local cusp regime) :ZR ER(t)ϖT(t)dt = +∞.(9.10) The global EF regime is enforced unconditionally by the explicit– formula analysis, whereas the local cusp regime would follow from the existence of an off–line zero with R ( β ) > 0. These regimes are mutually exclusive, and no unproved hypothesis mediates between them. Robustness under kernel and window perturbations is guaranteed by Section 7. Outlook. The method exemplifies a general paradigm: use only admissible test weights; extract global L2 control from explicit 104 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) identities; and locate a universal local obstruction whose presence forces divergence at fixed window scale. In the Lyapunov formulation encoded in Theorem B and implemented via the finite–time cascade in Theorem 1, the time–windowed contradiction propagates across compact kernel paths {Rτ : τ∈ [0 , τ∗ ] } and identifies x = 1 2 as the unique stationary cut of the time– averaged flux. Final statement. Within the classical analytic framework of ζ and its completion ξ , and under the explicit classical inputs cited above, the only configuration compatible with the Lyapunov/explicit–formula identities is that all nontrivial zeros lie on the critical line. Thus Theorems B and 1 hold, and therefore Theorem A (the Riemann Hypothesis) follows. ■(9.11) Acknowledgements and Declaration of Interests. The author thanks colleagues and early readers for comments on preliminary drafts, and is grateful to the broader analytic number theory community whose work underpins this manuscript. In particular, the classical developments of Riemann, Hadamard, de la Vallée Poussin, von Mangoldt, Landau, Titchmarsh, Hardy, Littlewood, Ingham, Selberg, and Weil form the foundation on which this work is built. The Clay Mathematics Institute is acknowledged for articulating the problem with a standard of clarity and rigor that has guided the present analysis. The author gratefully acknowledges the financial support of a private individual, Mr Raymond Williams, whose contributions enabled the completion of this manuscript. The author declares no conflicts of interest. Declaration of generative AI and AI-assisted technologies in the manuscript preparation process During the preparation of this work the author used OpenAI’s ChatGPT (GPT-5.1 and 5.2 Pro) to assist with L A T E X formatting, language polishing, and consistency checks for notation and referencing, as well as occasional suggestions for exposition and organisation. The tool did not replace mathematical reasoning or proof-writing: all mathematical statements, derivations, and proofs reflect the author’s own arguments. After using the tool, THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 105 the author reviewed and edited all content as needed and takes full responsibility for the scientific accuracy, originality, and integrity of the published article. Policy compliance note. This disclosure is intended to satisfy Elsevier’s policy on the responsible use of AI tools. The assistance was limited to editorial and expository support and did not substitute for human judgement in the development or verification of the mathematical content. Author’s Note (context and provenance) This manuscript grew out of an editorial suggestion, made in discussion of a broader programme (“Kairos Codex”), to isolate and prove one concrete claim to full classical standards. The present work does exactly that: it formulates a measurementonly framework in which a local divergence mechanism and a global explicit–formula bound are shown to be incompatible with off-line zeros. The author does not hold an academic appointment in analytic number theory. The work, however, is entirely classical in its ingredients and is presented so that every step can be audited with standard tools. Computational assistants were used for document preparation and routine consistency checks; they did not substitute for human reasoning, and no claim relies on outputs that cannot be verified by hand. The manuscript is offered for rigorous peer review. Its acceptance or rejection should turn solely on the correctness and clarity of the argument presented here. References [1] E. C. Titchmarsh, The Theory of the Riemann Zeta-Function, 2nd ed., revised by D. R. Heath-Brown, Oxford Univ. Press, 1986. [2] A. Ivić, The Riemann Zeta-Function: Theory and Applications, Dover, 2003. [3] A. E. Ingham, The Distribution of Prime Numbers, Cambridge Tracts in Mathematics 30, Cambridge Univ. Press, 1932. [4] H. L. Montgomery and R. C. Vaughan, Multiplicative Number Theory I: Classical Theory, Cambridge Studies in Advanced Mathematics 97, Cambridge Univ. Press, 2007. [5] A. Selberg, Contributions to the theory of the Riemann zeta-function, Arch. Math. Naturvid. 48 (1946), 89–155. 112 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Consequently, by monotone convergence of forms as ε↓0, ZI|h′(x)|2dx ≤cIqR[h]+∥h∥2 L2(I)for all h∈ D(qR).(A.23) Proof. Fix I⋐R . Since R∈S0 , there exists a small interval J∋x0,J⋐I, on which cJ|x−x0|2≤R(x)≤CJ|x−x0|2(x∈J)(A.24) for some positive constants cJ, CJ ; on the compact set I\J , continuity and nonnegativity of Rimply R(x)≥mI>0. Step 1: control on I\J .On I\J , RI,ε ( x ) ≥R ( x ) ≥mI> 0, so for all h∈ C∞ c(R), ZI\J|h′|2dx ≤m−1 IZI\J RI,ε(x)|h′|2dx ≤m−1 IqRI,ε [h],(A.25) and the same inequality extends to h∈ D(qRI,ε )by density. Step 2: control on J .On J , we exploit the quadratic structure. The one–dimensional Hardy–Poincaré inequality (applied separately on each side of x0and recombined) yields ZJ|h−⟨h⟩J|2dx ≤CHP ZJ|x−x0|2|h′(x)|2dx, (A.26) where ⟨h⟩J:= |J|−1RJh. Using R(x)≥cJ|x−x0|2on J, ZJ|x−x0|2|h′(x)|2dx ≤c−1 JZJ R(x)|h′(x)|2dx ≤c−1 JqRI,ε [h].(A.27) We also have the trivial bound RJ|⟨h⟩J|2dx ≤ |J|−1∥h∥2 L2(J). Combining these yields ZJ|h|2dx ≤2ZJ|h−⟨h⟩J|2dx+2 ZJ|⟨h⟩J|2dx ≤C′ JqRI,ε [h]+∥h∥2 L2(J), (A.28) with C′ Jindependent of ε. A standard scaling argument on the bounded interval J now gives ZJ|h′|2dx ≤C′′ JqRI,ε [h]+∥h∥2 L2(J),(A.29) where C′′ J depends only on J and finitely many S –seminorms of R, but not on ε∈(0,1]. Step 3: combine. Adding the bounds on I\J and on J gives ZI|h′|2dx ≤cIqRI,ε [h]+∥h∥2 L2(I)(A.30) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 113 for some cI> 0depending only on I and finitely many S –seminorms of R, and independent of ε. For the ε↓ 0statement, note that RI,ε ↓R pointwise and qRI,ε [ h ] ↓qR [ h ]for each fixed h . The monotone convergence theorem for closed forms (Kato [21, Thm. VIII.3.11]) then yields the same inequality with qRin place of qRI,ε .□ Use in the main text. This local coercivity controls localisation errors in Sections 5.7 and 5.9, and justifies distributional manipulations near the quadratic degeneracy at x0in the finite–time Lyapunov framework. Proposition A.2 (KLMN perturbations).Let V be a (possibly indefinite) form perturbation on D ( qR )with |V [ h ] | ≤ a qR [ h ] + b∥h∥2 L2 for some a < 1, b≥ 0. Then qR + V is closed and semibounded on D(qR), and its Friedrichs operator is self–adjoint. Proof. This is the KLMN theorem (Kato [21, Thm. X.17], Reed–Simon [19, Thm. X.12]). □ Use in the main text. This ensures stability of the EF decomposition and harmless lower–order corrections in Section 5.7, including those arising from decompositions of g ( ·, t )into regular and singular parts. Proposition A.3 (Form convergence & strong resolvent limit). If Rn, R ∈S0 with Rn→R in S ( R ), then qRn→qR in the sense of Mosco. Consequently, the associated self–adjoint operators HRn converge to HR in the strong resolvent sense, and e−tHRn→e−tHRstrongly on L2(R)for each t≥0.(A.31) Proof (Mosco). We recall the two Mosco conditions for forms qRnon L2(R): (M1) If hn⇀ h weakly in L2 and supnqRn [ hn ] <∞ , then qR [ h ] ≤ lim infn→∞ qRn[hn]. (M2) For every h∈ D ( qR )there exist hn∈ D ( qRn )with hn→h in L2and lim supn→∞ qRn[hn]≤qR[h]. (M1: liminf). Suppose hn⇀ h in L2 ( R )and supnqRn [ hn ] < ∞ . Then {R1/2 nh′ n} is bounded in L2 ( R ), so (up to a subsequence) R1/2 nh′ n⇀ w in L2 . Since hn⇀ h in L2 , we have for every ϕ∈ C∞ c(R),ZR hnϕ′dx →ZR h ϕ′dx. (A.32) Define the distributional derivative h′ by ⟨h′, ϕ⟩ := −RRh ϕ′dx . 114 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) To identify wwith R1/2h′, let ψ∈ C∞ c(R). Then ⟨w, ψ⟩= lim n→∞ZR R1/2 nh′ nψ dx = lim n→∞ZR h′ nR1/2 nψ dx =−lim n→∞ZR hn(R1/2 nψ)′dx. (A.33) Since R1/2 n→R1/2 in S ( R ), we have ( R1/2 nψ ) ′→ ( R1/2ψ ) ′ in L2, and {hn}is bounded in L2by weak convergence. Hence −lim n→∞ZR hn(R1/2 nψ)′dx =−ZR h(R1/2ψ)′dx =⟨h′, R1/2ψ⟩=⟨R1/2h′, ψ⟩. (A.34) Thus w = R1/2h′∈L2 ( R ), and h∈ D ( qR ). By weak lower semicontinuity of the L2–norm, qR[h] = ∥R1/2h′∥2 L2=∥w∥2 L2≤lim inf n→∞ ∥R1/2 nh′ n∥2 L2= lim inf n→∞ qRn[hn]. (A.35) (M2: limsup). Let h∈ D ( qR ). By Lemma A.2 there exist h(k)∈ C∞ c ( R )with ∥h(k)−h∥qR→ 0as k→ ∞ . Fix k . Since Rn→R in S ( R )and h(k) is smooth with compact support, dominated convergence gives qRn[h(k)] = ZR Rn|h(k)′|2dx −→ ZR R|h(k)′|2dx =qR[h(k)] (n→ ∞). (A.36) For each k pick nk so that |qRnk [ h(k) ] −qR [ h(k) ] | ≤ 2 −k . Define a sequence {um} by setting um = h(k) whenever nk≤m < nk+1 . Then um→hin L2(because h(k)→hin ∥·∥qR) and lim sup m→∞ qRm[um]≤lim sup k→∞ qRnk[h(k)]≤lim sup k→∞ qR[h(k)]+2−k=qR[h]. (A.37) This verifies (M2). By Kato’s Mosco–convergence theorem (Kato [21, Thm. VIII.3.11 & Cor. VIII.3.12], Reed–Simon [19, Thm. VIII.25]), (M1)–(M2) imply strong resolvent convergence HRn→HR and strong convergence of the associated semigroups. □ Use in the main text. This justifies the robustness claims in Section 7 (e.g. passage Rα→R after proving T –controlled bounds), and underpins the “measure, not modify” regulator removal in the Lyapunov/EF framework: the EF–bank bound is proved for regularised kernels and then passed to the limit. A.4. Auxiliary comparisons, compactness, and local identities. We collect routine but repeatedly used facts. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 115 Lemma A.7 (Form monotonicity and domains).If 0 ≤R1≤ R2 a.e., then for all h∈ C∞ c ( R ), qR1 [ h ] ≤qR2 [ h ]. Moreover, D(qR2)⊂ D(qR1)and qR1[h]≤qR2[h]for all h∈ D(qR2). Proof. The inequality on C∞ c is immediate from the definitions. The domain inclusion and extension of the inequality follow by completion and Fatou’s lemma. □ Lemma A.8 (Local Poincaré–Hardy control).Let I⋐R and let ψ∈ C∞ c(I)be a cutoff. Then for all h∈ D(qR), ZI|h−⟨h⟩I|2dx ≲ZI|x−x0|2|h′(x)|2dx ≲qR[ψh]+∥h∥2 L2(I),(A.38) where ⟨h⟩I := |I|−1RIh and the implicit constants depend on I and Ronly through finitely many S–seminorms. Proof. The first inequality is the 1D Hardy–Poincaré inequality, applied on each side of x0 and recombined. For the second, note that on I we have R ( x ) ∼κ|x−x0|2 and ψ≡ 1on a slightly smaller subinterval, so ZI|x−x0|2|h′(x)|2dx ≲ZR R(x)|(ψh)′(x)|2dx+∥h∥2 L2(I)=qR[ψh]+∥h∥2 L2(I), (A.39) using Lemma A.6 to control the term involving ψ′h.□ Lemma A.9 (Caccioppoli–type estimate).Let h∈ D ( HR )solve HRh = f∈L2 loc ( R )in the weak sense. Then for any η∈ C∞ c ( R ), ZR η2R|h′|2dx ≲ZR R|(ηh)′|2dx+ZR (η′)2R|h|2dx ≲∥ηf∥2 L2+∥h∥2 L2(supp η), (A.40) where the last inequality uses qR [ ηh ] = ⟨f, η2h⟩ and Cauchy–Schwarz/Young. Proof. Expanding (ηh)′gives R|(ηh)′|2=Rη2|h′|2+ 2Rηη′ℜ(h′h)+R(η′)2|h|2.(A.41) Integrate and absorb the cross term using Young’s inequality to obtain the first inequality. For the second, use the weak formulation ⟨HRh, η2h⟩ = qR [ h, η2h ] = ⟨f, η2h⟩ , then apply Cauchy–Schwarz and Young, and bound ∥ηh∥L2 by ∥h∥L2(supp η) . □ Lemma A.10 (Rellich compactness on compacts).If I⋐R , the embedding h∈ D(qR) : ∥h∥2 L2+qR[h]≤1,→L2(I)(A.42) is compact. 116 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Proof. By Lemma A.6, the set {h : ∥h∥2 L2 + qR [ h ] ≤ 1 } has uniformly bounded RI|h′|2dx , hence is bounded in H1 ( I ). Rellich–Kondrachov on the bounded interval I implies that the embedding into L2(I)is compact. □ A.5. Characterisation of the operator domain. Proposition A.4 (Weak/strong domain characterisation).Let HRbe the Friedrichs operator associated with qR. Then D(HR) = nh∈H1 R(R) : ∃f∈L2(R)with ZR Rh′ϕ′dx =ZR f ϕ dx ∀ϕ∈H1 R(R)o. (A.43) For h∈ D ( HR ), Rh′∈H1 loc ( R )with ( Rh′ ) ′ = −f∈L2 ( R ), and the flux is continuous at x0 (Lemma A.5). Conversely, any h∈L2 with h∈H1 R ( R ), Rh′∈H1 loc ( R )and ( Rh′ ) ′∈L2 ( R ) belongs to D(HR)with HRh=−(Rh′)′. Proof. The first characterisation is the standard variational definition of the operator associated to a closed form (Proposition A.1 and Kato [21, §VI.2]): given h∈ D ( qR ) = H1 R ( R ), we have h∈ D(HR)iff there exists f∈L2(R)such that qR[h, ϕ] = ⟨f, ϕ⟩L2for all ϕ∈ D(qR) = H1 R(R),(A.44) in which case HRh=f. If h∈ D ( HR ), Lemma A.4 shows that ( Rh′ ) ′ = −f∈L2 ( R ) in the distributional sense, hence Rh′∈H1 loc ( R ). Flux continuity at x0is Lemma A.5. Conversely, suppose h∈L2 ( R )satisfies h∈H1 R ( R ), Rh′∈ H1 loc(R)and (Rh′)′=−f∈L2(R). Then for any ϕ∈ C∞ c(R), ZR Rh′ϕ′dx =−ZR (Rh′)′ϕ dx =ZR f ϕ dx. (A.45) By density of C∞ c ( R )in H1 R ( R ), this identity extends to all ϕ∈H1 R(R), so h∈ D(HR)with HRh=f.□ A.6. Typical perturbations covered by KLMN. We record a convenient sufficient condition for the form–smallness hypothesis used in Section 5.7. Lemma A.11 (Local potentials are small after localisation). Let V∈L1 loc ( R ) + L∞ ( R ). Then for every compact I⋐R and δ > 0there is CI,δ such that for all h∈ D(qR), ZI V(x)|h(x)|2dx≤δ qR[h]+CI,δ ∥h∥2 L2(I).(A.46) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 117 Hence, by a partition of unity and Lemma A.6, any potential V with V−∈L1 loc ( R )sufficiently small on each piece and V+∈ L∞(R)is KLMN–admissible. Proof. Fix I⋐R and decompose V = V1 + V∞ with V1∈L1 loc ( R ), V∞∈L∞(R). On I, ZI V∞|h|2dx≤ ∥V∞∥L∞(I)∥h∥2 L2(I).(A.47) For V1, Hölder’s inequality gives ZI|V1||h|2dx ≤ ∥V1∥L1(I)∥h∥2 L∞(I).(A.48) Using the Sobolev embedding H1 ( I ) ,→L∞ ( I )and Lemma A.6, we may bound ∥h∥2 L∞(I) by CRI|h′|2 + ∥h∥2 L2(I) , which in turn is bounded by C′qR [ h ]+ ∥h∥2 L2(I) . Absorbing the qR [ h ]–term into δqR [ h ]by choosing the localisation scale (and hence ∥V1∥L1(I) ) sufficiently small, and adjusting the coefficient in front of ∥h∥2 L2(I) , yields ZI V|h|2dx≤δqR[h]+CI,δ∥h∥2 L2(I).(A.49) For the global KLMN statement, cover R by a finite–overlap partition of unity {ψj} subordinate to compact intervals Ij , apply the above estimate to each ψjh , and sum over j . The assumption that the negative part V− is small on each piece ensures that the resulting form perturbation has relative bound <1with respect to qR, so Proposition A.2 applies. □ Lemma A.12 (Windowed zero–sum envelope).Let R∈S0 and T > 0. With ϖT ( t ) = ( √π T ) −1e−t2/T2 and ZR ( ν, t )as in Appendix E, one has for all ν∈R, ZR|ZR(ν, t)|2ϖT(t)dt ≤CZ(R) (1+|ν|)21+log3(3+T)1+log2 2+|ν|−1, (A.50) with CZ ( R )depending on finitely many S –seminorms of R (and on the fixed cutoff ϑ ) and independent of T . For ν = 0 we interpret log (2 + |ν|−1 )as 0. If R ranges over a compact K⊂S0 , the same bound holds with a constant CZ ( K )uniform in R∈K. In particular, ZRZR|ZR(ν, t)|2ϖT(t)dt dν ≤C′ Z(R)1 + log3(3+T),(A.51) 118 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) for some C′ Z ( R ) <∞ depending only on R (and ϑ ) but independent of T, since RR(1+|ν|)−21 + log2(2+|ν|−1)dν < ∞. Proof. The first displayed inequality is exactly the conclusion of Theorem 2 in Appendix E, with the same convention at ν= 0. For the integrated bound, note that the integrand is nonnegative, so Tonelli’s theorem applies. Integrating the bound in ν gives ZRZR|ZR(ν, t)|2ϖT(t)dt dν ≤CZ(R)1+log3(3+T)ZR 1 + log2(2+|ν|−1) (1+|ν|)2dν. (A.52) The ν –integral is finite: split into |ν| ≥ 1and |ν| ≤ 1. For |ν| ≥ 1 one has log (2 + |ν|−1 ) ≤log 3, so the integrand is ≪ (1 + |ν| ) −2 , which is integrable. For |ν| ≤ 1the denominator is ≍ 1and the integrand is ≪ 1 + log2 (2 + |ν|−1 ); the latter is integrable near 0 since R1 0log2 (2 + ν−1 ) dν < ∞ (e.g. by the substitution ν = e−u , which yields an exponentially decaying weight e−u). Thus the ν –integral equals a finite absolute constant, and we may set C′ Z(R):=CZ(R)ZR 1 + log2(2+|ν|−1) (1+|ν|)2dν, (A.53) which is independent of T . If R ranges over a compact K⊂S0 , then CZ ( R )may be chosen uniformly (by Theorem 2), hence the same definition yields a uniform constant C′ Z(K).□ Remark A.2 (Domain of applicability).Lemma D.1 is a purely Fourier–in– x identity for the quadratic form qR and is used pointwise in t only... at those ordinates for which ER ( t ) = qR [ g ( ·, t )] is finite. In particular, no global L2 assumption is made on the map t7→ h ( ·, t ), and we never apply Plancherel in the t –variable. In the main text, the explicit–formula bounds for the windowed energy RRER ( t ) ϖT ( t ) dt are obtained by inserting the Gaussian window at the linear level in ξ′/ξ and then using Cauchy–Schwarz in ( σ, t ). The weighted Plancherel identity serves only as an x –side bookkeeping device for ER ( t )and is never invoked a priori in any regime where ER ( t )would be infinite. Connections back to the proof. – Lyapunov functional and kernel–side energy: ER ( t ) = qR [ g ( ·, t )] is well–posed on the closed form (Lemmas A.1, THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 119 A.2), and the semigroup e−sHR provides a canonical kernel–side evolution compatible with the finite–time Lyapunov framework. – EF linearisation and IBP: Weak identification HR = − ( Rh′ ) ′ , flux continuity, and core IBP (Lemmas A.4, A.3, A.5) justify moving derivatives off g even across the degeneracy (Section 5.7), in a way compatible with the windowed explicit–formula decomposition in Theorem B. – Local analysis near x = 1 2 :Regulator–stable coercivity and Poincaré–Hardy control (Lemmas A.6, A.8) are used in the neighbourhood–divergence analysis and to bound localisation errors (Section 5.9), uniformly in the time–window scale. – Perturbations and limits: KLMN stability (Proposition A.2) covers lower–order EF corrections; Mosco/strong resolvent convergence (Proposition A.3) implements the R –perturbation and Rα→R limits (Section 7), with strong semigroup convergence for the kernel–side evolutions. – Zero–block envelope: Lemma A.12 packages the windowed zero–sum bound from Appendix E in a form directly used in the EF energy bound, making explicit that the zero–block contribution has integrable low–frequency behaviour (via the harmless factor 1+ log2 (2+ |ν|−1 )), decays like (1+ |ν| ) −2 for large |ν| , and grows at most like 1 + log3 (3 + T )in the window scale, with constants depending only on finitely many S –seminorms of R (uniformly on compact kernel families). Bibliographic anchors. We invoke only standard results: Kato [21] (form representation, KLMN, monotone/Mosco convergence, resolvents/semigroups) and Reed–Simon [19] (complementary operator–theoretic statements). This completes Appendix A. 120 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Appendix B. Vertical–line envelopes and dominated convergence Aim. Record vertical–line bounds for Γ ′/ Γand ξ′/ξ on compact strips around the critical line, with explicit dependence on |t| ; give a local pole decomposition for ξ′/ξ near zeros; and state dominated–convergence/Fubini criteria for time–averaged pairings against the mass–one Gaussian window used throughout the paper. All statements align with the truncation convention at zero ordinates from Section 5.10, and are used only as fixed– T backstops for the finite–time Lyapunov/EF framework (Sections 5.7 and 5.9). They are formulated so that envelopes and dominated–convergence arguments remain stable under compact kernel families K⊂S0 and auxiliary parameters (e.g. α , τ ), with dependence only on finitely many Schwartz seminorms of R. Window convention. We work with the mass–one Gaussian from the main text, ϖT(t) := 1 √π T e−t2/T 2,ZR ϖT(t)dt = 1,(B.1) and note that the arguments below apply (with identical proofs) to any nonnegative mass–one window ωT ( t ) := T−1ω ( t/T ) with ω∈S ( R ); cf. Proposition 7.2. In particular, all dominated–convergence statements and Fubini/Tonelli interchanges in this appendix are window–robust. The genuinely T –uniform bounds used in the Lyapunov/explicit–formula analysis come from the blockwise EF estimates in Section 5.7, not from the crude envelopes recorded here. B.1. Uniform Stirling on compact strips. Lemma B.1 (Uniform Stirling).Fix ϵ∈ (0 ,1 2 ). For σ∈ [ 1 2− ϵ, 1 2+ϵ]and all t∈R, Γ′ Γσ+it 2≪ϵlog2+|t|,(B.2) with an implicit constant uniform in σon the strip. Proof. Let z = σ+it 2 . On any sector |arg z| ≤ π−δ the classical Stirling expansion gives log Γ( z )=( z−1 2 ) log z−z + Oδ (1) and hence Γ ′/ Γ( z ) = log z + Oδ (1 /|z| ). As σ ranges in a fixed compact interval about 1 2 , there exists δ ( ϵ ) ∈ (0 , π )so that z remains in such a sector for all t = 0. Then |log z|≍log (2 + |t| ) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 121 and |z|−1≪ 1. For bounded t ,Γ ′/ Γis smooth on compacta; enlarging the implicit constant covers all t.□ B.2. ξ′/ξ : vertical–line envelope and local pole decomposition. Recall ξ(s) = 1 2s(s−1) π−s/2Γ s 2ζ(s),ξ′ ξ(s) = 1 s+1 s−1−1 2log π+1 2 Γ′ Γs 2+ζ′ ζ(s). (B.3) Lemma B.2 (Vertical–line envelope for ξ′/ξ ).Fix ϵ∈ (0 ,1 2 ). For σ∈[1 2−ϵ, 1 2+ϵ]and all t∈Rwith ξ(σ+it)= 0, ξ′ ξ(σ+it)≪ϵ1 + log2+|t|.(B.4) Proof. The rational terms 1 /s and 1 / ( s− 1) are O (1) uniformly on the strip. By Lemma B.1, the gamma term is ≪ϵlog (2 + |t| ). On fixed strips around 1 2 , the classical bound ζ′/ζ ( σ + it ) = Oϵ ( log (2+ |t| )) holds at nonzeros (see Titchmarsh–Heath-Brown [1, Chs. III–IV] or Ivić [2, §8.2]). Summing the contributions gives the claim. □ Lemma B.3 (Local pole decomposition).Fix ϵ∈ (0 ,1 2 ). There exists Cϵ>0such that for σ∈[1 2−ϵ, 1 2+ϵ]and |t| ≥ 2, ξ′ ξ(σ+it) = X |γ−t|≤1 1 σ+it −ρ+Oϵ log(2 + |t|),(B.5) where the sum is over nontrivial zeros ρ = β + iγ of ζ , counted with multiplicity. Proof. Starting from the Hadamard product for ξ and differentiating log ξ , one obtains the usual partial fraction expansion for ζ′/ζ on fixed strips about 1 2(see [1, Ch. IV]): ζ′ ζ(s) = X |γ−t|≤1 1 s−ρ+Oϵlog(2 + |t|), s =σ+it, (B.6) valid for |t| ≥ 2and s away from zeros. Inserting this into the expression for ξ′/ξ and using Lemma B.1 for the gamma term, together with boundedness of 1 /s and 1 / ( s− 1) on the strip, shows that the additional contributions are Oϵ ( log (2+ |t| )), giving (B.5). □ Remark B.1 (A.E. interpretation and truncation).Both Lemma B.2 and (B.5) hold for a.e. t (with respect to dt ). At ordinates γ where σ + iγ is a zero of multiplicity m , the left-hand side is 128 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) For the L∞ bound, evaluate (C.8) at ξ = 0: b Rα (0) = √π1 2α−3/2 ; away from ξ = 0 the Gaussian factor decreases, so ∥b Rα∥∞≍ α−3/2. For the L1bound, use |b Rα(ξ)| ≪ α−3/2+α−5/2ξ2e−(πξ)2/α.(C.14) With the substitution u=πξ/√α, ZR|b Rα(ξ)|dξ ≪α−3/2√αZR (1+u2)e−u2du ≪α−1,(C.15) as claimed. □ Remark C.1 (No general Fourier–positivity).Even though Rα≥ 0and is even about x = 1 2 , the factor α 2− ( πξ ) 2 in (C.9) changes sign for |ξ|>1 πpα/2 . Thus b Rα is not nonnegative on R . This plays no role in the proof: the EF bounds in Section 5.7 and Appendix E require only that R, b R∈S ( R )with finitely many seminorms controlled, and that R ( 1 2 )vanishes to order 2. No global Fourier–positivity is used. Use in the main text. Formula (C.8) and Lemma C.2 provide explicit seminorm control for the model family Rα . They serve as a worked example of how EF–bank constants depend on finitely many Schwartz seminorms of a general R∈S0 , and as a robustness check in Section 7. When R ranges over a compact K⊂S0 , the seminorm bounds in Lemma C.2 are replaced in practice by K –dependent constants Cm,k ( K ), obtained by taking suprema of the relevant seminorms over K. C.4. Normalised distributional limit as α→ ∞ .The unnormalised family Rαhas mass ZR Rα(x)dx =ZR y2e−αy2dy =√π 2α−3/2.(C.16) The natural mass–one normalisation is e Rα(x) := 2 √πα3/2Rα(x) = 2 √πα3/2(x−1 2)2e−α(x−1 2)2.(C.17) Lemma C.3 (Approximate identity at x=1 2).As α→ ∞, e Rα=⇒δ1/2in S′(R),(C.18) i.e. for every ϕ∈S(R), ZRe Rα(x)ϕ(x)dx −→ ϕ(1 2).(C.19) THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 129 Proof. With y=x−1 2and dx =dy, ZRe Rα(x)ϕ(x)dx =2 √πα3/2ZR y2e−αy2ϕ 1 2+ydy. (C.20) Substitute z=√α y,dy =dz/√α, to obtain ZRe Rαϕ=2 √πZR z2e−z2ϕ 1 2+z/√αdz. (C.21) The integrand is dominated by Cz2e−z2 , which is integrable on R. By dominated convergence, the limit is ϕ(1 2)·2 √πZR z2e−z2dz =ϕ(1 2)·2 √π·√π 2=ϕ(1 2).(C.22) □ Lemma C.4 (Frequency–side limit).With e Rαas in (C.17), c e Rα(ξ)−→ e−πiξ in S′(R)and pointwise for each fixed ξ∈R. (C.23) Proof. By Lemma C.3, e Rα⇒δ1/2 in S′ ( R ). For any φ∈S ( R ), ZRc e Rα(ξ)φ(ξ)dξ =ZRe Rα(x)bφ(x)dx −→ bφ(1/2).(C.24) Under our convention, the Fourier transform of δ1/2is d δ1/2(ξ) = ZR δ1/2(x)e−2πixξ dx =e−πiξ,(C.25) and Re−πiξφ(ξ)dξ =bφ(1/2). Thus c e Rα⇒e−πiξ in S′(R). Pointwise convergence follows directly from (C.8) : multiplying by 2 √πα3/2gives c e Rα(ξ)=e−πiξ1−2(πξ)2 αe−(πξ)2/α −−−→ α→∞ e−πiξ (C.26) for each fixed ξ∈R.□ Use in the main text. The normalised limit shows that, for large α , the spatial kernel e Rα acts as an approximate identity at x = 1 2 , while on the frequency side the weights approach the pure phase e−πiξ dictated by centering at x = 1 2 . For small α , the unnormalised transform b Rα is concentrated near ξ = 0 with width ≍√α , as seen from (C.8) . These facts are used only as robustness checks and do not enter the EF–bank/cusp contradiction directly. 130 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) C.5. Summary of what is actually used. – Conventions/Plancherel (Definition C.1 and Lemma C.1): used throughout EF assembly and in the Dirichlet–Euler block via standard L2 –Plancherel. The weighted Plancherel identity for qR itself is recorded in Appendix D (Lemma D.1), while Appendix A records the domain-of-applicability caveat (Remark A.2); this identity is applied only pointwise in t. – Translation–evenness identity (C.4) : records that for kernels R with R (1 −x ) = R ( x ), the transform has the form e−πiξ times a real even amplitude. This symmetry is sufficient for all zero–block arguments in the explicit formula; no global Fourier–positivity is imposed. – Explicit transform/seminorm bounds (equations (C.8) – (C.9) and Lemma C.2): give concrete seminorm estimates for the model family Rα , illustrating the dependence of EF–bank constants on finitely many Schwartz seminorms of a general R∈S0 , and how these constants can be made uniform over compact kernel families K⊂S0. – Normalised limit (Lemmas C.3–C.4): e Rα⇒δ1/2 as α→ ∞ in space, and c e Rα→e−πiξ in frequency. This underpins the “approximate identity” language in the Riemann–von Mangoldt cross–checks in the main text and in the kernel robustness discussion (Section 7). This completes Appendix C. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 131 Appendix D. Contour shifts, EF admissibility, weighted Parseval, and interchanges Aim. We justify the admissibility of even Schwartz test functions in the Guinand–Weil explicit formula (EF), make explicit the contour shifts (with residue accounting) used to pass from Euler products to Fourier expansions, record the weighted Plancherel/- Parseval identity for the Lyapunov form qR , and formalise when one may interchange t –integration (against the mass–one Gaussian ϖT) with sums/integrals in the EF decomposition. We distinguish (i) fixed– T dominated–convergence interchanges, and (ii) T –controlled interchanges, the latter resting on the EF block bounds proved in Appendix E with their explicit polylogarithmic dependence on T . All statements are independent of RH and use only standard analytic number theory and Fourier analysis, in a fully test–side, Clay–compliant way. D.1. Weighted Plancherel for the Lyapunov form. We first record the basic “weighted Parseval” identity which underlies the decomposition of the Lyapunov functional into EF frequency blocks. Recall from Appendix A that for R∈S0we put qR[h] := ZR R(x)|h′(x)|2dx, D(qR)=H1 R(R).(D.1) Lemma D.1 (Weighted Plancherel identity).Let R∈S0 . For each h∈ D(qR)define Ψh(x) := R(x)1/2h′(x),FR[h](ν) := c Ψh(ν) = ZR R(x)1/2h′(x)e−2πixν dx, (D.2) where the integral is understood in the L2 –Fourier sense if necessary. Then qR[h] = ZR R(x)|h′(x)|2dx =ZR|FR[h](ν)|2dν. (D.3) In particular, for any measurable family h ( ·, t ) ∈ D ( qR )(e.g. the Lyapunov profile g(·, t)from Section 5.9), one has ER(t) := qR[h(·, t)] = ZR|FR[h(·, t)](ν)|2dν for all twith ER(t)<∞. (D.4) 132 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Remark D.1 (Plancherel at the distributional level).The identity in Lemma D.1 is first established for h∈ D ( qR )with qR [ h ] <∞ , i.e. when √R h′∈L2 ( R ). In applications to the Lyapunov functional we work instead with the distributional derivative h ( ·, t ) = g ( ·, t ) = log |ξ ( · + it ) |2 and with its weighted derivative Ψ g(·,t) = R1/2∂xg ( ·, t ), which is a tempered distribution in x for every fixed t . Pairing against Schwartz tests in x and using the usual extension of Plancherel to tempered distributions shows that the map h7→ FR [ h ]is well defined at the linear level for all t , and Lemma D.1 applies as an energy identity for those t where ER(t)<∞. In particular, times t at which ER ( t ) = + ∞ (such as the cusp ordinates t = γ with R ( β ) > 0in Section 5.9) lie outside the form domain and do not enter the windowed EF identities: by Lemma G.3 the set of such t has measure zero and is ignored by the Gaussian window ϖT . The EF decomposition and the zero–block bounds are therefore obtained purely at the linear distributional level, and the spatial/spectral L2 identification is used only on the full–measure set where ER ( t ) <∞ . In particular, the global windowed energies RRER ( t ) ϖT ( t ) dt appearing in Proposition 5.6 and Section 5.9 are always understood as Tonelli integrals of the nonnegative density R ( x ) |∂xg ( x, t ) |2ϖT ( t )over R2 , so the EF/Plancherel machinery controls the same double integral whose divergence is detected by the local cusp analysis. Remark D.2 (Frequency profile for the Lyapunov functional). In Section 5.7 we take h ( ·, t ) = g ( ·, t ) = log |ξ ( · + it ) |2 and write BR(ν, t):=FR[g(·, t)](ν) = ZR R(x)1/2∂xg(x, t)e−2πixν dx. (D.5) Lemma D.1 then states simply that ER(t) = qR[g(·, t)] = ZR|BR(ν, t)|2dν. (D.6) The centring at x = 1 2 only introduces a fixed phase e−πiν (Appendix C, (C.4) ) and plays no role in the norm identity or in the finite–time Lyapunov/EF decomposition into frequency blocks. See also Remark A.2 for how this identity is used in the main text. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 133 D.2. EF admissibility for even Schwartz tests. Definition D.1 (Fourier conventions and seminorms).For φ∈ S(R)we use bφ(u) := ZR φ(x)e−2πiux dx, φ(x) = ZRbφ(u)e2πiux du. (D.7) Schwartz seminorms are ∥φ∥A,B := sup x∈R (1+|x|)A|φ(B)(x)|, A, B ∈N0.(D.8) Proposition D.1 (Guinand–Weil explicit formula for S tests). Let φ∈S(R)be even. Then X ρbφ ρ−1 2 i=bφ 1 2i+bφ −1 2i(D.9) −∞ X n=1 Λ(n) √nφ(log n) + φ(−log n)+1 2πZ∞ −∞ℜΓ′ Γ1 4+it 2bφ(t)dt, where ρ runs over the nontrivial zeros of ζ with multiplicity. The zero - sum and the gamma integral converge absolutely; the Dirichlet–Euler sum is defined by symmetric summation (or via (D.12) below) and, for tests with any exponential decay, is absolutely convergent. All implied constants in auxiliary bounds depend only on finitely many seminorms ∥φ∥A,B , independent of any time-window parameter T. Remark D.3 (On absolute vs. conditional convergence).For a general φ∈S , the series Pn Λ( n ) n−1/2φ ( log n )is interpreted either by (i) the Mellin representation (D.12) below, which is absolutely convergent on vertical lines, or (ii) symmetric summation. If φ decays exponentially (e.g. φ = φ0∗gη with Gaussian gη ), then the Dirichlet–Euler series is absolutely convergent termwise. In our applications the EF is always used blockwise after pairing with bφ and/or with the mass–one window ϖT , in which case absolute integrability is ensured; see Proposition D.2. Proof of Proposition D.1 (heat regularisation and continuity). We first prove (D.9) for a real - analytic, exponentially decaying approximation to φ , and then pass to φ by continuity of both sides as tempered distributions. Step 1 (heat regularisation). Fix η∈ (0 , 1] and let gη ( x ) := e−πη2x2 , φη := φ∗gη . Then cφη ( u ) = bφ ( u ) η−1e−πu2/η2 , whence 134 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) cφη extends to an entire function with rapid decay on every fixed vertical line: ∀v∈Rfixed,cφη(t+iv)≪A,v (1+|t|)−Aeπv2/η2.(D.10) Moreover, φη→φin Sas η↓0. Step 2 (Mellin–Fourier identity). For Φ η ( s ) := cφη s−1 2 i and c > 1, 1 2πi Z(c) n−sΦη(s)ds =φη(log n) √n.(D.11) This follows by writing out cφη and applying Fubini; the decay in (D.10) controls the interchange. Step 3 (prime block as a vertical integral). Define P(φη) := ∞ X n=1 Λ(n) √nφη(log n) = 1 2πi Z(2) −ζ′ ζ(s) Φη(s)ds, (D.12) using (D.11) and absolute convergence of the Dirichlet series for −ζ′/ζ on ℜs>1. Step 4 (contour shift and residues). For Y→ ∞ shift the rectangle bounded by ℜs = 2 and ℜs = − 1. Bounds cφη ( t + iv ) ≪ (1+ |t| ) −A on each vertical line (with v fixed) and −ζ′/ζ ( σ + it ) ≪ log (2 + |t| )on σ∈ [ − 1 , 2] imply that the horizontal integrals vanish. Picking residues at s= 1 and at s=ρyields P(φη) = X ρ Φη(ρ)−Φη(1) + 1 2πi Z(−1) ζ′ ζ(s) Φη(s)ds. (D.13) Step 5 (functional equation on the left edge). From ξ′ ξ(s) = 1 s+1 s−1−1 2log π+1 2ψs 2+ζ′ ζ(s), ξ(s) = ξ(1 −s), we obtain, on ℜs=−1, ζ′ ζ(s)=−ζ′ ζ(1 −s) + 1 2log π−1 2ψs 2−1 s−1 s−1.(D.14) Insert (D.14) into (D.13) , change variables u = 1 −s in the −ζ′/ζ (1 −s )term (moving back to ℜu = 2), and shift the remaining line to ℜs = 1 2 . The residues at s = 0 , 1give cφη ( ±1 2i ), and the gamma contribution becomes 1 2πZ∞ −∞ ℜΓ′ Γ1 4+it 2cφη(t)dt, THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 135 with absolute convergence by vertical–line Stirling and rapid decay of cφη. Thus (D.9) holds for φη. Step 6 (passage to φ by continuity). Both sides of (D.9) define continuous linear functionals of φ in the Schwartz topology: the spectral side converges absolutely by N ( T ) ≪Tlog T and bφ ( t ) = OA ((1 + |t| ) −A ); the gamma integral is continuous by vertical–line Stirling; the prime block is the vertical integral in (D.12) , which is linear and continuous on S by the bound −ζ′/ζ (2 + it ) ≪log (2 + |t| )and rapid decay of bφ ( t−i (2 −1 2 )). Since φη→φ in S and (D.9) holds for each η , the identity extends to φ by continuity. (Equivalently, one may argue by density of real-analytic Swithin S.) □ Remark D.4 (Absolute convergence of the spectral side).Because N ( T ) := # {ρ : 0 <ℑρ≤T} = T 2πlog T + O ( T )and bφ ( t ) = OA ((1+ |t| ) −A )for every A , the series Pρbφ (( ρ−1 2 ) /i )converges absolutely: for any A>2, X |ℑρ|≥1bφ ρ−1 2 i≪Z∞ 1 log T TAdT < ∞.(D.15) D.3. Contour shift and residue accounting (Dirichlet–Euler block). Fix an even φ∈S(R)and put Φ(s) := bφ s−1 2 i,so that Φ(1/2+it) = bφ(t).(D.16) As in (D.12), define P(φ) := ∞ X n=1 Λ(n) √nφ(log n) = 1 2πi Z(2) −ζ′ ζ(s) Φ(s)ds, (D.17) where the right-hand side is the definition in the Schwartz case (Remark D.3), and coincides with the series whenever that series is absolutely convergent. Lemma D.2 (Contour shift for (D.17) ).Let Y→ ∞ and shift the vertical contour from 2 −iY to 2 + iY left to − 1 −iY to −1+iY , closing the rectangle. Then 1 2πiZ(2) −Z(−1) −ζ′ ζ(s) Φ(s)ds =X |ℑρ|≤Y Ress=ρ −ζ′ ζ(s)Φ(s)+ Ress=1 −ζ′ ζ(s)Φ(s), (D.18) 136 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) the horizontal integrals tending to 0. Consequently, P(φ) = X ρbφ ρ−1 2 i−Φ(1) + 1 2πi Z(−1) ζ′ ζ(s) Φ(s)ds. (D.19) Proof. For φη as in the proof of Proposition D.1 the statement follows from Cauchy’s theorem, since Φ η is entire and obeys the vertical decay (D.10) ; the bound −ζ′/ζ ( σ + it ) ≪log (2 + |t| ) on σ∈ [ − 1 , 2] then implies the horizontal integrals vanish as Y→ ∞ . The residues at s = ρ contribute Φ η ( ρ ) = cφη (( ρ− 1 2 ) /i )and the pole at s = 1 contributes − Φ η (1) = −cφη (1 / (2 i )). Finally pass to the limit η↓ 0as in Step 6 of the proof of Proposition D.1. □ Corollary D.1 (Accounting for trivial zeros via the gamma block).Using the functional equation as in (D.14) , the left - edge integral in (D.19) equals 1 2πi Z(2) −ζ′ ζ(s) Φ(1−s)ds +1 2πZ∞ −∞ ℜΓ′ Γ1 4+it 2bφ(t)dt −bφ −1 2i, (D.20) thereby producing the symmetric ±log n Dirichlet–Euler terms and leaving the trivial zeros of ζ accounted for by the gamma integral stated in (D.9). Remark D.5 (Normalisation of constants).Our convention absorbs the log π constant into the “endpoint evaluations” bφ ( ±1 2i ), resulting in a gamma block displayed with the real part of Γ ′/ Γ. This matches the conventions of Appendix C (and is equivalent to other common normalisations differing by a harmless multiple of φ(0)). D.4. Interchanges with the time window: fixed– T and T –uniform cases. Recall ϖT ( t ) = ( √πT ) −1e−t2/T2 (mass one). The window is an external averaging device; it never appears inside EF contour integrals and is applied only after the EF decomposition is in hand. Interchanges involving RR ( · ) ϖT ( t ) dt fall into two regimes. Fixed–Tdominated convergence. Lemma D.3 (Tonelli/dominated convergence at fixed scale). Let F ( x, t ) ≥ 0be measurable on [ ϵ, 1 −ϵ ] ×R , and suppose there is DT∈L1 ( R, ϖTdt )(depending on the fixed T > 0) such that F(x, t)≤DT(t)for all x∈[ϵ, 1−ϵ]. Then for any R∈S0: THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 137 (1) ZR ZR F ( x, t ) R ( x ) dx ϖT ( t ) dt < ∞ and this double integral equals either iterated integral (Tonelli). (2) Differentiation in t and limits in auxiliary parameters (e.g. α↓ 0in a family Fα ) may be moved under RF ϖTdt (dominated convergence). (3) If a series PkFk ( x, t )satisfies Fk≤DT termwise, then one may interchange Pkwith RϖT(Tonelli). Proof. Immediate from Tonelli and dominated convergence since DTϖTis integrable. □ T –uniform interchanges for EF blocks. In the main text and in Appendix E we work with the EF split into three “blocks”: BR(t)∈ { gamma block,Dirichlet–Euler block,zero–sum block }, (D.21) each depending linearly on a test R∈S0 (the precise form is recorded in Section 5.7). What is needed there is uniform L1 control in T , and, for the compact–path Lyapunov argument, uniformity over compact kernel families K⊂S0. Proposition D.2 ( T –uniform EF interchanges).There exist constants CΓ ( R ), CDE ( R ), CZ ( R ), depending only on finitely many seminorms of R, such that for all T > 0: ZR|BR(t)|ϖT(t)dt ≤CB(R),B ∈ {Γ,DE, Z}.(D.22) Consequently, interchanges of the ϖT –integral with the corresponding sums/integrals in each block are justified uniformly in T. In particular, for any index set I, ZX k∈I Bk(t)ϖT(t)dt ≤X k∈I Z|Bk(t)|ϖT(t)dt. (D.23) Moreover, if K⊂S0 is compact, then by Lemma C.2 (Appendix C) there exist constants CΓ ( K ) , CDE ( K ) , CZ ( K ) <∞ such that sup R∈KZR|BR(t)|ϖT(t)dt ≤CB(K),∀T > 0,B ∈ {Γ,DE, Z}. (D.24) Thus both Tonelli and dominated convergence apply uniformly in Tand uniformly over R∈K. Proof (repackaging of Appendix E). The gamma block is controlled by vertical - line Stirling, ℜ Γ ′/ Γ(1 / 4 + it/ 2) = log ( |t|/ 2) + 144 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) frequency decomposition (see the discussion preceding this subsection and §5.7). It is harmless because: (i) it does not alter the |ν|→∞ regime; (ii) it produces the needed integrable ν→ 0behaviour; and (iii) it is compatible with the Kη frequency smoothing used for robustness. Remark (minimal vanishing at the critical line). Any admissible kernel R∈S0 with vanishing of order ≥ 2at x = 1 2 yields the same mechanism; quadratic vanishing is the minimal requirement needed to remove the on–line pole and obtain the explicit–formula bounds used here. Proof. Fix t∈R and a zero ρ = β + iγ . Put ∆ := t−γ , a := | ∆ | , and u:= σ−β. Then 1 σ+it −ρ=u−isgn(∆) a u2+a2=πQa(u)−isgn(∆) Pa(u).(E.12) Pairing against Φν(σ) = pR(σ)e−2πiνσ yields ZR √R(σ)e−2πiνσ σ+it −ρdσ =πZR √R(β+u)e−2πiν(β+u)Qa(u)−isgn(∆) Pa(u)du =π e−2πiνβ \ √Rβ·(Qa−isgn(∆)Pa)(ν). By the product–to–convolution identity for Fourier transforms, \ √Rβ·(Qa−isgn(∆)Pa)(ν) = [ √Rβ∗\ Qa−isgn(∆)Pa(ν), (E.13) and Lemma E.1 gives \ Qa−isgn(∆)Pa(ξ) = −isgn(ξ)−isgn(∆)e−2πa|ξ|.(E.14) Hence the per–zero contribution equals πe−2πiνβ ZR [ √Rβ(ν−ξ)−isgn(ξ)−isgn(∆)e−2π|∆||ξ|dξ, (E.15) which is the raw coefficient κraw R ( ν ; β, ∆). Summing over zeros with multiplicity gives the raw EF–linearised zero block. Finally we apply the fixed ϑ –normalisation κR = κraw R− ϑ ( ν ) κraw R (0; β, ∆), which is equivalent to replacing the integrand [ √Rβ ( ν−ξ )by [ √Rβ ( ν−ξ ) −ϑ ( ν ) [ √Rβ ( −ξ ). This gives (E.8) and therefore (E.7) . The low–frequency estimate (E.11) is proved in the body of the lemma. □ THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 145 Remark E.1 (No use of RH).The representation (E.7) and the coefficient bounds stated in Lemma E.2 (together with the low– frequency envelope (E.11) ) are unconditional and rely only on the functional equation, the Poisson–kernel structure, and the admissibility R∈S0 . No information about the location or spacing of zeros (beyond unit–band counts) is used. E.2. Vertical envelopes for Poisson–weighted zero sums. Let N ( u ; 1) denote the number of nontrivial zeros ρ = β + iγ with ordinates γ∈ ( u, u + 1], counted with multiplicity. The unconditional Riemann–von Mangoldt formula on unit intervals states that N(u; 1) ≪log(2 + |u|),(E.16) with an absolute implied constant. Lemma E.3 (Poisson–weighted zero sum, multiplicity m ( ρ )). For ν∈R\{0}and t∈R, X ρ m(ρ)e−4π|ν||t−γ|≪1 + log(2 + |t|)1 + 1 + log(2 + |ν|−1) |ν|, (E.17) where the sum is over all nontrivial zeros ρ = β + iγ with multiplicity. The implied constant is absolute. In particular, no pointwise decay in |ν| is claimed or possible uniformly in t : at t=γthe left side is ≥1for every ν. Proof. Partition zeros by distance of ordinates from t . For each integer k≥0, let Ak(t) := nρ=β+iγ :k≤ |t−γ|< k + 1o.(E.18) Then X ρ m(ρ)e−4π|ν||t−γ|≤∞ X k=0 e−4π|ν|kX ρ∈Ak(t) m(ρ) = ∞ X k=0 e−4π|ν|kNk(t), (E.19) where Nk ( t ) := Pρ∈Ak(t)m ( ρ ). By (E.16) , each band Ak ( t )lies in O(1) unit intervals, hence Nk(t)≪log(2 + |t|+k)≪log(3 + |t|) + log(1 + k).(E.20) Therefore ∞ X k=0 e−4π|ν|kNk(t)≪1 + log(3 + |t|)∞ X k=0 e−4π|ν|k+∞ X k=0 e−4π|ν|klog(1 + k). (E.21) 146 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) The geometric series satisfies Pk≥0e−4π|ν|k≪ 1 + |ν|−1 . For the logarithmic sum, compare to the integral: ∞ X k=0 e−4π|ν|klog(1+k)≪1+Z∞ 0 e−4π|ν|xlog(1+x)dx ≪1+1 + log(2 + |ν|−1) |ν|, (E.22) by integration by parts (or the change of variables y = |ν|x ). Combining these bounds gives the claim. □ Lemma E.4 (Poisson–weighted zero sum, multiplicity m ( ρ ) 2 ). For ν∈R\{0}and t∈R, X ρ m(ρ)2e−4π|ν||t−γ|≪1 + log2(2+|t|)1 + 1 + log(2 + |ν|−1) |ν|, (E.23) where the sum is over all nontrivial zeros ρ = β + iγ with multiplicity. The implied constant is absolute. Proof. Decompose into the same bands Ak(t). On each band, X ρ∈Ak(t) m(ρ)2≤X ρ∈Ak(t) m(ρ)2=Nk(t)2,(E.24) and Nk(t)≪log(2 + |t|+k)as above. Hence X ρ m(ρ)2e−4π|ν||t−γ|≤∞ X k=0 e−4π|ν|kNk(t)2≪∞ X k=0 e−4π|ν|klog2(2+|t|+k). (E.25) Using log2 (2+ |t| + k ) ≪log2 (3+ |t| )+ log2 (1+ k )and repeating the geometric/integral comparison (with log2 (1 + k )in place of log (1 + k )) yields the stated bound, with the same unavoidable 1 /|ν| scale at low frequency and no false pointwise ν –decay uniformly in t.□ E.3. Main windowed mean–square bound. We now bound the Gaussian–windowed mean square of ZR(ν, t). Theorem 2 (Windowed zero–sum lemma).Fix R∈S0 and T > 0. With the mass–one Gaussian ϖT ( t ) = ( √π T ) −1e−t2/T2 one has, for all ν∈R, ZR|ZR(ν, t)|2ϖT(t)dt ≤CZ(R) (1+|ν|)21+log3(3+T)1+log2 2+|ν|−1, (E.26) where CZ ( R )depends on finitely many S –seminorms of R (and on the fixed cutoff ϑ ) and is independent of T ; all T –dependence THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 147 in this bound appears through the explicit factor 1 + log3 (3 + T ). (For ν = 0 we interpret log (2 + |ν|−1 )as 0on the right-hand side.) If R ranges in a compact set K⊂S0 , the same bound holds with a constant CZ(K)uniform in R∈K. Proof. We treat ν = 0; the case ν = 0 is harmless because ZR(0, t) = 0 by the built–in ϑ–normalisation (see Lemma E.2). From (E.7) we have |ZR(ν, t)| ≤ sup β∈[0,1],∆∈R|κR(ν;β, ∆)|X ρ m(ρ)e−2π|ν||t−γ|=: KR(ν)S(ν, t). (E.27) By (E.11) , KR ( ν ) ≪R|ν| for |ν| ≤ 1. By the Schwartz regularity in ν stated in Lemma E.2, one has KR ( ν ) ≪R (1+ |ν| ) −2 for all ν . For large |ν| , the supremum KR ( ν ) := supβ∈[0,1],∆∈R|κR ( ν ; β, ∆) | is controlled by dropping the Poisson factor in (E.8) , since e−2π|∆||ξ|≤ 1for all ∆, so the ν –decay of [ √Rβ yields the stated envelope for KR ( ν ); combined with the uniform low–frequency bound (E.11) , this controls all ν . Combining these two bounds gives KR(ν)≪R|ν| (1+|ν|)2(ν∈R).(E.28) Thus ZR|ZR(ν, t)|2ϖT(t)dt ≪RKR(ν)2ZR S(ν, t)2ϖT(t)dt. (E.29) Apply Cauchy–Schwarz over the zero index: S(ν, t)2=X ρ m(ρ)e−2π|ν||t−γ|2≤A(ν, t)B(ν, t),(E.30) where A(ν, t) := X ρ m(ρ)2e−4π|ν||t−γ|, B(ν, t) := X ρ e−4π|ν||t−γ|.(E.31) No cancellation between distinct zeros is assumed; all bounds proceed by absolute values, Cauchy–Schwarz, and the unit–band counting input. By Lemma E.4 and Lemma E.3 (with m(ρ)≡1), A(ν, t)B(ν, t)≪1+log3(2+|t|)1+ 1 + log(2 + |ν|−1) |ν|2.(E.32) 148 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) Multiplying by KR ( ν ) 2≪R|ν|2 (1 + |ν| ) −4 and using the elementary estimate |ν|21+ 1 + log(2 + |ν|−1) |ν|2≪(1+|ν|)21+log2 2+|ν|−1,(E.33) we obtain the pointwise bound |ZR(ν, t)|2≪R 1 + log2(2+|ν|−1) (1+|ν|)21 + log3(2+|t|).(E.34) Integrating in t against ϖT and using Appendix B, Lemma B.4 with k= 3 gives ZR1 + log3(2+|t|)ϖT(t)dt ≪1 + log3(3+T),(E.35) which yields (E.26) . Uniformity over compact K⊂S0 follows from the dependence of the constants on finitely many seminorms of R.□ Remark E.2 (What is and is not T –uniform).The bound (E.26) is uniform in T > 0up to the harmless factor  1 + log3 (3 + T )  , which is the natural growth one expects from vertical envelopes of ξ′/ξ under the Gaussian window. For the Lyapunov/EF contradiction we only need, for each fixed T , finiteness of the windowed zero–block energy and a ν –decay sufficient to ensure the convergence of the frequency integral; both are supplied by (E.26) . The additional factor 1+ log2 (2+ |ν|−1 )is harmless: it is integrable at ν = 0 and does not affect any fixed– T contradiction. Lemma E.5 (Schur estimate for the zero–block).Under the hypotheses of Theorem 2 one has sup ν∈R (1+|ν|)2 1 + log2(2+|ν|−1)ZR|ZR(ν, t)|2ϖT(t)dt ≤CZ(R)1+log3(3+T), (E.36) with CZ ( R )as in (E.26) . If R ranges in a compact set K⊂S0 , the same bound holds with a constant CZ ( K )uniform in R∈K . Proof. Immediate from (E.26). □ E.4. Robustness variants (windows). Proposition E.1 (Mass–one Schwartz windows).Let ω∈S ( R ) be nonnegative with RRω = 1, and set ωT ( t ) := T−1ω ( t/T ). Then Theorem 2 holds with ϖT replaced by ωT , with a constant THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 149 CZ ( R, ω )independent of T and with the same explicit factor 1 + log3(3+T)in the bound, namely ZR|ZR(ν, t)|2ωT(t)dt ≤CZ(R, ω) (1+|ν|)21+log3(3+T)1+log2 2+|ν|−1, (E.37) where CZ ( R, ω )depends on finitely many seminorms of R and ω and is uniform in T > 0. If R∈K⊂S0 and ω vary in compact families, the constants are uniform on those families. Proof. The argument follows the proof of Theorem 2 verbatim once the zero sums have been bounded in terms of log (2 + |t| ) and log(2 + |ν|−1). Since ω∈S ( R )is fixed, nonnegative, and satisfies RRω = 1, its rescalings ωT ( t ) = T−1ω ( t/T )obey the uniform moment bound ZR1 + log3(2+|t|)ωT(t)dt ≪ω1 + log3(3+T),(E.38) uniformly for T > 0, by rapid decay of ω and the change of variables t=Tu. Replacing ϖT by ωT therefore affects only the window moment estimate; all frequency–side bounds are unchanged. The stated inequality follows with a constant CZ ( R, ω )depending on finitely many seminorms of Rand ωand independent of T.□ E.5. Where and how this lemma is used. In §5.7 the EF assembly decomposes the (windowed) energy into three blocks: ZR ER(t)ϖT(t)dt =BΓ[R;T]+BDE[R;T]+BZ[R;T].(E.39) Appendix D identifies ER ( t )with R|BR ( ν, t ) |2dν by weighted Parseval (Lemma D.1) and records EF admissibility and contour calculus. The present appendix supplies a windowed mean– square bound on the zero block BZ [ R ; T ], with explicit large– |ν| decay and a harmless integrable low–frequency factor 1 + log2 (2 + |ν|−1 ), together with controlled logarithmic T –growth. This ensures that the associated frequency integrals converge absolutely for each fixed T , and remain uniformly controlled over compact kernel families K⊂S0. Together with the bounds for the gamma and prime blocks (Appendix D and §5.7), this provides the global EF–bound used in the Lyapunov contradiction and its phase–locked variant: all zero contributions are controlled at the level required for 150 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) the fixed–sign flux argument, without any assumption on zero spacings beyond unit–band counting. This completes Appendix E. THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 151 Appendix F. Explicit–formula normalisation and the limit α↓0 Aim. Show that admissible kernels can be approximated by Gaussian–mollified, jet–pinned kernels that remain admissible (for α small) and converge to the target in the Schwartz topology; prove that each explicit–formula (EF) block depends continuously on the kernel in that topology so that, after establishing T –uniform EF bounds, one may pass to the limit α↓ 0. The order of regulators is: –first, fix Rand obtain EF bounds uniformly in T; –only then pass to α↓0for a mollified sequence R(α)→R. At no stage are ζor ξmodified (“measure, not modify”). Notation. Write S(R)for the Schwartz space and S0:= nR∈S(R) : Ris real, even about 1 2, R(1 2) = R′(1 2)=0, R′′(1 2)>0, and there exists a fixed √R∈S(R)with (√R)2=Ro. (F.1) For seminorms we use ∥f∥S;(a,b):= sup x∈R (1+|x|)af(b)(x)(a, b ∈N0).(F.2) F.1. Gaussian mollification with jet pinning preserves admissibility. We adopt the approximate–identity normalisation for the Gaussian: ϕα(y) := 1 α√πe−(y/α)2(α > 0),(F.3) so that ϕα∈S,ϕαis even, has mass 1, and c ϕα(ξ)=e−(παξ)2−−→ α↓01for each ξ∈R.(F.4) For f:R→Cdefine the centred mollification about x=1 2by (Mαf)(x) := (f(1 2+·)∗ϕα)x−1 2=ZR f1 2+yϕα x−1 2−ydy. (F.5) Remark F.1 (Why we pin the jet at x = 1 2 ).Raw convolution Mα preserves evenness and smoothness, and ( Mαf ) ′ ( 1 2 ) = 0 whenever f is even about 1 2 . However in general ( Mαf )( 1 2 )  = f ( 1 2 )and ( Mαf ) ′′ ( 1 2 )  = f′′ ( 1 2 ). Since admissibility in S0 requires the precise jet f ( 1 2 ) = f′ ( 1 2 )=0and f′′ ( 1 2 ) > 0, we pin these 152 PROF. ELIAHI PRIEST HON.DS.C(UFSEI) entries after mollification by subtracting a fixed, even bump that equals 1near 1 2(to fix the value) and—if desired—adding a localised quadratic bump (to fix the curvature). This local renormalisation stays inside S and leaves all EF manipulations intact. Fix once and for all two even bump functions centred at 1 2 : choose χ0, χ2∈C∞ c(R), even, with χj(0) = 1, and set ψ0(x):=χ0x−1 2, ψ2(x):=χ2x−1 2(x−1 2)2. Then ψ0, ψ2∈Sare even about 1 2and satisfy ψ0(1 2) = 1, ψ′ 0(1 2) = ψ′′ 0(1 2)=0, ψ2(1 2) = ψ′ 2(1 2)=0, ψ′′ 2(1 2)=2. Define the pinned Gaussian mollification of Rby R(α)(x) := (MαR)(x)−(MαR) 1 2ψ0(x)+R′′(1 2)−(MαR)′′(1 2) 2ψ2(x), α > 0. (F.6) On the frequency side this reads d R(α)(ξ) = b R(ξ)e−(παξ)2−Aαc ψ0(ξ) + Bαc ψ2(ξ), Aα:= (MαR) 1 2, Bα:= R′′(1 2)−(MαR)′′(1 2) 2. (F.7) Proposition F.1 (Pinned mollification preserves S0 and converges in S).If R∈S0, then: (i) R(α)∈S(R)is real and even about 1 2for every α > 0. (ii) There exists α0=α0(R)>0such that, for all 0< α ≤α0, R(α)∈S0,(F.8) i.e. R(α) ( 1 2 ) = ( R(α) ) ′ ( 1 2 ) = 0,( R(α) ) ′′ ( 1 2 ) = R′′ ( 1 2 ) > 0, and R(α)has the same unique quadratic minimum at 1 2as R. (iii) As α↓0, R(α)−−→ α↓0Rin S(R),(F.9) and in particular (R(α))(k)(1 2)→R(k)(1 2)for each k∈N0. Proof. Since ϕα is even of mass one and (F.5) is a centred convolution, Mα maps real, 1 2 –even functions to real, 1 2 –even functions and preserves the Schwartz class. The bump corrections in (F.6) are Schwartz and even, hence R(α)∈S ( R )is real and 1 2 –even, proving (i). THE RIEMANN HYPOTHESIS VIA A LYAPUNOV DYNAMIC CASCADE 153 For the jet at 1 2we use the defining properties of ψ0and ψ2: R(α) 1 2= (MαR) 1 2−(MαR) 1 2ψ0 1 2= 0,(F.10) (R(α))′1 2= (MαR)′1 2−(MαR) 1 2ψ′ 0 1 2+R′′(1 2)−(MαR)′′(1 2) 2ψ′ 2 1 2= 0, (F.11) and (R(α))′′1 2= (MαR)′′1 2−(MαR) 1 2ψ′′ 01 2+R′′(1 2)−(MαR)′′(1 2) 2ψ′′ 21 2=R′′1 2>0. (F.12) Thus the value, slope, and curvature at x = 1 2 are pinned exactly. Because R is even about x = 1 2 with a unique quadratic minimum there, its Taylor expansion has the form R 1 2+y=R′′(1 2) 2y2+O(|y|3).(F.13) Standard properties of convolution with the approximate identity ϕαgive MαR→Rin C2 loc(R)as α↓0.(F.14) In particular, near x=1 2we have MαR 1 2+y=MαR 1 2+MαR′′(1 2) 2y2+O(|y|3).(F.15) Subtracting the constant term ( MαR )( 1 2 )and adding the quadratic correction in (F.6), one obtains R(α) 1 2+y=R′′(1 2) 2y2+O(|y|3),(F.16) with the implicit constant independent of α for α sufficiently small. Thus R(α) has the same local quadratic minimum at x=1 2as R. Away from 1 2 , the corrections in (F.6) are supported in a fixed compact set and tend to 0uniformly (together with all derivatives) as α↓ 0, since ( MαR )( 1 2 ) →R ( 1 2 ) = 0 and ( MαR ) ′′ ( 1 2 ) →R′′ ( 1 2 ). As R is nonnegative with a unique quadratic minimum at 1 2 , continuity and compactness give a positive lower bound for R on the complement of a small neighbourhood of 1 2 ; for α small enough the corrections are too small to change the sign there. Combining this with the local expansion above yields an α0 ( R ) > 0such that R(α)≥ 0and has the same unique quadratic minimum at 1 2 for all 0 < α ≤α0 , i.e. R(α)∈S0 , proving (ii).