A real-valued Lévy process describes random evolution with stationary and independent increments. Brownian motion, compound-Poisson risk models and infinite-activity jump processes belong to this class. Their analytical theory is a foundation for insurance risk, mathematical finance and stochastic control. Lévy–Khintchine representation and the construction of Poisson random measures identify the possible laws; fluctuation theory describes first passage, ruin and reflection. The latter results are used in Avram, Palmowski and Pistorius (2007) to study optimal dividend barriers.
The LeanLevy formalisation provides a machine-checked starting point for characteristic functions, Lévy–Khintchine theory, càdlàg paths, jump random measures and pure-jump processes. This broader mission adapts those foundations into Prove2Me and extends them with source-backed fluctuation and stochastic-calculus results needed by dividend control. It is a single, multi-source formalisation campaign: the frozen LeanLevy version is the source for its original results, and subsequent topics have their own mathematical provenance.
A Lévy process starts from zero and has independent stationary increments and càdlàg sample paths. Its characteristic exponent is defined by
A Lévy–Khintchine triple comprises drift, a nonnegative Gaussian-variance coefficient and a Lévy measure satisfying and . An infinitely divisible law has convolution roots of every positive integer order. A Poisson random measure records jump counts on measurable sets; its compensated version subtracts its intensity.
For a spectrally negative Lévy process, positive jumps are excluded. Write on the nonnegative Laplace domain. For a discount rate , the -scale function is characterised on the positive half-line by its Laplace transform
under the customary existence and normalisation assumptions. First-passage times, reflected paths, regulators and Stieltjes dividend measures are defined explicitly, with time in . Admissible dividend policies are adapted, nondecreasing controls subject to the original solvency and payment conditions.
Formalise finite-measure Fourier transforms, positive-definite characteristic functions, Bochner representation, Lévy continuity, consistent finite-dimensional laws, Kolmogorov extension and càdlàg versions. Establish Lévy–Khintchine representation, converse, characterisation and uniqueness with explicit truncation and integrability conventions. Construct pure-jump processes and account separately for Brownian/Gaussian components where the stronger construction requires them.
Construct Poisson random measures and independent scattering; establish thinning, Campbell identities, measurable jump evaluation, compensated stochastic integrals, isometries and genuine adapted martingales. Distinguish finite-activity compound-Poisson results from infinite-activity and unbounded-variation claims.
Establish stopping-time measurability for passage and ruin events, right-continuous filtration and strong-Markov prerequisites, suitable exponential martingales, optional stopping under its actual integrability or boundedness hypotheses, and stopped-process identities. Statements must distinguish crossing and creeping at the boundary and account for possible infinite stopping times.
Construct and identify -scale functions, their Laplace transforms, positivity and uniqueness; prove the two-sided exit identity with the correct stopping conventions:
for the stated domain and admissible process assumptions. Develop Esscher changes of measure and scale-function asymptotics required to establish finiteness of candidate optimal barriers.
Formalise the relevant origin limits, bounded- versus unbounded-variation behaviour, differentiability and higher regularity under the precise Gaussian or smoothness hypotheses. When excursion theory is used, include its local-time, height-measure and atomlessness assumptions rather than infer differentiability from an arbitrary atomless cumulative measure.
Define Skorokhod reflection and its regulator, predictable or càdlàg control conventions, Stieltjes discounting and strong-Markov renewal identities. Derive the positive-barrier scale-function valuation and its zero-barrier limits under correct initial-capital and ruin conventions. Avoid stopping times that ask a nonnegative reflected reserve to become negative.
Establish finite-activity pathwise change of variables, compensated jump Itô formulas, localisation and integrability, the Lévy generator, stopped Dynkin identities and harmonicity for functions meeting the required regularity conditions. Explicitly separate finite variation, unbounded variation and Gaussian contributions.
Establish admissible finite-variation dividend-control calculus and conditional supermartingale/verification bounds. For a suitable candidate , the familiar inequalities involve and on the relevant reserve regions. Unrestricted barrier optimality requires the original additional generator inequality above the candidate barrier; it must never be inferred from optimality among barrier strategies alone.
Prove type- and source-faithful bridges between this mission's process, exponent, stopping-time and scale-function conventions and those used by the existing Avram Dividend mission. Path-law equivalence transports measurable path events and their laws; filtration-dependent assertions require independent, explicit compatibility hypotheses. The deliverable is a collection of reusable, actually Proved declarations importable under the same Prove2Me verification environment.
A completed development would supply machine-checked infrastructure for first-passage probabilities, scale functions, reflection, discounted rewards and jump-driven stochastic verification. These are the reusable inputs to the dividend-barrier analysis of Avram, Palmowski and Pistorius (2007) and other Lévy-process problems. Formalisation of a named theorem is complete only after Prove2Me accepts its proof and all theorem dependencies resolve, including cross-mission imports.
The transition from finite-dimensional laws to measurable càdlàg path events requires careful exceptional sets, time indexing and filtration handling. Optional stopping needs genuine uniform integrability or localisation. Scale functions are defined analytically but used at random passage times, requiring both transform uniqueness and pathwise stopping identities. Unbounded-variation regularity can require excursion arguments. Reflection creates finite-variation Stieltjes reward terms, and the stochastic verification bound mixes continuous, compensated-jump and singular-control contributions. Each branch must preserve the exact assumptions and distinctions on which the corresponding mathematical theorem depends.
Use real-valued reserve processes, nonnegative real time, the frozen characteristic-function and jump-truncation conventions where compatible, and the pinned Mathlib revision of this proposal. Explain and prove any necessary conversions to the conventions of other sources. The existing LeanLevy results retain their attribution and original scope. Subsequent milestones must cite their distinct sources and include explicit domains, integrability, regularity, measurability, boundary, filtration and discount assumptions.
A vacuous process model, a fixed zero-jump case, an assumed conclusion, a strengthened hypothesis that simply embeds the desired result or a false equality of filtrations does not count as the general theorem. Reuse non-deprecated Mathlib and Prove2Me declarations with compatible environments, preserving canonical identities where possible. The proposal is a private, unverified draft until its statements compile and proofs receive authoritative remote verification. Local Lean compilation is excluded from this workflow.
theorem ProbabilityTheory.IsPoissonPointFamily.indepFun_pieceBlocks [MeasureTheory.IsProbabilityMeasure μ]
(hd : IsPoissonPointFamily K X m μ) (n : ℕ) :
IndepFun
(fun ω (i : {i : ℕ ⊕ ℕ × ℕ // pointFamilyBlock i ≤ n}) => pointFamilyCombined K X i.1 ω)
(fun ω (i : {i : ℕ ⊕ ℕ × ℕ // pointFamilyBlock i = n + 1}) => pointFamilyCombined K X i.1 ω)
μ := by sorrySource-header theorem from the pinned LeanLevy library. The original mathematical assertion is retained with explicit names and source-compatible universe declarations. Its original proof has not been transplanted or verified.
No open leaves. Every sub-goal is proved or awaiting decomposition.