Small-noise and system-size approximations (Gardiner, Ch. 6 and Ch. 7) #
Index of the approximation layer built on the dCKE layer of SHSLib.Gardiner
(C. W. Gardiner, Handbook of Stochastic Methods, §6.1, §6.3, §7.1.2, §7.2, §7.5, with the
covariance equations of §4.4.6 and §4.4.9). Conventions are those of SHSLib/Gardiner.lean:
Sp n = EuclideanSpace ℝ (Fin n), pd/pd2 are partial derivatives defined through fderiv,
and fpRHS is the Fokker–Planck right-hand side (3.5.6).
Master equations are mostly studied through their backward generator
(L f)(x) = Σ_j W_j(x) [f(x + r_j) - f(x)] (or its integral version); this is the operator
appearing in the derivation "as in Sect. 3.4" and it is where all error terms are stated.
The limit and integral interchanges the text makes silently are named hypotheses
(hswap, …), as in the dCKE layer.
Approximation guarantees with explicit error terms #
| Guarantee | Lean |
|---|---|
master-equation generator vs. linear-noise operator, O(Ω^{-1/2}) (1-d) | Gardiner.vkGen_sub_lnaGen_le, Gardiner.vkGen_tendsto_lnaGen |
| same, multivariate (chemical networks) | Gardiner.mvkGen_sub_mlnaGen_le |
same, for X ⇌ A (instance) | Gardiner.xa_vkGen_sub_lnaGen_le |
small-noise generator vs. Ornstein–Uhlenbeck generator, O(ε) | Gardiner.snGen_sub_ouGen_le |
| Kramers–Moyal truncation error via the next jump moment | Gardiner.jumpGen_sub_kmGen_le, Gardiner.finJumpGen_sub_finKM_le |
same, on ℝⁿ (third absolute jump moment) | Gardiner.mjGen_sub_mjKM_le, Gardiner.chem_KM_le |
rate O(√δ) of the limit (7.2.6) | Gardiner.scaled_jumpGen_rate, Gardiner.scaled_jumpGen_tendsto |
rate O(δ) of the stationary limit (7.2.20)–(7.2.21) | Gardiner.bdStat_log_rate, Gardiner.bdStat_ou_rate |
critical slow-time generator, O(Ω^{-1/(q+1)}) | Gardiner.critGen_sub_critLim_le, Gardiner.cubic_critGen_sub_critLim_le |
Kurtz coupling ⇒ characteristic functions, O(V^{-1/2} log V) | Gardiner.kurtz_charFun_le, Gardiner.kurtz_charFun_lna_le |
| exactness of the LNA mean and variance for linear propensities | Gardiner.lna_exact_of_linear, Gardiner.xa_lna_exact, Gardiner.mlna_exact, Gardiner.cme_lna_exact |
| (6.3.48) with explicit error | Gardiner.ex_c1_approx |
§6.1 and §6.3: the small noise expansion of the Fokker–Planck equation #
§7.1.2: the exact solution of X ⇌ A #
| Gardiner | Lean |
|---|---|
| (7.1.24)–(7.1.26) | Gardiner.xaME |
| (7.1.27)–(7.1.28) | Gardiner.xaOp, Gardiner.coeff_xaOp, Gardiner.SolvesXA, Gardiner.solvesXA_iff |
| (7.1.34)–(7.1.37) | Gardiner.xaGen, Gardiner.xa_exact_solution, Gardiner.xaGen_zero, Gardiner.coeff_xaGen, Gardiner.xaGenFun, Gardiner.xaGenFun_hasDerivAt_t |
| (7.1.38)–(7.1.40) | Gardiner.xaGenFun_deriv_one, Gardiner.xaGenFun_deriv2_one, Gardiner.xa_variance_formula, Gardiner.xa_mean_variance_exact |
| (7.1.41)–(7.1.43) | Gardiner.xa_moment_identities, Gardiner.xa_factorial_moment_identity |
| (7.1.44)–(7.1.45) | Gardiner.xaGenFun_tendsto, Gardiner.poisson_stationary |
| (7.1.49)–(7.1.55) | Gardiner.poissonPMF, Gardiner.poisson_solves_xaME, Gardiner.solvesXA_poisson, Gardiner.xaMean |
§7.2.1–7.2.2: jump-process approximation of a diffusion; Kramers–Moyal #
| Gardiner | Lean |
|---|---|
| (7.2.1)–(7.2.3) | Gardiner.scaledW, Gardiner.ScaledKernelHyp |
| (7.2.4) | Gardiner.integral_mul_scaledW, Gardiner.scaledW_moment_zero/one/two |
| (7.2.5) | Gardiner.scaledW_tendsto_zero |
| (7.2.6) with rate | Gardiner.scaled_jumpGen_rate, Gardiner.scaled_jumpGen_tendsto |
| (7.2.9)–(7.2.13) | Gardiner.randomWalk_moments, Gardiner.randomWalk_rate |
| (7.2.14)–(7.2.15) | Gardiner.poisson_moments, Gardiner.poisson_no_diffusion_limit |
| (7.2.16)–(7.2.18) | Gardiner.bdRate, Gardiner.bd_moments, Gardiner.bd_rate |
| (7.2.19) | Gardiner.bdStat, Gardiner.bdStat_flux, Gardiner.bdStat_stationary, Gardiner.bdStat_eq_ratio |
| (7.2.20)–(7.2.21) with rate | Gardiner.bdStat_log_rate, Gardiner.bdStat_ou_rate |
| (7.2.22)–(7.2.23) | Gardiner.km_master_change_of_variables |
| (7.2.25)–(7.2.26) | Gardiner.jumpMoment, Gardiner.kmGen, Gardiner.finJumpMoment, Gardiner.finKM |
| truncation error | Gardiner.jumpGen_sub_kmGen_le, Gardiner.finJumpGen_sub_finKM_le |
| (7.2.27) | Gardiner.scaledW_jumpMoment_abs_le, Gardiner.scaledW_jumpMoment_tendsto_zero |
§7.2.3: van Kampen's system size expansion #
§7.2.4–7.2.5: Kurtz's theorem and critical fluctuations #
| Gardiner | Lean |
|---|---|
| (7.2.64) | the hypothesis hcouple |
| (7.2.65)–(7.2.67) | Gardiner.kurtz_phase, Gardiner.kurtz_charFun_le, Gardiner.kurtz_charFun_tendsto, Gardiner.kurtz_charFun_lna_le |
| (7.2.68), (7.2.70)–(7.2.75) | Gardiner.critGen, Gardiner.critLim, Gardiner.critGen_sub_critLim_le |
| (7.2.71), (7.2.73), (7.2.77) | Gardiner.critical_exponent, Gardiner.critical_scales |
| (7.2.69) | Gardiner.critical_cubic, Gardiner.critical_cubic_solution, Gardiner.cubic_critGen_sub_critLim_le |
| (7.2.76) | Gardiner.critical_mean_eq |
§7.5: combinatorial kinetics, chemical Fokker–Planck equation, linear noise approximation #
| Gardiner | Lean |
|---|---|
| (7.5.3)–(7.5.5) | Gardiner.stoich |
| (7.5.6)–(7.5.7) | Gardiner.orthant_forward, Gardiner.orthant_backward, Gardiner.propCount_ne_zero_iff |
| (7.5.8) | Gardiner.propCount, Gardiner.propPoly, Gardiner.propPoly_natCast, Gardiner.tplus, Gardiner.tminus |
| (7.5.9) | Gardiner.cme, Gardiner.cme_eq, Gardiner.latME, Gardiner.tsum_mul_latME |
| (7.5.30) → (7.5.31) | Gardiner.mjGen_sub_mjKM_le, Gardiner.chem_KM_le |
| (7.5.31) | Gardiner.chemFPE |
| (7.5.32) | Gardiner.chemDrift, Gardiner.chemDiff, Gardiner.chemDiff_nonneg, Gardiner.chem_latDrift, Gardiner.chem_latDiff, Gardiner.chem_mjDrift, Gardiner.chem_mjDiff |
| linearisation about the deterministic solution | Gardiner.mvkGen, Gardiner.mlnaGen, Gardiner.mvkGen_sub_mlnaGen_le |
| (4.4.51) | Gardiner.lyapunov_iff, Gardiner.lna_cov_stationary |
| (4.4.83)–(4.4.85) | Gardiner.meanLin, Gardiner.covLin, Gardiner.mlna_exact, Gardiner.cme_lna_exact |
X ⇌ A as a network | Gardiner.xa_chem_affine |
Errata (formalized in corrected form; see the module docstrings) #
- (7.2.35), (7.2.40): the drift is
α̃₁'(φ) z, notα̃₁(φ) z(SystemSize). - (7.2.56), (7.2.58):
k₁fork₂in the numerator andδ + δ²forδ - δ²(ReactionXAvk). - (6.3.19)–(6.3.21):
M¹in place ofM⁰in theÃ₁terms; (6.3.28):r ≥ 1, notr ≠ 1(SmallNoiseMoments). - (6.3.43):
e^{λᵢ t}, note^{-λᵢ t}; (6.3.51): the factor-2λ₁(SmallNoiseExample). - (6.3.54): coefficients of the
ε²andε⁴terms (SmallNoiseStationary). - (7.2.4):
α₂ = ∫ y²Φ + δA²Q; the text's value is the leading order (KramersMoyal).