Truncated Laplace transforms of bounded functions #
For f : ℝ → ℂ, this file studies the Laplace transform G z = ∫ t in Ici 0, f t * exp (-z t)
and its truncations G_T z = ∫ t in 0..T, f t * exp (-z t):
- each truncation is entire (
differentiable_intervalIntegral_mul_cexp_neg_mul), and, forfbounded on[0, ∞), the full transform is holomorphic on0 < re z(differentiableOn_integral_mul_cexp_neg_mul), being a locally uniform limit of truncations; - if
‖f‖ ≤ Mon[0, ∞), the tail bound‖G z - G_T z‖ ≤ M exp (-T re z) / re zholds for0 < re z(norm_integral_sub_intervalIntegral_mul_cexp_neg_mul_le), and if‖f‖ ≤ Mon[0, T], then‖G_T z‖ ≤ M exp (-T re z) / (-re z)forre z < 0(norm_intervalIntegral_mul_cexp_neg_mul_le_of_re_neg).
The substitution x = exp t identifies ∫ x in u..v, x ^ (-z - 1) with the truncated Laplace
transform ∫ t in log u..log v, exp (-z t) of the constant function 1
(intervalIntegral_ofReal_cpow_neg_sub_one); this links Dirichlet series, via integrals against
x ^ (-z - 1), to Laplace transforms of step functions t ↦ a ⌊exp t⌋₊.
These are the analytic inputs of Newman's contour argument for his Tauberian theorem.
Multiplying by the exponential kernel cexp (-z t) does not change interval integrability.
Multiplying by the exponential kernel cexp (-z t) does not change almost-everywhere strong
measurability.
Truncated Laplace transforms are entire. The function
z ↦ ∫ t in a..b, f t * exp (-z t) is complex differentiable everywhere. If f is not
interval integrable on [a, b], it is identically 0.
Substitution x = exp t. For 0 < u and 0 < v, the integral of x ^ (-z - 1) over
[u, v] is the truncated Laplace transform ∫ t in log u..log v, exp (-z t) of the constant
function 1.
For 0 < re z, the Laplace integrand of a function bounded on [0, ∞) is integrable there.
Tail bound. For 0 < re z and 0 ≤ T, the Laplace transform of a function bounded by
M on [0, ∞) differs from its truncation at T by at most M exp (-T re z) / re z.
Left half-plane bound. For re z < 0 and 0 ≤ T, the truncated Laplace transform of a
function bounded by M on [0, T] is at most M exp (-T re z) / (-re z).