Derivatives of the Fourier transform #
In this file we compute the Fréchet derivative of the Fourier transform of f
, where f
is a
function such that both f
and v ↦ ‖v‖ * ‖f v‖
are integrable. Here the Fourier transform is
understood as an operator (V → E) → (W → E)
, where V
and W
are normed ℝ
-vector spaces
and the Fourier transform is taken with respect to a continuous ℝ
-bilinear
pairing L : V × W → ℝ
and a given reference measure μ
.
We also investigate higher derivatives: Assuming that ‖v‖^n * ‖f v‖
is integrable, we show
that the Fourier transform of f
is C^n
.
We also study in a parallel way the Fourier transform of the derivative, which is obtained by tensoring the Fourier transform of the original function with the bilinear form. We also get results for iterated derivatives.
A consequence of these results is that, if a function is smooth and all its derivatives are
integrable when multiplied by ‖v‖^k
, then the same goes for its Fourier transform, with
explicit bounds.
We give specialized versions of these results on inner product spaces (where L
is the scalar
product) and on the real line, where we express the one-dimensional derivative in more concrete
terms, as the Fourier transform of -2πI x * f x
(or (-2πI x)^n * f x
for higher derivatives).
Main definitions and results #
We introduce two convenience definitions:
VectorFourier.fourierSMulRight L f
: givenf : V → E
andL
a bilinear pairing betweenV
andW
, then this is the functionfun v ↦ -(2 * π * I) (L v ⬝) • f v
, fromV
toHom (W, E)
. This is essentiallyContinuousLinearMap.smulRight
, up to the factor- 2πI
designed to make sure that the Fourier integral offourierSMulRight L f
is the derivative of the Fourier integral off
.VectorFourier.fourierPowSMulRight
is the higher order analogue for higher derivatives:fourierPowSMulRight L f v n
is informally(-(2 * π * I))^n (L v ⬝)^n • f v
, in the space of continuous multilinear mapsW [×n]→L[ℝ] E
.
With these definitions, the statements read as follows, first in a general context
(arbitrary L
and μ
):
VectorFourier.hasFDerivAt_fourierIntegral
: the Fourier integral off
is differentiable, with derivative the Fourier integral offourierSMulRight L f
.VectorFourier.differentiable_fourierIntegral
: the Fourier integral off
is differentiable.VectorFourier.fderiv_fourierIntegral
: formula for the derivative of the Fourier integral off
.VectorFourier.fourierIntegral_fderiv
: formula for the Fourier integral of the derivative off
.VectorFourier.hasFTaylorSeriesUpTo_fourierIntegral
: under suitable integrability conditions, the Fourier integral off
has an explicit Taylor series up to orderN
, given by the Fourier integrals offun v ↦ fourierPowSMulRight L f v n
.VectorFourier.contDiff_fourierIntegral
: under suitable integrability conditions, the Fourier integral off
isC^n
.VectorFourier.iteratedFDeriv_fourierIntegral
: under suitable integrability conditions, explicit formula for then
-th derivative of the Fourier integral off
, as the Fourier integral offun v ↦ fourierPowSMulRight L f v n
.VectorFourier.pow_mul_norm_iteratedFDeriv_fourierIntegral_le
: explicit bounds for then
-th derivative of the Fourier integral, multiplied by a power function, in terms of corresponding integrals for the original function.
These statements are then specialized to the case of the usual Fourier transform on
finite-dimensional inner product spaces with their canonical Lebesgue measure (covering in
particular the case of the real line), replacing the namespace VectorFourier
by
the namespace Real
in the above statements.
We also give specialized versions of the one-dimensional real derivative (and iterated derivative)
in Real.deriv_fourierIntegral
and Real.iteratedDeriv_fourierIntegral
.
Send a function f : V → E
to the function f : V → Hom (W, E)
given by
v ↦ (w ↦ -2 * π * I * L (v, w) • f v)
. This is designed so that the Fourier transform of
fourierSMulRight L f
is the derivative of the Fourier transform of f
.
Instances For
The w
-derivative of the Fourier transform integrand.
Main theorem of this section: if both f
and x ↦ ‖x‖ * ‖f x‖
are integrable, then the
Fourier transform of f
has a Fréchet derivative (everywhere in its domain) and its derivative is
the Fourier transform of smulRight L f
.
The Fourier integral of the derivative of a function is obtained by multiplying the Fourier
integral of the original function by -L w v
.
The formal multilinear series whose n
-th term is
(w₁, ..., wₙ) ↦ (-2πI)^n * L v w₁ * ... * L v wₙ • f v
, as a continuous multilinear map in
the space W [×n]→L[ℝ] E
.
This is designed so that the Fourier transform of v ↦ fourierPowSMulRight L f v n
is the
n
-th derivative of the Fourier transform of f
.
Equations
Instances For
Decomposing fourierPowSMulRight L f v n
as a composition of continuous bilinear and
multilinear maps, to deduce easily its continuity and differentiability properties.
The iterated derivative of a function multiplied by (L v ⬝) ^ n
can be controlled in terms
of the iterated derivatives of the initial function.
Variant of hasFTaylorSeriesUpTo_fourierIntegral
in which the smoothness index is restricted
to ℕ∞
(and so are the inequalities in the assumption hf
). Avoids normcasting in some
applications.
If ‖v‖^n * ‖f v‖
is integrable for all n ≤ N
, then the Fourier transform of f
is C^N
.
If ‖v‖^n * ‖f v‖
is integrable for all n ≤ N
, then the n
-th derivative of the Fourier
transform of f
is the Fourier transform of fourierPowSMulRight L f v n
,
i.e., (L v ⬝) ^ n • f v
.
The Fourier integral of the n
-th derivative of a function is obtained by multiplying the
Fourier integral of the original function by (2πI L w ⬝ )^n
.
The k
-th derivative of the Fourier integral of f
, multiplied by (L v w) ^ n
, is the
Fourier integral of the n
-th derivative of (L v w) ^ k * f
.
One can bound the k
-th derivative of the Fourier integral of f
, multiplied by (L v w) ^ n
,
in terms of integrals of iterated derivatives of f
(of order up to n
) multiplied by ‖v‖ ^ i
(for i ≤ k
).
Auxiliary version in terms of the operator norm of fourierPowSMulRight (-L.flip) ⬝
. For a version
in terms of |L v w| ^ n * ⬝
, see pow_mul_norm_iteratedFDeriv_fourierIntegral_le
.
One can bound the k
-th derivative of the Fourier integral of f
, multiplied by (L v w) ^ n
,
in terms of integrals of iterated derivatives of f
(of order up to n
) multiplied by ‖v‖ ^ i
(for i ≤ k
).
The Fréchet derivative of the Fourier transform of f
is the Fourier transform of
fun v ↦ -2 * π * I ⟪v, ⬝⟫ f v
.
The Fréchet derivative of the Fourier transform of f
is the Fourier transform of
fun v ↦ -2 * π * I ⟪v, ⬝⟫ f v
.
The Fourier integral of the Fréchet derivative of a function is obtained by multiplying the
Fourier integral of the original function by 2πI ⟪v, w⟫
.
If ‖v‖^n * ‖f v‖
is integrable, then the Fourier transform of f
is C^n
.
If ‖v‖^n * ‖f v‖
is integrable, then the n
-th derivative of the Fourier transform of f
is
the Fourier transform of fun v ↦ (-2 * π * I) ^ n ⟪v, ⬝⟫^n f v
.
The Fourier integral of the n
-th derivative of a function is obtained by multiplying the
Fourier integral of the original function by (2πI L w ⬝ )^n
.
One can bound ‖w‖^n * ‖D^k (𝓕 f) w‖
in terms of integrals of the derivatives of f
(or order
at most n
) multiplied by powers of v
(of order at most k
).
The Fourier integral of the Fréchet derivative of a function is obtained by multiplying the
Fourier integral of the original function by 2πI x
.