Line derivatives #
We define the line derivative of a function f : E β F
, at a point x : E
along a vector v : E
,
as the element f' : F
such that f (x + t β’ v) = f x + t β’ f' + o (t)
as t
tends to 0
in
the scalar field π
, if it exists. It is denoted by lineDeriv π f x v
.
This notion is generally less well behaved than the full FrΓ©chet derivative (for instance, the composition of functions which are line-differentiable is not line-differentiable in general). The FrΓ©chet derivative should therefore be favored over this one in general, although the line derivative may sometimes prove handy.
The line derivative in direction v
is also called the Gateaux derivative in direction v
,
although the term "Gateaux derivative" is sometimes reserved for the situation where there is
such a derivative in all directions, for the map v β¦ lineDeriv π f x v
(which doesn't have to be
linear in general).
Main definition and results #
We mimic the definitions and statements for the FrΓ©chet derivative and the one-dimensional derivative. We define in particular the following objects:
LineDifferentiableWithinAt π f s x v
LineDifferentiableAt π f x v
HasLineDerivWithinAt π f f' s x v
HasLineDerivAt π f s x v
lineDerivWithin π f s x v
lineDeriv π f x v
and develop about them a basic API inspired by the one for the FrΓ©chet derivative.
We depart from the FrΓ©chet derivative in two places, as the dependence of the following predicates on the direction would make them barely usable:
- We do not define an analogue of the predicate
UniqueDiffOn
; - We do not define
LineDifferentiableOn
norLineDifferentiable
.
Results that do not rely on a topological structure on E
f
has the derivative f'
at the point x
along the direction v
in the set s
.
That is, f (x + t v) = f x + t β’ f' + o (t)
when t
tends to 0
and x + t v β s
.
Note that this definition is less well behaved than the total FrΓ©chet derivative, which
should generally be favored over this one.
Equations
- HasLineDerivWithinAt π f f' s x v = HasDerivWithinAt (fun (t : π) => f (x + t β’ v)) f' ((fun (t : π) => x + t β’ v) β»ΒΉ' s) 0
Instances For
f
has the derivative f'
at the point x
along the direction v
.
That is, f (x + t v) = f x + t β’ f' + o (t)
when t
tends to 0
.
Note that this definition is less well behaved than the total FrΓ©chet derivative, which
should generally be favored over this one.
Equations
- HasLineDerivAt π f f' x v = HasDerivAt (fun (t : π) => f (x + t β’ v)) f' 0
Instances For
f
is line-differentiable at the point x
in the direction v
in the set s
if there
exists f'
such that f (x + t v) = f x + t β’ f' + o (t)
when t
tends to 0
and x + t v β s
.
Equations
- LineDifferentiableWithinAt π f s x v = DifferentiableWithinAt π (fun (t : π) => f (x + t β’ v)) ((fun (t : π) => x + t β’ v) β»ΒΉ' s) 0
Instances For
f
is line-differentiable at the point x
in the direction v
if there
exists f'
such that f (x + t v) = f x + t β’ f' + o (t)
when t
tends to 0
.
Equations
- LineDifferentiableAt π f x v = DifferentiableAt π (fun (t : π) => f (x + t β’ v)) 0
Instances For
Line derivative of f
at the point x
in the direction v
within the set s
, if it exists.
Zero otherwise.
If the line derivative exists (i.e., β f', HasLineDerivWithinAt π f f' s x v
), then
f (x + t v) = f x + t lineDerivWithin π f s x v + o (t)
when t
tends to 0
and x + t v β s
.
Equations
- lineDerivWithin π f s x v = derivWithin (fun (t : π) => f (x + t β’ v)) ((fun (t : π) => x + t β’ v) β»ΒΉ' s) 0
Instances For
Line derivative of f
at the point x
in the direction v
, if it exists. Zero otherwise.
If the line derivative exists (i.e., β f', HasLineDerivAt π f f' x v
), then
f (x + t v) = f x + t lineDeriv π f x v + o (t)
when t
tends to 0
.
Instances For
Alias of the forward direction of hasLineDerivAt_iff_tendsto_slope_zero
.
Results that need a normed space structure on E
Converse to the mean value inequality: if f
is line differentiable at xβ
and C
-lipschitz
on a neighborhood of xβ
then its line derivative at xβ
in the direction v
has norm
bounded by C * βvβ
. This version only assumes that βf x - f xββ β€ C * βx - xββ
in a
neighborhood of x
.
Converse to the mean value inequality: if f
is line differentiable at xβ
and C
-lipschitz
on a neighborhood of xβ
then its line derivative at xβ
in the direction v
has norm
bounded by C * βvβ
. This version only assumes that βf x - f xββ β€ C * βx - xββ
in a
neighborhood of x
.
Converse to the mean value inequality: if f
is line differentiable at xβ
and C
-lipschitz
then its line derivative at xβ
in the direction v
has norm bounded by C * βvβ
.
Converse to the mean value inequality: if f
is C
-lipschitz
on a neighborhood of xβ
then its line derivative at xβ
in the direction v
has norm
bounded by C * βvβ
. This version only assumes that βf x - f xββ β€ C * βx - xββ
in a
neighborhood of x
.
Version using lineDeriv
.
Converse to the mean value inequality: if f
is C
-lipschitz on a neighborhood of xβ
then its line derivative at xβ
in the direction v
has norm bounded by C * βvβ
.
Version using lineDeriv
.
Converse to the mean value inequality: if f
is C
-lipschitz then
its line derivative at xβ
in the direction v
has norm bounded by C * βvβ
.
Version using lineDeriv
.