PhyslibSearch

Physlib.Mathematics.Calculus.AdjFDeriv

32 declarations

definition

Adjoint Fréchet derivative (Df(x))(dy)(Df(x))^*(dy)

#adjFDeriv

Given a function f:EFf: E \to F between two inner product spaces EE and FF over a field k\mathbb{k}, let Df(x):EFDf(x): E \to F denote the Fréchet derivative of ff at the point xEx \in E. The adjoint Fréchet derivative at xx is the adjoint of this linear map, denoted (Df(x)):FE(Df(x))^*: F \to E. For a vector dyFdy \in F, the function returns the value (Df(x))(dy)E(Df(x))^*(dy) \in E, which is the unique vector satisfying Df(x)v,dyF=v,(Df(x))dyE\langle Df(x) \cdot v, dy \rangle_F = \langle v, (Df(x))^* \cdot dy \rangle_E for all vEv \in E.

theorem

adjFDeriv f x=f\text{adjFDeriv } f \ x = f' if ff has an adjoint Fréchet derivative ff' at xx

#adjFDeriv

Let EE and FF be inner product spaces over a field k\mathbb{k}. For a function f:EFf: E \to F and a point xEx \in E, if ff has an adjoint Fréchet derivative ff' at xx (expressed by the proposition `HasAdjFDerivAt 𝕜 f f' x`), then the defined adjoint Fréchet derivative adjFDeriv k f x\text{adjFDeriv } \mathbb{k} \ f \ x is equal to ff'.

theorem

Differentiability at xx implies existence of adjoint Fréchet derivative (Df(x))(Df(x))^*

#hasAdjFDerivAt

Let EE and FF be complete generalized inner product spaces over a field k\mathbb{k}. If a function f:EFf: E \to F is Fréchet differentiable at a point xEx \in E, then ff possesses an adjoint Fréchet derivative at xx, which is given by the adjoint of its Fréchet derivative, denoted as adjFDeriv k f x=(Df(x))\text{adjFDeriv } \mathbb{k} \ f \ x = (Df(x))^*.

theorem

The mapping (L,y)Ly(L, y) \mapsto L^* y is a bounded bilinear map over R\mathbb{R}

#isBoundedBilinearMap_real

Let XX and YY be real inner product spaces. The map that sends a pair (L,y)L(X,Y)×Y(L, y) \in \mathcal{L}(X, Y) \times Y to LyL^* y, where L:YXL^*: Y \to X is the adjoint of the continuous linear map LL, is a bounded bilinear map over R\mathbb{R}.

theorem

fCn+1    ((x,y,dz)(Dyf(x,y))dz)Cnf \in C^{n+1} \implies ((x, y, dz) \mapsto (D_y f(x, y))^* dz) \in C^n

#contDiffAt_deriv

Let EE be a real normed space, and let FF and GG be complete real generalized inner product spaces. Suppose f:E×FGf: E \times F \to G is a function such that for every xEx \in E and yFy \in F, the partial map fx:yf(x,y)f_x: y \mapsto f(x, y) has an adjoint Fréchet derivative f(x,y):GFf'(x, y): G \to F at yy. If ff is Cn+1C^{n+1} as a function of (x,y)(x, y), then the map (x,y,dz)f(x,y)dz(x, y, dz) \mapsto f'(x, y) \cdot dz from E×F×GE \times F \times G to FF is CnC^n. Here, the adjoint Fréchet derivative f(x,y)f'(x, y) is the adjoint of the linear map Dyf(x,y):FGD_y f(x, y): F \to G, satisfying Dyf(x,y)v,dzG=v,f(x,y)dzF\langle D_y f(x, y) \cdot v, dz \rangle_G = \langle v, f'(x, y) \cdot dz \rangle_F for all vFv \in F and dzGdz \in G.

theorem

f(x)=(Df(x))(1)\nabla f(x) = (Df(x))^*(1)

#gradient_eq_adjFDeriv

Let UU be an inner product space over a field k\mathbb{k} (where k\mathbb{k} is typically R\mathbb{R} or C\mathbb{C}). For any function f:Ukf: U \to \mathbb{k} that is Fréchet differentiable at a point xUx \in U, the gradient f(x)\nabla f(x) is equal to the adjoint Fréchet derivative of ff at xx applied to the unit 1k1 \in \mathbb{k}. That is, f(x)=(Df(x))(1)\nabla f(x) = (Df(x))^*(1), where (Df(x))(Df(x))^* denotes the adjoint of the Fréchet derivative Df(x):UkDf(x): U \to \mathbb{k}.

theorem

The identity map has the identity as its adjoint Fréchet derivative at any point xx

#hasAdjFDerivAt_id

Let EE be an inner product space over a field k\mathbb{k}. For any xEx \in E, the identity function f:EEf : E \to E defined by f(x)=xf(x) = x has an adjoint Fréchet derivative at xx which is the identity function f(dx)=dxf'(dx) = dx.

theorem

The Adjoint Fréchet Derivative of the Identity Function is the Identity Map

#adjFDeriv_id

Let EE be an inner product space over a field k\mathbb{k}. The adjoint Fréchet derivative of the identity function f:EEf: E \to E, defined by f(x)=xf(x) = x, is the identity map at every point. Specifically, for any point xEx \in E and any vector dxEdx \in E, the adjoint Fréchet derivative satisfies (adjFDeriv k f x) dx=dx(\text{adjFDeriv } \mathbb{k} \ f \ x) \ dx = dx.

theorem

The Adjoint Fréchet Derivative of the Identity Map is the Identity Map

#adjFDeriv_id'

Let EE be an inner product space over a field k\mathbb{k}. The adjoint Fréchet derivative of the identity function id:EE\text{id} : E \to E is the identity map at every point. Specifically, for any point xEx \in E and any vector dxEdx \in E, the adjoint Fréchet derivative satisfies (adjFDeriv k id x) dx=dx(\text{adjFDeriv } \mathbb{k} \ \text{id} \ x) \ dx = dx.

theorem

The adjoint Fréchet derivative of a constant function is zero

#hasAdjFDerivAt_const

Let EE and FF be inner product spaces over a field k\mathbb{k}. For any point xEx \in E and any value yFy \in F, the constant function f:EFf: E \to F defined by f(z)=yf(z) = y has an adjoint Fréchet derivative at xx given by the zero linear map dz0dz \mapsto 0.

theorem

The adjoint Fréchet derivative of a constant function is zero

#adjFDeriv_const

Let EE and FF be inner product spaces over a field k\mathbb{k}. For any constant yFy \in F, let f:EFf: E \to F be the constant function f(x)=yf(x) = y. The adjoint Fréchet derivative of ff is zero at every point xEx \in E, meaning that for all xEx \in E and all vectors dyFdy \in F, (adjFDeriv k f x)(dy)=0(\text{adjFDeriv } \mathbb{k} \ f \ x)(dy) = 0.

theorem

Chain Rule for Adjoint Fréchet Derivatives

#comp

Let E,FE, F, and GG be inner product spaces over a field k\mathbb{k}. Let g:EFg: E \to F and f:FGf: F \to G be functions. Suppose gg has an adjoint Fréchet derivative gg' at a point xEx \in E, and ff has an adjoint Fréchet derivative ff' at the point g(x)g(x). Then the composition fgf \circ g has an adjoint Fréchet derivative at xx given by the composition of the adjoint derivatives gfg' \circ f'.

theorem

Chain Rule for Adjoint Fréchet Derivatives: (D(fg)(x))=(Dg(x))(Df(g(x)))(D(f \circ g)(x))^* = (Dg(x))^* \circ (Df(g(x)))^*

#adjFDeriv_comp

Let E,FE, F, and GG be complete inner product spaces over a field k\mathbb{k}. Suppose g:EFg: E \to F is a function differentiable at a point xEx \in E and f:FGf: F \to G is a function differentiable at g(x)g(x). The adjoint Fréchet derivative of the composition fgf \circ g at xx is the composition of the adjoint Fréchet derivatives of gg at xx and ff at g(x)g(x). That is, for any dyGdy \in G: \[ (\text{adjFDeriv } \mathbb{k} \ (f \circ g) \ x)(dy) = (\text{adjFDeriv } \mathbb{k} \ g \ x)((\text{adjFDeriv } \mathbb{k} \ f \ (g(x)))(dy)) \] In terms of the adjoint of the Fréchet derivative operator DD, this is expressed as: \[ (D(f \circ g)(x))^* = (Dg(x))^* \circ (Df(g(x)))^* \]

theorem

The adjoint Fréchet derivative of x(f(x),g(x))x \mapsto (f(x), g(x)) is (dy,dz)f(dy)+g(dz)(dy, dz) \mapsto f'(dy) + g'(dz)

#prodMk

Let E,FE, F, and GG be inner product spaces over a field k\mathbb{k}. Suppose f:EFf: E \to F and g:EGg: E \to G are functions such that at a point xEx \in E, ff has an adjoint Fréchet derivative f:FEf': F \to E and gg has an adjoint Fréchet derivative g:GEg': G \to E. Then the map h:EF×Gh: E \to F \times G defined by h(x)=(f(x),g(x))h(x) = (f(x), g(x)) has an adjoint Fréchet derivative at xx given by the map (dy,dz)f(dy)+g(dz)(dy, dz) \mapsto f'(dy) + g'(dz). Here, the product space F×GF \times G is equipped with the standard product inner product (y1,z1),(y2,z2)=y1,y2F+z1,z2G\langle (y_1, z_1), (y_2, z_2) \rangle = \langle y_1, y_2 \rangle_F + \langle z_1, z_2 \rangle_G.

theorem

The adjoint Fréchet derivative of x(f(x))1x \mapsto (f(x))_1 is dyf(dy,0)dy \mapsto f'(dy, 0)

#fst

Let E,FE, F, and GG be generalized inner product spaces over a field k\mathbb{k}. If a function f:EF×Gf: E \to F \times G has an adjoint Fréchet derivative f:F×GEf' : F \times G \to E at a point xEx \in E, then the map x(f(x))1x \mapsto (f(x))_1 (the projection onto the first component) has an adjoint Fréchet derivative at xx given by the function dyf(dy,0)dy \mapsto f'(dy, 0).

theorem

The adjoint Fréchet derivative of x(f(x))1x \mapsto (f(x))_1 is dy(Df(x))(dy,0)dy \mapsto (Df(x))^*(dy, 0)

#adjFDeriv_fst

Let E,FE, F, and GG be complete generalized inner product spaces over a field k\mathbb{k}. For a function f:EF×Gf: E \to F \times G that is Fréchet differentiable at a point xEx \in E, the adjoint Fréchet derivative of its first component f1:EFf_1: E \to F (where f1(x)=(f(x))1f_1(x) = (f(x))_1) at xx is given by the linear map dy(Df(x))(dy,0)dy \mapsto (Df(x))^*(dy, 0). Here, (Df(x)):F×GE(Df(x))^*: F \times G \to E denotes the adjoint Fréchet derivative of ff at xx. The product space F×GF \times G is equipped with the standard product inner product (y1,z1),(y2,z2)=y1,y2F+z1,z2G\langle (y_1, z_1), (y_2, z_2) \rangle = \langle y_1, y_2 \rangle_F + \langle z_1, z_2 \rangle_G.

theorem

The adjoint Fréchet derivative of the first projection is a(a,0)a \mapsto (a, 0)

#adjFDeriv_prod_fst

Let EE and FF be complete generalized inner product spaces over a field k\mathbb{k}. For the projection map π1:F×EF\pi_1: F \times E \to F defined by π1(y,z)=y\pi_1(y, z) = y, its adjoint Fréchet derivative at any point xF×Ex \in F \times E is the linear map that sends aFa \in F to the pair (a,0)F×E(a, 0) \in F \times E.

theorem

The adjoint Fréchet derivative of x(f(x))2x \mapsto (f(x))_2 is dzf(0,dz)dz \mapsto f'(0, dz)

#snd

Let E,FE, F, and GG be inner product spaces over a field k\mathbb{k}. Let f:EF×Gf: E \to F \times G be a function and xEx \in E. If ff has an adjoint Fréchet derivative f:F×GEf': F \times G \to E at xx, then the map xπ2(f(x))x \mapsto \pi_2(f(x)) (the projection onto the second component) also has an adjoint Fréchet derivative at xx, which is given by the linear map dzf(0,dz)dz \mapsto f'(0, dz).

theorem

Adjoint Fréchet derivative of x(f(x))2x \mapsto (f(x))_2 is dy(Df(x))(0,dy)dy \mapsto (Df(x))^*(0, dy)

#adjFDeriv_snd

Let E,FE, F, and GG be complete generalized inner product spaces over a field k\mathbb{k}. Let f:EF×Gf: E \to F \times G be a function that is differentiable at a point xEx \in E. Then the adjoint Fréchet derivative of the map x(f(x))2x \mapsto (f(x))_2 (the projection of ff onto its second component) at xx is given by the linear map that sends dyGdy \in G to (Df(x))(0,dy)(Df(x))^*(0, dy), where (Df(x)):F×GE(Df(x))^*: F \times G \to E denotes the adjoint Fréchet derivative of ff at xx.

theorem

The adjoint Fréchet derivative of proj2\text{proj}_2 is a(0,a)a \mapsto (0, a)

#adjFDeriv_prod_snd

Let EE and FF be complete inner product spaces over a field k\mathbb{k}. The adjoint Fréchet derivative of the second projection map proj2:F×EE\text{proj}_2 : F \times E \to E, defined by proj2(y,z)=z\text{proj}_2(y, z) = z, at any point xF×Ex \in F \times E is the linear map from EE to F×EF \times E that sends aEa \in E to (0,a)(0, a).

theorem

The adjoint Fréchet derivative of f(x,y)f(x, y) is (fx,fy)(fx', fy')

#hasAdjFDerivAt_uncurry

Let E,FE, F, and GG be inner product spaces over a field k\mathbb{k}. Let f:EFGf: E \to F \to G be a function, and let (x,y)E×F(x, y) \in E \times F. Suppose that: 1. The uncurried function (x,y)f(x,y)(x, y) \mapsto f(x, y) is differentiable at (x,y)(x, y). 2. The partial map xf(x,y)x' \mapsto f(x', y) has an adjoint Fréchet derivative fx:GEfx': G \to E at xx. 3. The partial map yf(x,y)y' \mapsto f(x, y') has an adjoint Fréchet derivative fy:GFfy': G \to F at yy. Then the uncurried function (x,y)f(x,y)(x, y) \mapsto f(x, y) has an adjoint Fréchet derivative at (x,y)(x, y) given by the map dz(fx(dz),fy(dz))dz \mapsto (fx'(dz), fy'(dz)) from GG into E×FE \times F.

theorem

The adjoint Fréchet derivative of f(x,y)f(x, y) is the pair of partial adjoint derivatives ((D1f),(D2f))((D_1f)^*, (D_2f)^*)

#adjFDeriv_uncurry

Let E,FE, F, and GG be complete inner product spaces over a field k\mathbb{k}. Let f:EFGf: E \to F \to G be a function such that the uncurried function f^(x,y)=f(x,y)\hat{f}(x, y) = f(x, y) is differentiable at a point (x,y)E×F(x, y) \in E \times F. The adjoint Fréchet derivative of f^\hat{f} at (x,y)(x, y), denoted (Df^(x,y)):GE×F(D\hat{f}(x, y))^*: G \to E \times F, is given by: (Df^(x,y))(dz)=((D1f(x,y))(dz),(D2f(x,y))(dz))(D\hat{f}(x, y))^*(dz) = \left( (D_1 f(x, y))^*(dz), (D_2 f(x, y))^*(dz) \right) for all dzGdz \in G, where (D1f(x,y)):GE(D_1 f(x, y))^*: G \to E is the adjoint Fréchet derivative of the partial map xf(x,y)x' \mapsto f(x', y) at xx, and (D2f(x,y)):GF(D_2 f(x, y))^*: G \to F is the adjoint Fréchet derivative of the partial map yf(x,y)y' \mapsto f(x, y') at yy.

theorem

The adjoint Fréchet derivative of f-f is f-f'

#neg

Let EE and FF be inner product spaces over a field k\mathbb{k}. If a function f:EFf: E \to F has an adjoint Fréchet derivative ff' at a point xEx \in E, then the function xf(x)x \mapsto -f(x) also has an adjoint Fréchet derivative at xx, given by the map dyf(dy)dy \mapsto -f'(dy).

theorem

(D(f)(x))=(Df(x))(D(-f)(x))^* = -(Df(x))^*

#adjFDeriv_neg

Let EE and FF be complete inner product spaces over a field k\mathbb{k}. If a function f:EFf: E \to F is Fréchet differentiable at a point xEx \in E, then the adjoint Fréchet derivative of the negated function xf(x)x \mapsto -f(x) at xx is equal to the negation of the adjoint Fréchet derivative of ff at xx. In terms of the adjoint operator (Df(x))(Df(x))^*, this is expressed as: (D(f)(x))=(Df(x)) (D(-f)(x))^* = -(Df(x))^* Specifically, for any vector dyFdy \in F, the value is (Df(x))(dy)-(Df(x))^*(dy).

theorem

f+gf + g has adjoint Fréchet derivative f+gf' + g'

#add

Let EE and FF be inner product spaces over a field k\mathbb{k}. If f,g:EFf, g: E \to F are functions such that ff has an adjoint Fréchet derivative ff' at xEx \in E and gg has an adjoint Fréchet derivative gg' at xx, then the sum function f+gf + g (defined by xf(x)+g(x)x \mapsto f(x) + g(x)) has an adjoint Fréchet derivative at xx given by f+gf' + g' (defined by dyf(dy)+g(dy)dy \mapsto f'(dy) + g'(dy)).

theorem

(D(f+g)(x))=(Df(x))+(Dg(x))(D(f+g)(x))^* = (Df(x))^* + (Dg(x))^*

#adjFDeriv_add

Let EE and FF be complete inner product spaces over a field k\mathbb{k}. If f,g:EFf, g: E \to F are functions differentiable at xEx \in E, then the adjoint Fréchet derivative of their sum f+gf + g at xx is the sum of their individual adjoint Fréchet derivatives at xx. That is, \[ (D(f+g)(x))^* = (Df(x))^* + (Dg(x))^*, \] where (Df(x))(Df(x))^* denotes the adjoint of the Fréchet derivative of ff at xx.

theorem

The adjoint Fréchet derivative of fgf - g is fgf' - g'

#sub

Let EE and FF be inner product spaces over a field k\mathbb{k}. Suppose f,g:EFf, g: E \to F are functions that have adjoint Fréchet derivatives f,g:FEf', g': F \to E at a point xEx \in E, respectively. Then the function fgf - g (defined by xf(x)g(x)x \mapsto f(x) - g(x)) has an adjoint Fréchet derivative at xx given by fgf' - g' (defined by dyf(dy)g(dy)dy \mapsto f'(dy) - g'(dy)).

theorem

Adjoint Fréchet Derivative of fgf - g is (Df(x))(Dg(x))(Df(x))^* - (Dg(x))^*

#adjFDeriv_sub

Let EE and FF be complete inner product spaces over a field k\mathbb{k}. Suppose f,g:EFf, g: E \to F are functions that are Fréchet differentiable at a point xEx \in E. Then the adjoint Fréchet derivative of the difference fgf - g at xx is equal to the difference of the adjoint Fréchet derivatives of ff and gg at xx. That is, for any vector dyFdy \in F: adjFDeriv k (fg) x (dy)=adjFDeriv k f x (dy)adjFDeriv k g x (dy) \text{adjFDeriv } \mathbb{k} \ (f - g) \ x \ (dy) = \text{adjFDeriv } \mathbb{k} \ f \ x \ (dy) - \text{adjFDeriv } \mathbb{k} \ g \ x \ (dy) Using the notation (Df(x))(Df(x))^* for the adjoint of the Fréchet derivative, this can be written as: (D(fg)(x))=(Df(x))(Dg(x)) (D(f - g)(x))^* = (Df(x))^* - (Dg(x))^*

theorem

Adjoint Fréchet Derivative of the Product g(x)f(x)g(x)f(x)

#smul

Let EE and FF be generalized inner product spaces over a field k\mathbb{k} (where k\mathbb{k} is R\mathbb{R} or C\mathbb{C}). Suppose f:EFf: E \to F is a map with an adjoint Fréchet derivative ff' at xEx \in E, and g:Ekg: E \to \mathbb{k} is a map with an adjoint Fréchet derivative gg' at xx. Then the adjoint Fréchet derivative of the scalar-vector product xg(x)f(x)x \mapsto g(x)f(x) at the point xx is the linear map FEF \to E given by: dyg(x)f(dy)+g(dy,f(x)) dy \mapsto \overline{g(x)} f'(dy) + g'(\overline{\langle dy, f(x) \rangle}) where z\overline{z} denotes the conjugate of zkz \in \mathbb{k} and ,\langle \cdot, \cdot \rangle denotes the inner product on FF.

theorem

Adjoint Fréchet Derivative of the Product g(x)f(x)g(x)f(x)

#adjFDeriv_smul

Let EE and FF be complete inner product spaces over a field k\mathbb{k} (where k\mathbb{k} is R\mathbb{R} or C\mathbb{C}). If f:EFf: E \to F and g:Ekg: E \to \mathbb{k} are differentiable at a point xEx \in E, then the adjoint Fréchet derivative of the scalar-vector product xg(x)f(x)x \mapsto g(x)f(x) at xx is the linear map FEF \to E given by: dyg(x)(Df(x))(dy)+(Dg(x))(dy,f(x)) dy \mapsto \overline{g(x)} (Df(x))^*(dy) + (Dg(x))^*(\overline{\langle dy, f(x) \rangle}) where (Df(x))(Df(x))^* and (Dg(x))(Dg(x))^* denote the adjoint Fréchet derivatives of ff and gg at xx respectively, and ,\langle \cdot, \cdot \rangle denotes the inner product on FF.

theorem

Adjoint Fréchet Derivative of the Inner Product (x1,x2)x1,x2(x_1, x_2) \mapsto \langle x_1, x_2 \rangle

#inner

Let EE be a real generalized inner product space. For the inner product function f:E×ERf: E \times E \to \mathbb{R} defined by f(x1,x2)=x1,x2f(x_1, x_2) = \langle x_1, x_2 \rangle, the adjoint Fréchet derivative at any point x=(x1,x2)E×Ex = (x_1, x_2) \in E \times E is the linear map from R\mathbb{R} to E×EE \times E defined by: yy(x2,x1) y \mapsto y \cdot (x_2, x_1) for any yRy \in \mathbb{R}.

theorem

Adjoint Fréchet Derivative of Inner Product (x1,x2)x1,x2(x_1, x_2) \mapsto \langle x_1, x_2 \rangle is yy(x2,x1)y \mapsto y(x_2, x_1)

#adjFDeriv_inner

Let EE be a real generalized inner product space. For the inner product function f:E×ERf: E \times E \to \mathbb{R} defined by f(x1,x2)=x1,x2f(x_1, x_2) = \langle x_1, x_2 \rangle, the adjoint Fréchet derivative at any point x=(x1,x2)E×Ex = (x_1, x_2) \in E \times E is the linear map from R\mathbb{R} to E×EE \times E defined by: yy(x2,x1) y \mapsto y \cdot (x_2, x_1) for any yRy \in \mathbb{R}.