Jacobian matrix and determinant

From Wikipedia, the free encyclopedia - View original article

 
  (Redirected from Jacobian)
Jump to: navigation, search

In vector calculus, the Jacobian matrix (pron.: /ɨˈkbiən/, /jɨˈkbiən/) is the matrix of all first-order partial derivatives of a vector- or scalar-valued function with respect to another vector. Suppose F : \mathbb{R}^n \rightarrow \mathbb{R}^m is a function from Euclidean n-space to Euclidean m-space. Such a function is given by m real-valued component functions, F_1(x_1,\ldots,x_n),\ldots,F_m(x_1,\ldots,x_n). The partial derivatives of all these functions (if they exist) can be organized in an m-by-n matrix, the Jacobian matrix J of F, as follows:

J=\begin{bmatrix} \dfrac{\partial F_1}{\partial x_1} & \cdots & \dfrac{\partial F_1}{\partial x_n} \\ \vdots & \ddots & \vdots \\ \dfrac{\partial F_m}{\partial x_1} & \cdots & \dfrac{\partial F_m}{\partial x_n}  \end{bmatrix}.

This matrix is also denoted by J_F(x_1,\ldots,x_n) and \frac{\partial(F_1,\ldots,F_m)}{\partial(x_1,\ldots,x_n)}. If (x_1, \ldots , x_n) are the usual orthogonal Cartesian coordinates, the i th row (i = 1, ..., m) of this matrix corresponds to the gradient of the ith component function Fi: \left(\nabla F_i\right). Note that some books define the Jacobian as the transpose of the matrix given above.

The Jacobian determinant (often simply called the Jacobian) is the determinant of the Jacobian matrix (if m=n).

These concepts are named after the mathematician Carl Gustav Jacob Jacobi.

Contents

Jacobian matrix

The Jacobian of a function describes the orientation of a tangent plane to the function at a given point. In this way, the Jacobian generalizes the gradient of a scalar valued function of multiple variables which itself generalizes the derivative of a scalar-valued function of a scalar. In other words, the Jacobian for a scalar valued multivariable function is the gradient and that of a scalar valued function of scalar is simply its derivative. Likewise, the Jacobian can also be thought of as describing the amount of "stretching" that a transformation imposes. For example, if (x_2,y_2)=f(x_1,y_1) is used to transform an image, the Jacobian of f, J(x_1,y_1) describes how much the image in the neighborhood of (x_1,y_1) is stretched in the x and y directions.

If a function is differentiable at a point, its derivative is given in coordinates by the Jacobian, but a function doesn't need to be differentiable for the Jacobian to be defined, since only the partial derivatives are required to exist.

The importance of the Jacobian lies in the fact that it is a factor in one term of the best linear approximation to a differentiable function near a given point. In this sense, the Jacobian is the derivative of a multivariate function.

If p is a point in Rn and F is differentiable at p, then its derivative is given by JF(p). In this case, the linear map described by JF(p) is the best linear approximation of F near the point p, in the sense that

F(\mathbf{x}) = F(\mathbf{p}) + J_F(\mathbf{p})(\mathbf{x}-\mathbf{p}) + o(\|\mathbf{x}-\mathbf{p}\|)

for x close to p and where o is the little o-notation (for x\to p) and \|\mathbf{x}-\mathbf{p}\| is the distance between x and p.

Compare this to a Taylor series for a scalar function of a scalar argument, truncated to first order:

f(x) = f(p) + f'(p) ( x - p ) + o(x-p).

In a sense, both the gradient and Jacobian are "first derivatives" — the former the first derivative of a scalar function of several variables, the latter the first derivative of a vector function of several variables. In general, the gradient can be regarded as a special version of the Jacobian: it is the Jacobian of a scalar function of several variables.

The Jacobian of the gradient has a special name: the Hessian matrix, which in a sense is the "second derivative" of the scalar function of several variables in question.

Inverse

According to the inverse function theorem, the matrix inverse of the Jacobian matrix of an invertible function is the Jacobian matrix of the inverse function. That is, for some function F : RnRn and a point p in Rn,

 J(F^{-1}(p)) = [ J(F(p)) ]^{-1}.\

It follows that the (scalar) inverse of the Jacobian determinant of a transformation is the Jacobian determinant of the inverse transformation.

Uses

Dynamical systems

Consider a dynamical system of the form x' = F(x), where x' is the (component-wise) time derivative of x, and F : RnRn is continuous and differentiable. If F(x0) = 0, then x0 is a stationary point (also called a critical point, not to be confused with a fixed point). The behavior of the system near a stationary point is related to the eigenvalues of JF(x0), the Jacobian of F at the stationary point.[1] Specifically, if the eigenvalues all have real part with a magnitude less than 0, then the system is stable in the operating point, if any eigenvalue has a real part with a magnitude greater than 0, then the point is unstable. If the largest real part of the eigenvalues is equal to 0, the Jacobian matrix does not allow for an evaluation of the stability.

Image Jacobian

In computer vision the image Jacobian is known as the relationship between the movement of the camera and the apparent motion of the image (Optical Flow).

A point in the space with 3D coordinates X=\left( X,\text{ }Y,\text{ }Z \right)\text{ } in the frame of the camera, is projected in the image as a 2D point with coordinates x=\left( x,\text{ }y \right) , the relation between them is (neglecting the intrinsic parameters of the camera, and assuming focal distance 1) :

\left\{ \begin{matrix}    x=\frac{X}{Z}\\    y=\frac{Y}{Z} \\ \end{matrix} \right.

Differentiating this

\left\{ \begin{matrix}    \dot{x}=\frac{{\dot{X}}}{Z}-\frac{X\dot{Z}}{Z^{2}}=\frac{\dot{X}-x\dot{Z}}{Z}  \\    \dot{y}=\frac{{\dot{Y}}}{Z}-\frac{Y\dot{Z}}{Z^{2}}=\frac{\dot{Y}-y\dot{Z}}{Z}  \\ \end{matrix}\right.\qquad\qquad\dot{X}=-v_{c}-\omega _{c}\times X\Leftrightarrow \left\{ \begin{matrix}    \dot{X}=-v_{x}-\omega _{y}Z+\omega _{z}Y  \\    \dot{Y}=-v_{y}-\omega _{z}X+\omega _{x}Z  \\    \dot{Z}=-v_{z}-\omega _{x}Y+\omega _{y}X  \\ \end{matrix}\right.


Grouping these equations

\left\{ \begin{matrix}    \dot{x}=\frac{-v_{x}}{Z}+\frac{xv_{z}}{Z}+xy\omega _{x}-(1+x^{2})\omega _{y}+y\omega _{z}  \\    \dot{y}=\frac{-v_{y}}{Z}+\frac{yv_{z}}{Z}+(1+y^{2})\omega _{x}-xy\omega _{y}-x\omega _{z}  \\ \end{matrix} \right.\ \ \ \ \ \,\,\,\,\,\,\,\,\,\, \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,


Finally for a given pixel in the image with coordinates :(x,y) the apparent motion[2]  :(u,v)

\begin{bmatrix} u\\v \end{bmatrix} = \begin{bmatrix} -\frac{1}{Z}  & 0 &  \frac{x}{Z} & xy  & -(x^2+1)      &y\\ 0  & -\frac{1}{Z} &  \frac{y}{Z} & y^2+1 & -xy     &-x\\ \end{bmatrix} \begin{bmatrix} V_x  \\ V_y  \\ V_z \\ \omega_x \\ \omega_y \\ \omega_z \\ \end{bmatrix}

Newton's method

A system of coupled nonlinear equations can be solved iteratively by Newton's method. This method uses the Jacobian matrix of the system of equations.

The following is the detail code in MATLAB (although there is a built in 'jacobian' command)

    function s = jacobian(f, x, tol)    % f is a multivariable function handle, x is a starting point    if nargin == 2        tol = 10^(-5);    end    while 1        % if x and f(x) are row vectors, we need transpose operations here        y = x' - jacob(f, x)\f(x)';             % get the next point        if norm(f(y))<tol                       % check error tolerate            s = y';            return;        end        x = y';    end   
    function j = jacob(f, x)            % approximately calculate Jacobian matrix    k = length(x);    j = zeros(k, k);    x2 = x;    dx = 0.001;    for m = 1: k         x2(m) = x(m)+dx;        j(m, :) = (f(x2)-f(x))/dx;    % partial derivatives in m-th row        x2(m) = x(m);    end 

Jacobian determinant

If m = n, then F is a function from n-space to n-space and the Jacobian matrix is a square matrix. We can then form its determinant, known as the Jacobian determinant. The Jacobian determinant is sometimes simply called "the Jacobian."

The Jacobian determinant at a given point gives important information about the behavior of F near that point. For instance, the continuously differentiable function F is invertible near a point pRn if the Jacobian determinant at p is non-zero. This is the inverse function theorem. Furthermore, if the Jacobian determinant at p is positive, then F preserves orientation near p; if it is negative, F reverses orientation. The absolute value of the Jacobian determinant at p gives us the factor by which the function F expands or shrinks volumes near p; this is why it occurs in the general substitution rule.

Uses

The Jacobian determinant is used when making a change of variables when evaluating a multiple integral of a function over a region within its domain. To accommodate for the change of coordinates the magnitude of the Jacobian determinant arises as a multiplicative factor within the integral. Normally it is required that the change of coordinates be done in a manner which maintains an injectivity between the coordinates that determine the domain. Similarly, the Jacobian determinant represents the factor by which volumes change when space is distorted according to some function. The Jacobian determinant, as a result, is usually well-defined. The Jacobian can also be used to solve systems of differential equations at an equilibrium point or approximate solutions near an equilibrium point.

Examples

Example 1. The transformation from spherical coordinates (r, θ, φ) to Cartesian coordinates (x1, x2, x3), is given by the function F : R+ × [0,π] × [0,2π) → R3 with components:

 x_1 = r\, \sin\theta\, \cos\phi \,
 x_2 = r\, \sin\theta\, \sin\phi \,
 x_3 = r\, \cos\theta. \,

The Jacobian matrix for this coordinate change is

J_F(r,\theta,\phi) =\begin{bmatrix} \dfrac{\partial x_1}{\partial r} & \dfrac{\partial x_1}{\partial \theta} & \dfrac{\partial x_1}{\partial \phi} \\[3pt] \dfrac{\partial x_2}{\partial r} & \dfrac{\partial x_2}{\partial \theta} & \dfrac{\partial x_2}{\partial \phi} \\[3pt] \dfrac{\partial x_3}{\partial r} & \dfrac{\partial x_3}{\partial \theta} & \dfrac{\partial x_3}{\partial \phi} \\ \end{bmatrix}=\begin{bmatrix}  	\sin\theta\, \cos\phi &  r\, \cos\theta\, \cos\phi  & -r\, \sin\theta\, \sin\phi \\ 	\sin\theta\, \sin\phi &  r\, \cos\theta\, \sin\phi  &  r\, \sin\theta\, \cos\phi \\  	\cos\theta            &  -r\, \sin\theta            &             0                                \end{bmatrix}.

The determinant is r2 sin θ. As an example, since dV = dx1 dx2 dx3 this determinant implies that the differential volume element dV = r2 sin θ dr . Nevertheless this determinant varies with coordinates. To avoid any variation the new coordinates can be defined as w_{1}=\frac{r^{3}}{3},\ w_{2}=-\cos\theta,\ w_{3}=\phi.\, [3] Now the determinant equals 1 and volume element becomes r^{2}dr\ \sin\theta\ d\theta\ d\phi=dw_{1}dw_{2}dw_{3}\,.

Example 2. The Jacobian matrix of the function F : R3R4 with components

 y_1 = x_1 \,
 y_2 = 5x_3 \,
 y_3 = 4x_2^2 - 2x_3 \,
 y_4 = x_3 \sin(x_1) \,

is

J_F(x_1,x_2,x_3) =\begin{bmatrix} \dfrac{\partial y_1}{\partial x_1} & \dfrac{\partial y_1}{\partial x_2} & \dfrac{\partial y_1}{\partial x_3} \\[3pt] \dfrac{\partial y_2}{\partial x_1} & \dfrac{\partial y_2}{\partial x_2} & \dfrac{\partial y_2}{\partial x_3} \\[3pt] \dfrac{\partial y_3}{\partial x_1} & \dfrac{\partial y_3}{\partial x_2} & \dfrac{\partial y_3}{\partial x_3} \\[3pt] \dfrac{\partial y_4}{\partial x_1} & \dfrac{\partial y_4}{\partial x_2} & \dfrac{\partial y_4}{\partial x_3} \\ \end{bmatrix}=\begin{bmatrix} 1 & 0 & 0 \\ 0 & 0 & 5 \\ 0 & 8x_2 & -2 \\ x_3\cos(x_1) & 0 & \sin(x_1) \end{bmatrix}.

This example shows that the Jacobian need not be a square matrix.

Example 3.

x\,=r\,\cos\,\phi;
y\,=r\,\sin\,\phi.

J(r,\phi)=\begin{bmatrix} {\partial x\over\partial r} & {\partial x\over \partial\phi}   \\ {\partial y\over \partial r} & {\partial y\over \partial\phi}   \end{bmatrix}=\begin{bmatrix} {\partial (r\cos\phi)\over \partial r} & {\partial (r\cos\phi)\over \partial \phi}   \\ {\partial(r\sin\phi)\over \partial r} & {\partial (r\sin\phi)\over \partial\phi}   \end{bmatrix}=\begin{bmatrix} \cos\phi & -r\sin\phi   \\ \sin\phi & r\cos\phi   \end{bmatrix}

The Jacobian determinant is equal to r. This shows how an integral in the Cartesian coordinate system is transformed into an integral in the polar coordinate system:

\iint_A dx\, dy= \iint_B r \,dr\, d\phi.

Example 4. The Jacobian determinant of the function F : R3R3 with components

\begin{align}   y_1 &= 5x_2 \\   y_2 &= 4x_1^2 - 2 \sin (x_2x_3) \\   y_3 &= x_2 x_3 \end{align}

is

\begin{vmatrix}   0 & 5 & 0 \\   8 x_1 & -2 x_3 \cos(x_2 x_3) & -2x_2\cos(x_2 x_3) \\   0 & x_3 & x_2 \end{vmatrix} = -8 x_1 \cdot \begin{vmatrix}   5 & 0 \\   x_3 & x_2 \end{vmatrix} = -40 x_1 x_2.

From this we see that F reverses orientation near those points where x1 and x2 have the same sign; the function is locally invertible everywhere except near points where x1 = 0 or x2 = 0. Intuitively, if you start with a tiny object around the point (1,1,1) and apply F to that object, you will get an object set with approximately 40 times the volume of the original one.

See also

Notes

  1. ^ D.K. Arrowsmith and C.M. Place, Dynamical Systems, Section 3.3, Chapman & Hall, London, 1992. ISBN 0-412-39080-9.
  2. ^ Fermín, Leonardo; Medina et al (2009). "Estimation of Velocities Components using Optical Flow and Inner Product". Lecture Notes in Computer Science 396: 349-358.
  3. ^ Taken from http://www.sjcrothers.plasmaresources.com/schwarzschild.pdf - On the Gravitational Field of a Mass Point according to Einstein’s Theory by K. Schwarzschild - arXiv:physics/9905030 v1 (text of the original paper, in Wikisource).

External links