Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 21 views

    Randomfield KL Cond

    Uploaded by

    Slim Salim
    The document introduces random fields and Gaussian random fields. It defines a random field as a stochastic process indexed by a spatial variable. A Gaussian random field is characterized by its covariance function and forms an infinite-dimensional vector space. The derivative of a Gaussian field is again a Gaussian field whose covariance function is related to the second partial derivative of the original covariance function. Integrals of random fields can be defined as limits of Riemann sums, and convolutions of kernels and random fields produce new random fields.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Randomfield KL Cond

    Uploaded by

    Slim Salim
    21 views3 pages
    The document introduces random fields and Gaussian random fields. It defines a random field as a stochastic process indexed by a spatial variable. A Gaussian random field is characterized by its covariance function and forms an infinite-dimensional vector space. The derivative of a Gaussian field is again a Gaussian field whose covariance function is related to the second partial derivative of the original covariance function. Integrals of random fields can be defined as limits of Riemann sums, and convolutions of kernels and random fields produce new random fields.

    Original Description:

    random field

    Original Title

    Randomfield Kl Cond

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document introduces random fields and Gaussian random fields. It defines a random field as a stochastic process indexed by a spatial variable. A Gaussian random field is characterized by its covariance function and forms an infinite-dimensional vector space. The derivative of a Gaussian field is again a Gaussian field whose covariance function is related to the second partial derivative of the original covariance function. Integrals of random fields can be defined as limits of Riemann sums, and convolutions of kernels and random fields produce new random fields.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    21 views3 pages

    Randomfield KL Cond

    Uploaded by

    Slim Salim
    The document introduces random fields and Gaussian random fields. It defines a random field as a stochastic process indexed by a spatial variable. A Gaussian random field is characterized by its covariance function and forms an infinite-dimensional vector space. The derivative of a Gaussian field is again a Gaussian field whose covariance function is related to the second partial derivative of the original covariance function. Integrals of random fields can be defined as limits of Riemann sums, and convolutions of kernels and random fields produce new random fields.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 3

    Introduction to Random Fields.

    Moo K. Chung
    mchung@stat.wisc.edu

    February 1, 2007

    Suppose we measure temperature Y at position x and time t in a classroom


    M ∈ R3 . Since every measurement will be error-prone, we model the temperature
    as
    Y (x, t) = µ(x, t) + ǫ(x, t)
    where µ is the unknown signal and ǫ is the measurement error. The measurement
    error can be modelled as a random variable. So at each point (x, t) ∈ M ⊗ R+ ,
    measurement error ǫ(x, t) is a random variable. The collection of random variables

    {ǫ(x, t) : (x, t) ∈ M ⊗ R+ }

    is called a stochastic process. A stochastic process that is indexed by a spatial


    variable is called a random field. A formal measure theoretic definition can be
    found in Geometry of Random Fields by Adler (1980) and Introduction to the
    Theory of Random Processes by Gikhman and Skorokhod (1969).

    1 Gaussian random fields


    P
    A random vector X = (X1 , · · · , Xm ) is multivariate normal if i ci Xi is Gaus-
    sian for every possible choice of ci . A random fields ǫ(x) ∈ RN is a Gaussian
    random field if ǫ1 (x1 ), · · · , ǫm (xm ) is multivariate normal for any xi ∈ M ⊂ RN .
    ǫ is a mean zero Gaussian field if Eǫ(x) = 0 for all x. The covariance function of
    mean zero field is defined as

    R(x, y) = Eǫ(x)ǫ(y).

    The variance field of ǫ is given by R(x, x). Mean zero Gaussian field is com-
    pletely characterized by the covariance function. A Gaussian random vector field
    is defined similarly.
    e(x) = (e1 (x), · · · , en (x))′

    1
    is a n-dimensional Gaussian vector field if ei are Gaussian fields. We may gener-
    alize it further to matrix and tensor fields.
    Two fields e1 and e2 are independent if e1 (x) and e2 (y) are independent for
    every x and y. For mean zero Gaussian fields, e1 and e2 are independent if and
    only if the cross-covariance function
    R(x, y) = Ee1 (x)e2 (y)
    vanishes for all x and y.
    Let G be a collection of Gaussian random fields. Suppose e1 , e2 ∈ G. forms
    a vector space if c1 e1 + c2 e2 ∈ G for all c1 and c2 . It is trivial to see that G is
    an infinite-dimensional vector space. However, there exists a finite vector space
    Gp that is the closest to G in the least-squares sense. For any linear operator f ,
    f (G) ⊂ L. We show this for the derivatives of Gaussian fields.

    2 Derivative Fields
    The sequence of random fields Xh (t) converges to X(t) as h → 0 in mean-square
    if
    lim E|Xh (t) − X(t)|2 = 0.
    h→0
    We denote it by limh→0 Xh (t) = X(t) if there is no ambiguity. The convergence
    in the mean square implies the convergence in mean. Note that

    2  2
    E|Xh (t) − X(t)|2 = −Var Xh (t) − X(t) − EXh (t) − X(t) .
    

    Now let Xh → X in mean squre. The left hand side is positive and converges to
    zero while the right hand side negative. Hence each term in the right hand side
    should converges to zero.
    We define derivative in mean square as
    dX(t) X(t + h) − X(t)
    = lim .
    dt h→0 h
    Note that if X(t) and X(t + h) are Gaussian random fields, X(x + h) − X(t) is
    again Gaussian and hence the limit on the right hand side is again Gaussian. If R is
    the covariance function of the mean zero Gaussian field X, the covariance function
    of its derivative is given by
    h dX(t) dX(s) i ∂ 2 R(t, s)
    E = .
    dt ds ∂t∂s
    Partial derivatives are defined similarly.

    2
    3 Integration of Fields
    We define the integration of a random field as the limit of Riemann sum of the field.
    Let ∪ni=1 Ωi be a partition of Ω ⊂ Rn , i.e. Ω = ∪ni=1 Ωi and Ωi ∩ Ωj = ∅ if i 6= j.
    Let ti ∈ Ωi and µ(Ωi ) be the volume of Ωi . Then we define the integration of a
    random field as
    Z n
    X
    X(t) dt = lim X(ti )µ(Ωi ).
    Ω all µ(Ωj )→0 i=1
    Multiple integration is defined similarly. When we integrate a Gaussian field,
    it is the limit of a linear combination of Gaussian random variables so it is again
    Gaussian.
    Consider the following integral
    Z
    Y (t) = K(t, s)X(s) ds.

    where K is called the kernel of the integral. Define convolution between kernel K
    and random field X as the above integral.

    Y (t) = K ∗ X(t). (1)

    Suppose
    R the kernel to be isotropic probability density, i.e. K(t, s) = K(t −
    s) and K(t) dt = 1. Further we may assume K to be unimodal with some
    parameter σ such that
    lim K(t; σ) → δ(t),
    σ→0

    the Dirac-delta function. The Dirac-delta function is defined as a function that


    satisfies
    Z
    δ(t) = 0, t 6= 0 and δ(t) dt = 1.

    Since the Dirac-delta function satisfies


    Z
    δ(t − s)f (s) ds = f (t),

    it can be easily seen that


    lim K(·; σ) ∗ X → X.
    σ→0

    Theory on linear filter will continue.

    You might also like