Using High Order Finite Differences/Definitions and Basics


Definitions and Basics

edit

Vector Norms

edit

Definition of a Vector Norm

edit

The most ordinary kind of vectors are those consisting of ordered n-tuples of real or complex numbers. They may be written in row                           

or column       

forms. Commas or other seperators of components or coordinates may or may not be used. When a vector has many elements, notations like

    or       are often used.

A most popular notation to indicate a vector is   .

Vectors are usually added component-wise, for

    and    ,

  .

Scalar multiplication is defined by    .

A vector norm is a generalization of ordinary absolute value     of a real or complex number.

For     and   vectors, and   ,  a scalar, a vector norm is a real value     associated with a vector for which the following properties hold.

 .

Common Vector Norms

edit

The most commonly used norms are:

 .

Any two norms on   dimensional vectors of complex numbers are topologically equivalent in the sense that, if     and     are two different norms, then there exist positive constants     and     such that   .

Inner Product of Vectors

edit

The inner product (or dot product), of two vectors

    and    ,

is defined by

  ,

or when     and     are complex valued, by

 

 .

(1.3.3.0)


It is often indicated by any one of several notations:

     or     .

Besides the dot product, other inner products are defined to be a rule that sssigns to each pair of vectors   ,  a complex number with the following properties.

 

 

 

for      is real valued and positive

 

and

 

An inner product defines a norm by


  .

(1.3.3.1)


Inequalities Involving Norms

edit

The Cauchy Schwarz and Holder's inequalities are commonly employed.

 

     for   

Matrix Norms

edit

Definition of a Matrix Norm

edit

The most ordinary kind of matices are those consisting of rectangular arrays of real or complex numbers. They may be written in element form             and be considered as collections of column  or  row  vectors.

Matrices are usually added element-wise, for

 

Scalar multiplication is defined by    .

The notation     means that all elements of     are identically zero.

A matrix norm is a generalization of ordinary absolute value     of a real or complex number, and can be considered a type of a vector norm.

For     and   matrices, and   ,  a scalar, a matrix norm is a real value     associated with a matrix for which the following properties hold.

 .

Common Matrix Norms

edit

The most commonly used norms are:

 .

Like vector norms any two matrix norms on   matrices of complex numbers are topologically equivalent in the sense that, if     and     are two different norms, then there exist positive constants     and     such that   .

The norm     on matrices     is an example of an induced norm. An induced norm is for a vector norm    defined by

 .

This could cause the same subscript notation to be used fot two different norms, sometimes.

Positive definite matrices

edit

   matrix     is said to be positive definite if for any vector   

 

for some positive constant    , not depending on    .

The property of being positive definite insures the numerical stability of a variety of common numerical techniques used to solve the equation    .

Taking     then

  .

so that

      and        .

This gives

 

and

  .

Consistency of Norms

edit

A matrix norm     and a vector norm     are said to be consistent when

  .

When     is the matrix norm induced by the vector norm     then the two norms will be consistent.

When the two norms are not consistent there will still be a positive constant     such that

  .

Finite Difference Operators

edit

approximation of a first derivative

edit

The defition of the derivative of a function gives the first and simplest finite difference.

 .

So the finite difference 

 

can be defined. It is an approximation to     when    is near  .

The finite difference approximation     for     is said to be of order   ,  if there exists     such that

 ,

when    is near  .

For practical reasons the order of a finite difference will be described under the assumption that     is sufficiently smooth so that it's Taylor's expansion up to some order exists. For example, if

 

then

 

so that

 ,

meaning that the order of the approximation of     by     is   .

The finite difference so far defined is a 2-point operator, since it requires 2 evaluations of  . If

 

then another 2-point operator

 

can be defined. Since

 ,

this     is of order 2, and is referred to as a centered difference operator. Centered difference operators are usually one order of accuracy higher than un-centered operators for the same number of points.


More generally, for     points     a finite difference operator

 

is usually defined by choosing the coefficients     so that     has as high of an order of accuracy as possible. Considering

 .

Then

 .

where the     are the right hand side of the Vandermonde system

 

and

 .

When the     are chosen so that 

 

then

 .

so that the operator is of order   .

At the end of the section a table for the first several values of   ,  the number of points, will be provided. The discussion will move on to the approximation of the second derivative.

approximation of a second derivative

edit

The definition of the second derivative of a function

 .

used together with the finite difference approximation for the first derivative

 

gives the finite difference 

 

In view of

 

for the operator just defined

 .

If instead, the difference operator

 

is used

 

If the other obvious possibility is tried

 

In view of

 ,

 .

So    

is a second order centered finite difference approximation for   .


The reasoning applied to the approximation of a first derivative can be used for the second derivative with only a few modifications.

For     points     a finite difference operator

 

is usually defined by choosing the coefficients     so that     has as high of an order of accuracy as possible. Considering

 .

Then

 .

where the     are the right hand side of the Vandermonde system

 

and

 .

When the     are chosen so that 

 

then

 .

so that the operator is of order   .

The effect of centering the points will be covered somewhere below.

approximation of higher derivatives

edit

Although approximations to higher derivatives can be defined recursively from those for derivatives of lower order, the end result is the same finite difference operators. The Vandermonde type system will be used again for this purpose.

 .

The number of points needed to approximate     by finite differences is at least   .


For     points     a finite difference operator

 

is usually defined by choosing the coefficients     so that    approximates     to as high of an order of accuracy as possible. Considering

 .

Then

 .

where the     are the right hand side of the Vandermonde system

 

and

 .

When the     are chosen so that 

 

then

 .

so that the operator is of order   .

An alternative analysis is to require that the finite difference operator differentiates powers of     exactly, up to the highest power possible.

effect of the placement of points

edit

Usually the     are taken to be integer valued, since the points are intended to coincide with those of some division of an interval or 2 or 3 dimensional domain. If these points and hence     are chosen with only accuracy in mind, then a higher accuracy of only one order can be achieved.

So start by seeing how high is the accuracy that     can be approximated with three points.

 

Then accuracy of order 4 can not be achieved, because it would require the solution of

 

which can not be solved since the matrix

 

is non-singular. The possibility of an     being    can be ruled out otherwise.

For accuracy of order 3

 

So the matrix

 

is singular and     are the roots of some polynomial 

  .

Two examples are next.

 

 


To see what the accuracy that     can be approximated to with three points.

 

Then accuracy of order 3 can not be achieved, because it would require the solution of

 

which can not be solved since the matrices

       and       

would both need to be singular.

If the matrix

 

is singular, then     are the roots of some polynomial 

  ,

implying

 

meaning that elementary row operations can transform

 

to

 

which is non-singular.

Conversely, if     are the roots of some polynomial 

  , then

 

can be solved and     approximated to an order of 2 accuracy.

See how high is the accuracy that     can be approximated with     points.

 

Then accuracy of order     can not be achieved, because it would require the solution of

 

which can not be solved since the matrix

 

is non-singular. The possibility of an     being    can be ruled out otherwise, because, for example, if   ,  then the non-singularity of the block

 

would force   .

For accuracy of order   

 

So the matrix

 

is singular and     are the roots of some polynomial 

  .

The progression for the second, third, ... derivatives goes as follows.

If     are the roots of some polynomial 

  

then the system

 

can be solved, and

 

approximates     to an order of accuracy of   .

If     are the roots of some polynomial 

  

then the system

 

can be solved, and

 

approximates     to an order of accuracy of   .

Now, the analysis is not quite done yet. Returning to the approximation of   .  If for the polynomial

 

it were that   ,  then the system can be solved for one more order of accuracy. So the question arises as to whether polynomials of the form

 

exist that have     distinct real roots. When     there is not. So consider   .

 

If     has 4 distinct real roots, then

 

has 3 distinct real roots, which it does not. So the order of approximation can not be improved. This is generally the case.

Returning to the approximation of   .  If for the polynomial

 

it were that   ,  then the system can be solved for one more order of accuracy. So the question arises as to whether polynomials of the form

 

exist that have     distinct real roots.

If     has     distinct real roots, then

 

has     distinct real roots, which it does not. So the order of approximation can not be improved.

For functions of a complex variable using roots of unity, for example, may obtain higher orders of approximation, since complex roots are allowed.

centered difference operators

edit

For     points     a finite difference operator

 

is said to be centered when the points are symmetrically placed about   .

 

When     is odd   .

To find the centered difference operators, consider

 .

Then

 .

where the     are the right hand side of the over-determined system

 

and

 .

When the     are chosen so that 

 

then

 .

so that the operator is of order   .

Since, for the centered case, the system is over-determined, some restriction is needed for the system to have a solution. A solution occurs when the     are the roots of a polynomial

 

with

 .

Observing that when     is even

 


and when     is odd

 .

So when     is even     has     for all odd    and when     is odd     has     for all even   .

So a centered difference operator will achieve the one order extra of accuracy when the number of points     is even and the order of the derivative     is odd or when the number of points     is odd and the order of the derivative     is even.


Shift Operators

edit

Shift of a Function

edit

Trigometric polynomials

edit

Let 

  

(3.2.0)


be trigometric polynomial defined on   .

Define the inner product on such trigometric polynomials by


 

 .

(3.2.1)


In light of the orthogonalities

 ,

and         when   ,

inner products can be calculated easily.


 

 .

(3.2.2)


and for

 

 

the inner product is given by


 

 .

(3.2.3)


Definition of a shift operator

edit

Define the shift operator     on     by


 

 .

(3.3.0)


Since

 

and

 ,

so that


 

 .

(3.3.1)


Approximation by trigometric polynomials

edit

Let     be a function defined on and periodic with respect to the interval   .  That is   .

The     degree trigometric polynomial approximation to     is given by

 

where


 

       and       .

(3.4.0)


   approximates     in the sense that

 

is minimized over all trigometric polynomials, of degree     or less, by   .

In fact

 

 .

The term in the center

   

 ,

so that


 

 

 .

(3.4.1)


 

linearity property

(3.4.2)


If     and     are the     degree trigometric polynomial approximations to     and   ,  then the     degree trigometric polynomial approximation to     is given by   .

This follows immediately from (3.4.0) since

       and       .

Fundamental Property

edit
 

(3.5.0)


Generally if     is the     degree trigometric polynomial approximation to a function   periodic on   ,  then     is the     degree trigometric polynomial approximation to   .

To see this calculate the trigometric polynomial approximations for   .

 

 

 

 

 ,

where

       and       .

 

 

 

 

 .

Comparing the results with (3.3.1) finishes the observation.

Another detail of use is

 

 

which is


 

 .

(3.5.1)


Error Estimation

edit

A result used in error estimation is

  .

(3.6.0)


When     is a sine polynomial

 

then


  

(3.6.1)


and


  .

(3.6.2)


Simple Functions

edit

Let   

so that

 

is a partition of   .

The function     is said to be simple on   ,  if


  

(3.7.0)


Of particular interest is when the points have equal spacing   .

The intent is to make estimates of   .

Begin by making an odd extension of     to    by setting     and continue the definition by extending     periodically.

Then approximate     with a sine polynomial

 

where

 .

When     is large enough so that some     are divided by    then, for   

 ,

and letting   ,

  ,

so that

  .

Now, return to the sum

 

with     for   .

If     and   ,  then   ,  and in this case

   and   .

So for      with    ,

 

  .

(3.7.1)


Next observe that if     is the     degree sine polynomial approximation to     then     is the     degree trigometric polynomial approximation to   .  The assumption that     is odd and periodic is still in effect.

Finally the intended results follow.

 

 

 

 

so that

 .

Making use of (3.7.1)

 .

Being that simple functions can be approximated by trigometric polynomials to arbitrary accuracy,


 

 .

(3.7.2)


Now, for   ,  and using the definition of the simple function   

 

To find the sum for     list tthe values of     over the values of     on the whole interval   .

 

This gives

 .

So the inequality follows


 

 .

(3.7.3)



Specialized Norms

edit

Energy and Heat Norms

edit

There are a number of less well known, but important norms. These norms are important in the analysis of many physical problems and are used in error estimation for finite difference and finite element methods. Examples are the energy and heat norms.

These norms are usually expressed in an integral form.

 

 

When       the inequality below holds.

 

This follows from a completely elementary, but lengthy calculation, as shown in appendix a). When additional assumptions on     are made this inequality can be improved somewhat.

 

See appendix b) for an explanation.

Discrete Heat Norms

edit

In the analysis of finite difference methods for partial differential equations it is useful to have discrete analogs of norms like the energy and heat norms.

In an attempt not to have notation too cumbersome some indices are suppressed when from the discussion it is clear what they refer to. When     is a     dimensional vector of complex numbers, finite difference operators are defined as discrete approximations or analogs of derivatives.

The most appropriate definition of a discrete energy or heat norm may vary due to differences in the handling of initial or boudary conditions. So for this reason the reader should make appropriate adjustments, when needed, to apply the same reasoning to another problem.

Before the discrete versions of energy or heat norms can be defined,  finite differece operations need to be defined and explained. This was done in the section on finite difference operators.


The next discrete version of the preceding inequality has important applications to the estimation of the error when approximating a second derivative with a finite difference operator.

If     then

 

with

 

and generally the following under-estimate holds.

 

See appendix c) for a proof.

The inequality can be improved for general     by using (3.7.3) with     increased by 2.

 

If     then

 

and

 .


appendix a)

edit

When       the inequality below holds.

 

First apply the Cauchy Schwarz inequality.

 ,

Next observe the integrals on the right are increasing with   .

 

Integrate, make cancellations, and reapply the first inequality.

  .

  .

  .

After integrating again the inequality is improved.

  .

  .

Now, assume the inequality immediately below holds for some   .

 

After substituting the inequality above into the next

  .

the following observations are made.

  .

  .

  .

  .

  .

Repeating this iteration leads to a sequence   

which converges to   ,  the solution of   .

So:

 

and

  .

For    ,

  .

  .

To handle the part   

 ,

 

  .

  .

  .

Reasoning as before this inequality can be strengthend to

  .

  .

  .

Adding the results of the two main calculations

  .

appendix b)

edit

When       the inequality below holds.

 


Assume the conditions that     can be approximated by a trigometric polynomial

 .

such that     is also approximated by the polynomials derivative

 

That is to say, given   ,  there exixts  a trigometric sine polynomial   ,  such that

    and    .

Now,

 

  and  

 .

So it is easy to see that

 .

In this case the inequality is sharp, since it holds with equality for   .

Since

 ,

and  

 ,

the inequality holds for     and   .

appendix c)

edit

If     then

 

with

 

and generally the following under-estimate holds.

 

The cases for 1 and 2 are simple.

 

and

 

with equality, when   .

To prove the general under-estimate do as follows.

Use     and apply the Cauchy Schwartz inequality together with inequality ().

 

 

 

(1)

So for   ,

 

and

 .

After inserting the inequality above into (1)

 .

So for   ,

 

 

and after using formula ()

 .

Using     the right end of the sum is estimated using nearly the same procedure.

 

 

 

So for   ,

 ,

 ,

and

 .

So for   ,

 ,

 

 

 .

Combining the two results:

 

 

 .

When     is odd, use     so that the inequality becomes

 

 

 .

When     is even, use     so that the inequality becomes

 

 

 .

Special Formulas

edit

The rule named L´Hospital's Rule named by a French math teacher who published it without proof, because he said you need to be taken to the hospital if you can't prove it yourself, states.

If     and     then,

 .

The rule named Leibniz's Rule states how to differentiate a product of two functions     times.

 

There are a number of ways to find the formulas for the sums of the powers of the first     integers. One of the most convenient is as follows. Begin with the geometric formula.

 

By convention   .

Apply Leibniz's Rule rule to the right hand side of the equivalent identity

 

to differentiate both sides     times.

 

Upon noticing that only the first two terms of the sum on the right are non-zero

 .

Now take the limit as     to establish the formula

 .

From term by term differentiation.

 

and

 .

More generally

 .

and thus

 .

A progression of formulas follow by setting   .

 

 .

 .

and

 .

 

After adjusting the index of summation and increasing     by     the progression has the form

 .

Letting   

 .

 


The formula called  summation by parts  is a discrete analog of  integration by parts . It is verified simply by inspection.

 

One form of the statement of the Fundamental Theorem of Integral Calculus is

For     it holds   .

Another rule which slightly generalizes the Fundamental Theorem of Integral Calculus says

 .


The following simple inequality can be used to bound sums of products.

 

and also

 ,

so that

 .

Letting     and    

 .

The particular case with     gives

 .

save scratch

edit



 

 

 

 

Now, suppose for some   

 

 

So for   

 

 

 

 


  zzzzzzzzzz

So for   

 

 

 

 

 

 

xxxxxxxxxxxxxxxx