OU blog

Personal Blogs

Valentin Fadeev

Self ex-spline-atory

Visible to anyone in the world
Edited by Valentin Fadeev, Thursday, 27 Mar 2014, 10:06

M832 did not really fit in the "big picture" of my study plan this year, not only because I am notoriously bad at numerical calculations. Having invested a lot of effort and sleepless nights in developing intuition for the behaviour of analytic functions I was suddenly confronted by the cubic splines which seemed to have all those properties, that well-mannered functions would be never allowed to possess.

Nicely, but artificially glued together of several pieces of cubics, smooth only up to the second derivative, vanishing on the entire intervals, now this is what seems really counter-intuitive. Be that as it may, the TMA deadlines had to be met.

The following is the kind of problem I got stuck with for a while. Obviously I am not replicating a TMA question here, giving an extended solution to one of the problems from the Course Notes. So the task is to express a function, say f of x equals x squared in terms of cubic B-splines on the entire real axis. I am omitting a lot of background material focusing on one particular idea that arises in the solution.

equation left hand side x squared equals right hand side n ary summation from p equals negative normal infinity to normal infinity over lamda sub p times cap b sub p of x

Since cap b sub p cubed of x has a supporting interval equation left hand side open p comma sum with, 3 , summands p plus three plus one close equals right hand side open p comma p plus four close of length 4 outside which it vanishes, we can start by expressing the function in terms of cap b sub negative three comma cap b sub negative two comma cap b sub negative one comma cap b sub zero on open 0.1 close and then try to extend the result. Calculating the expressions for the splines on open zero comma one close :

cap b sub negative three of x equals negative one divided by 24 times open x minus one close cubed

cap b sub negative two of x equals one divided by 24 times open three times x cubed minus six times x squared plus four close

cap b sub negative one of x equals one divided by 24 times open sum with, 4 , summands negative three times x cubed plus three times x squared plus three times x plus one close

cap b sub zero of x equals one divided by 24 times x cubed

multiplying by the respective coefficients, summing and equating powers of x on each side we arrive at the following system of equations:

case statement case 1column 1 plus plus minus minus plus plus minus minus lamda minus minus three times times three lamda minus minus two times times three lamda minus minus one lamda zero equals equals zero case 2column 1 plus plus minus minus times times three lamda minus minus three times times six lamda minus minus two times times three lamda minus minus one equals equals zero case 3column 1 plus plus minus minus times times three lamda minus minus three times times three lamda minus minus one equals equals 24 case 4column 1 plus plus plus lamda minus minus three times times four lamda minus minus two lamda minus minus one equals equals zero

Having the solution (guaranteed by the Schoenberg-Whitney theorem): equation left hand side open lamda sub negative three comma lamda sub negative two comma lamda sub negative one comma lamda sub zero close equals right hand side open eight divided by three comma negative four divided by three comma eight divided by three comma 44 divided by three close

Now we want to find all coefficients on each of the intervals open xi sub p comma xi sub p plus four close for the points equation left hand side xi sub i equals right hand side i times h semicolon i equals prefix plus minus of one comma prefix plus minus of two full stop full stop full stop full stop . From the general expression for the B-spline it can be deduced that

cap b sub p of xi sub p plus one equals one divided by 24 times h

cap b sub p of xi sub p plus two equals one divided by six times h

cap b sub p of xi sub p plus three equals one divided by 24 times h

which for h equals one leads to the following recurrence relation:

equation left hand side sum with, 3 , summands lamda sub j minus one plus four times lamda sub j minus two plus lamda sub j minus three equals right hand side 24 times j squared

or, after changing the index

sum with, 3 , summands lamda sub j plus four times lamda sub j plus one plus lamda sub j plus two equals 24 left parenthesis j plus three times right parenthesis squared

Now here is the trick that I came up with and that was not (at least not explicitly) described in the Course Notes or the set book. The last expression can be thought of as a "second-order linear inhomogeneous recurrence relation". The advantage of this approach is that the structure of the solution instantly becomes clear.

The general solution of the corresponding homogeneous relation

sum with, 3 , summands lamda sub j plus four times lamda sub j plus one plus lamda sub j plus two equals zero

is derived in the course notes, using the standard method of solving this type of recurrencies and is given by the following expression:

equation left hand side lamda sub j super h equals right hand side alpha times open negative two minus Square root of three close super j plus beta times open negative two plus Square root of three close super j

It can also be found using generating functions. Not surprisingly it depends on 2 arbitary constants, as it takes 2 initial terms, lamda sub zero and lamda sub negative one to reconstruct the whole sequence from the three-term recurrency. Applying the general ideas from the linear systems we deduce that in order to obtain the general solution of the inhomogeneous recurrency we have to add a particular solution to the expression above.

Since the RHS is the quadratic polynomial it makes sence to look for the particular solution in the form:

equation left hand side lamda sub j super p equals right hand side sum with, 3 , summands a times j squared plus b times j plus c

Substituting this into the original recurrency and gatherig together the powers of j we obtain:

equation left hand side sum with, 5 , summands six times a times j squared plus open 12 times a plus six times b close times j plus eight times a plus six times b plus six times c equals right hand side sum with, 3 , summands 24 times j squared plus 144 times j plus 216

which after equating powers gives the solution equation left hand side open a comma b comma c close equals right hand side open four comma 16 comma 44 divided by three close

Thus the general solution of the inhomogeneous equation is given by the following formula:

equation left hand side lamda sub j equals right hand side sum with, 5 , summands alpha times open negative two minus Square root of three close super j plus beta times open negative two plus Square root of three close super j plus four times j squared plus 16 times j plus 44 divided by three

Now we can use the values of lamda sub zero and lamda sub negative one to determine the constants (bearing in mind that equation left hand side open negative two minus Square root of three close times open negative two plus Square root of three close equals right hand side negative one ):

equation left hand side 44 divided by three equals right hand side sum with, 3 , summands alpha plus beta plus 44 divided by three

equation left hand side eight divided by three equals right hand side negative alpha times open negative two plus Square root of three close minus beta times open negative two minus Square root of three close plus eight divided by three

which gives equation sequence alpha equals beta equals zero . Thus finally:

equation left hand side lamda sub j equals right hand side sum with, 3 , summands four times j squared plus 16 times j plus 44 divided by three

which is the solution of the original problem.

Permalink Add your comment
Share post
Valentin Fadeev

Recurrencies and summing by parts

Visible to anyone in the world
Edited by Valentin Fadeev, Thursday, 22 Jul 2010, 12:23

This question was posed in sci.math.research group on Google:

http://groups.google.com/group/sci.math.research/browse_thread/thread/48ca096ec5f08a39?hl=en

After latexing the mind-bending plain text of the discussion it looks like this:

a sub one equals one

equation left hand side a sub n equals right hand side one divided by one minus b super n times n ary summation from m equals one to n minus one over a sub m times matrix row 1 n row 2 m times open one minus b close super n minus m times b super m of asterisk operator

assuming b not equals one . The author's conjecture was that for large n :

a sub n b minus one divided by natural log of b

Summing by parts is quite a standard device. Though, like with integration by parts the difficult part often is use it at the right moment. Whittakker and Watson ascribe it's systematic introduction to Abel. Probably the best account of it is given in "Concrete Mathematics". The authors introduce "definite sums" which are effectively the sums with an omitted last term:

equation left hand side n ary summation from a to b over f of x times delta times x equals right hand side n ary summation from x equals a to x equals b minus one over f of x

The cryptic delta is added solely to enhance the analogy with definite integrals.

The general formula is the following:

equation left hand side n ary summation from n equals a to n equals b minus one over u times normal cap delta times v equals right hand side u times v vertical line sub n equals a super n equals b minus n ary summation from n equals a to n equals b minus one over cap e times v times normal cap delta times u

Where equation left hand side normal cap delta times f of x equals right hand side f times open x plus one close minus f of x is the difference operator and equation left hand side cap e times f of x equals right hand side f times open x plus one close is the shift operator. The formula is easily proved by evaluating normal cap delta times u times v

equation left hand side n ary summation from a to b over normal cap delta times f of x equals right hand side f of b minus f of a

Part of the sum (*) on the right prompts for the binomial formula. Hence, it would be good to pull it out of the sum. Let's try:

equation sequence u equals a sub m times normal cap delta times v equals matrix row 1 n row 2 m times open one minus b close super n minus m times b super m

equation left hand side normal cap delta times u equals right hand side a sub m plus one minus a sub m

equation sequence v equals n ary summation from m equals one to x over matrix row 1 n row 2 m times open one minus b close super n minus m times b super m vertical line sub one super n equals one postfix minus left parenthesis one minus b times right parenthesis super n minus n times b left parenthesis one minus b times right parenthesis super n minus one

For

equation left hand side n ary summation from m equals one to n over matrix row 1 n row 2 m times open one minus b close super n minus m times b super m equals right hand side

equation left hand side equals right hand side n ary summation from m equals zero to n over open matrix row 1 n row 2 m times open one minus b close super n minus m times b super m close postfix minus left parenthesis equation left hand side one minus b times right parenthesis super n equals right hand side left parenthesis one minus b plus b times right parenthesis super n postfix minus left parenthesis equation left hand side one minus b times right parenthesis super n equals right hand side one postfix minus left parenthesis one minus b times right parenthesis super n

Putting it all together:

equation left hand side n ary summation from m equals one to n minus one over a sub m times matrix row 1 n row 2 m times open one minus b close super n minus m times b super m equals right hand side

equation left hand side equals right hand side a sub n left parenthesis one postfix minus left parenthesis one minus b times right parenthesis super n minus n times b times open one minus b times right parenthesis super n minus one close minus n ary summation from m equals one to n minus one over open a sub m plus one minus a sub m close left parenthesis one postfix minus left parenthesis equation left hand side one minus b times right parenthesis super n minus n times b times open one minus b times right parenthesis super n minus one close equals right hand side

equation left hand side equals right hand side a sub n left parenthesis one postfix minus left parenthesis one minus b times right parenthesis super n minus n times b times open one minus b times right parenthesis super n minus one close postfix minus left parenthesis one postfix minus left parenthesis equation left hand side one minus b times right parenthesis super n minus n times b times open one minus b times right parenthesis super n minus one close times open a sub n minus one close equals right hand side one postfix minus left parenthesis one minus b times right parenthesis super n minus n times b left parenthesis one minus b times right parenthesis super n minus one

This gives an equation for a sub n :

equation left hand side a sub n times open one minus b super n close equals right hand side one postfix minus left parenthesis one minus b times right parenthesis super n minus n times b left parenthesis one minus b times right parenthesis super n minus one

equation left hand side a sub n equals right hand side one postfix minus left parenthesis one minus b times right parenthesis super n minus n times b left parenthesis one minus b times right parenthesis super n minus one divided by one minus b super n

Permalink 2 comments (latest comment by Valentin Fadeev, Sunday, 25 Jul 2010, 21:50)
Share post

This blog might contain posts that are only visible to logged-in users, or where only logged-in users can comment. If you have an account on the system, please log in for full access.

Total visits to this blog: 54126