20
St. John's University of Tanzania MAT210 NUMERICAL ANALYSIS 2013/14 Semester II INTERPOLATION Splines Kaw, Chapter 5.05

SJUT/MAT210/Interpolation/Splines 2013-14S2

Embed Size (px)

DESCRIPTION

Lecture slides based on Autar Kaw's Numerical Methods text, Chapter 5.05, found at http://nm.mathforcollege.com

Citation preview

Page 1: SJUT/MAT210/Interpolation/Splines 2013-14S2

St. John's University of Tanzania

MAT210 NUMERICAL ANALYSIS2013/14 Semester II

INTERPOLATIONSplines

Kaw, Chapter 5.05

Page 2: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 2 of 20

● Direct, Newton Divided Difference & Lagrangian Interpolation● Two approaches for finding the same nth order

polynomial fit for all points in an data set● Is splines just another way to do the same

● NO!● It is Piecewise polynomial interpolation● Each piece can be linear, quadratic or cubic

Introduction

Page 3: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 3 of 20

How do we avoid this?

Page 4: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 4 of 20

By observing● This function has distinct regions

● The interval from x ≈ -1 to -0.5● The interval from x ≈ -0.5 to -0.1● The interval from x ≈ -0.1 to +0.1● The interval from x ≈ 0.1 to 0.5● The interval from x ≈ 0.5 to 1

Though there is some symmetry...

● It would be better to fit different functions to different intervals

Page 5: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 5 of 20

Piecewise Polynomials

Rather than interpolating n+1 points with a single polynomial of degree n, put different polynomials on each interval

S(x)={s0(x) , x∈[x0 , x1)s1(x) , x∈[x1,x2)

sn−1(x) , x∈[xn−1 , xn]}where the sj are polynomials of (usually) small degree

Page 6: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 6 of 20

Interpretation● Piecewise linear = connect the dots● Piecewise quadratic

= parabolas between the dots● But wait

● Two points uniquely define a line– linear is understandable

● Three points are needed for a parabola– How is the other degree of freedom set?

Page 7: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 7 of 20

Splines● In the connect the dots linear case,

the curve is not “smooth”● Add “smoothness” into the requirement

● Draftsmen achieved this smoothness with splines - a flexible strip of metal or wood

Page 8: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 8 of 20

Splines● In the connect the dots linear case,

the curve is not “smooth”● Add “smoothness” into the requirement

● Draftsmen achieved this smoothness with splines - a flexible strip of metal or wood

● Mathematicians achieve it by matching derivatives at the end points of the intervals

Page 9: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 9 of 20

Linear Splines

Page 10: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 10 of 20

Linear Splines

Slope between points

Page 11: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 11 of 20

v(16) … Again

The linear case is unchanged

No surprise there

Page 12: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 12 of 20

Quadratic Splines● Now things get interesting

● How to find all the coefficients?● 3n coefficients, n equations, n continuity at

end points, whence the other n?

Page 13: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 13 of 20

2n from continuity

Each curve must pass through both endpoints

Page 14: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 14 of 20

n-1 from smoothness

a1x2+b1 x+c1⇒2 a1 x+b1

a2 x2+b2 x+c2⇒2 a2 x+b2

Must match at n-1 interior points

2 a1 xi+b1=2 a2 xi+b2∀ i ∈ [1 ,n−1]

Page 15: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 15 of 20

One more assumption● This is 3n unknowns and 3n -1 equations● Need to set one more condition● Generally set the first spline to be linear

● a1 = 0

● Now use any technique to solve simultaneous linear equations

Page 16: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 16 of 20

Revisiting the Rocket

Page 17: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 17 of 20

The continuous derivatives

The draftsman is bending his spline!

Page 18: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 18 of 20

The Final Matrix

Page 19: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 19 of 20

The Solution

Page 20: SJUT/MAT210/Interpolation/Splines 2013-14S2

MAT210 2013/14 Sem II 20 of 20

Going Deeper● The overall curve is smooth and the

accuracy can be quite good● Cubic is better, more common

– See that next time

● What about finding the distance traveled?● From 11 to 14s?● From 11 to 16s?● From 0 to 30s?