Unit 3: Solution of Overdetermined Linear System of Equations (Curve Fitting) In this section, we consider linear systems of the form Ax = b with A ∈ R m×n with m >> n. Führer: FMA421 2010/11 38 Curve Fitting Example We would like to fit the data i ti yi 0 -1.0 1.0 1 -0.5 0.5 2 0.0 0.0 3 0.5 0.5 4 1.0 2.0 by a quadratic polynomial of the form 2 p(t) = a2t + a1t + a0. Führer: FMA421 2010/11 39 Curve Fitting Example (Cont.) We set up interpolation conditions p(ti) = yi: 1 1 1 1 1 t0 t1 t2 t3 t4 t20 t21 a0 t22 a1 t23 a2 t24 y0 y 1 = y2 . y3 y4 This leads to a nonsquare linear system. Führer: FMA421 2010/11 40 Least Squares Solution What do we mean when we look for a solution of a linear system without a solution? Find an x̂ with kAx̂ − bk2 = min kAx − bk2 = min kr(x)k2 x x with the residual vectors r(x) := b − Ax. Führer: FMA421 2010/11 41 Length of a vector By kvk2 we mean the (Euclidean) length of a vector: T kvk2 = (v v) 1/2 = n X 2 1/2 . vi i=1 (called also the 2-norm of v , see help norm ) Führer: FMA421 2010/11 42 Is there a unique solution? Problem: Find an x̂ with kAx̂ − bk2 = min kAx − bk2 = min kr(x)k2 x x A necessary condition: d 2 kr(x)k2 dx x=x̂ = 0. Führer: FMA421 2010/11 43 Is there a unique solution? (Cont.) From 2 T T T T T T T kr(x)k2 = r r = (b − Ax) (b − Ax) = b b − 2x A b + x A Ax we take the first derivative w.r.t. x. Flerdim-course: T T A Ax̂ − A b = 0 Führer: FMA421 2010/11 44 Normal Equations T T A Ax̂ − A b = 0 We can see them as a linear system Ax = b̄ where b̄ is the orthogonal projection of b onto the range space of A (bildrummet, kolonnrummet). Führer: FMA421 2010/11 45 Normal Equations (Cont.) Consider the range space R(A). It is spanned by the columns of A. T T A (b − Ax) = A r = 0 Thus r⊥A. This justifies the name ”normal” equations. r b A x Im(A) Führer: FMA421 2010/11 46 In MATLAB Alternative 1: x=(A0 ∗ A)\ A0*b Alternative 2: x=A\ b there are even more stable methods. Führer: FMA421 2010/11 47 Oat Porridge Problem Führer: FMA421 2010/11 48 Oat Porridge Problem (Cont.) We pose the following questions: • How much water is needed for three portions? • How much water is needed for 300 portions? To answer these questions, we first have to set up a correct mathematical model and then make a least squares fit to determine the unknown parameters in the model. Führer: FMA421 2010/11 49 Oat Porridge Problem (Cont.) The discussion of the different models will take place in the lecture - not here. Führer: FMA421 2010/11 50
© Copyright 2026 Paperzz