11.5. LEAST SQUARES 283

Lemma 11.5.1 Let V andW be finite dimensional inner product spaces and let A : V →Wbe linear. For each y ∈W there exists x ∈ V such that

|Ax− y| ≤ |Ax1 − y|

for all x1 ∈ V. Also, x ∈ V is a solution to this minimization problem if and only if x is asolution to the equation, A∗Ax = A∗y.

Proof: By Theorem 11.2.4 on Page 276 there exists a point, Ax0, in the finite dimen-sional subspace, A (V ) , of W such that for all x ∈ V, |Ax− y|2 ≥ |Ax0 − y|2 . Also, fromthis theorem, this happens if and only if Ax0 − y is perpendicular to every Ax ∈ A (V ) .Therefore, the solution is characterized by (Ax0 − y,Ax) = 0 for all x ∈ V which is thesame as saying (A∗Ax0 −A∗y, x) = 0 for all x ∈ V. In other words the solution is obtainedby solving A∗Ax0 = A∗y for x0. ■

Consider the problem of finding the least squares regression line in statistics. Supposeyou have given points in the plane, {(xi, yi)}ni=1 and you would like to find constants mand b such that the line y = mx + b goes through all these points. Of course this will beimpossible in general. Therefore, try to find m, b such that you do the best you can to solvethe system 

y1...

yn

 =

x1 1...

...

xn 1

(m

b

)

which is of the form y = Ax. In other words try to make

∣∣∣∣∣∣∣∣A(m

b

)−

y1...

yn

∣∣∣∣∣∣∣∣2

as small

as possible. According to what was just shown, it is desired to solve the following for m andb.

A∗A

(m

b

)= A∗

y1...

yn

 .

Since A∗ = AT in this case,( ∑ni=1 x

2i

∑ni=1 xi∑n

i=1 xi n

)(m

b

)=

( ∑ni=1 xiyi∑ni=1 yi

)Solving this system of equations for m and b,

m =− (∑n

i=1 xi) (∑n

i=1 yi) + (∑n

i=1 xiyi)n

(∑n

i=1 x2i )n− (

∑ni=1 xi)

2

and

b =− (∑n

i=1 xi)∑n

i=1 xiyi + (∑n

i=1 yi)∑n

i=1 x2i

(∑n

i=1 x2i )n− (

∑ni=1 xi)

2 .

One could clearly do a least squares fit for curves of the form y = ax2 + bx + c in thesame way. In this case you solve as well as possible for a, b, and c the system

x21 x1 1...

......

x2n xn 1

 a

b

c

 =

y1...

yn

using the same techniques.

11.5. LEAST SQUARES 283Lemma 11.5.1 Let V and W be finite dimensional inner product spaces and let A: V + Wbe linear. For each y € W there exists x € V such that|Az — y| < |Aa, — y|for alla, € V. Also, x € V is a solution to this minimization problem if and only if x is asolution to the equation, A* Ax = A*y.Proof: By Theorem 11.2.4 on Page 276 there exists a point, Azo, in the finite dimen-sional subspace, A(V), of W such that for all x € V,|Ax —y|? > |Azo —y|?. Also, fromthis theorem, this happens if and only if Avo — y is perpendicular to every Ax € A(V).Therefore, the solution is characterized by (Avo — y, Ax) = 0 for all « € V which is thesame as saying (A* Avo — A*y,x) = 0 for all a € V. In other words the solution is obtainedby solving A* Axp = A*y for xo.Consider the problem of finding the least squares regression line in statistics. Supposeyou have given points in the plane, {(x;,y;)}/_, and you would like to find constants mand 6 such that the line y = mx + 6 goes through all these points. Of course this will beimpossible in general. Therefore, try to find m, b such that you do the best you can to solvethe systemYi xy 1— . m7 , ( bYn Ln 12Y1which is of the form y = Ax. In other words try to make |A ( ; — : as smallYnas possible. According to what was just shown, it is desired to solve the following for m andb.Y1wa(t Jaa :b .YnSince A* = A” in this case,ete oe ei mi \_ in Pithie Fi n b iat YiSolving this system of equations for m and 8,— (ey 1;) (ey yi) + (ey Liyi)N(S27, 3) n — (ai)= (doer @i) Doin Vii + Oop Yi) Doe2 :(ie @7) n — (Sy 2)One could clearly do a least squares fit for curves of the form y = ax? + ba + c in thesame way. In this case you solve as well as possible for a,b, and c the systemandxt rv =i a Y1b =v2 ay, 1 c Ynusing the same techniques.