288 CHAPTER 12. SERIES AND TRANSFORMS

Then using this fact as needed, consider the following computation in which I will try tochoose bk to make ∫

π

−π

∣∣∣∣∣ f (θ)− n

∑k=−n

bkeikθ

∣∣∣∣∣2

dθ (12.13)

as small as possible. Remember that |z|2 = zz̄ whenever z is a complex number. Using thisand doing routine computations,∫

π

−π

∣∣∣∣∣ f (θ)− n

∑k=−n

bkeikθ

∣∣∣∣∣2

=∫

π

−π

| f (θ)|2 dθ −2Re∫

π

−π

n

∑k=−n

f (θ)bkeiθ dθ +2π

n

∑k=−n|bk|2

=∫

π

−π

| f (θ)|2 dθ −2(2π)Ren

∑k=−n

akbk +2π

n

∑k=−n|bk|2

Note that if bk = ak, this would equal∫π

−π

| f (θ)|2 dθ −2(2π)n

∑k=−n|ak|2 +2π

n

∑k=−n|ak|2 =

∫π

−π

| f (θ)|2 dθ −2π

n

∑k=−n|ak|2

In the general case, it follows from the Cauchy Schwarz inequality,

≥∫

π

−π

| f (θ)|2 dθ −2(2π)

(n

∑k=−n|ak|2

)1/2( n

∑k=−n|bk|2

)1/2

+2π

n

∑k=−n|bk|2

≥∫

π

−π

| f (θ)|2 dθ −2π

(n

∑k=−n|ak|2 +

n

∑k=−n|bk|2

)+2π

n

∑k=−n|bk|2

=∫

π

−π

| f (θ)|2 dθ −2π

n

∑k=−n|ak|2

Therefore, the expression in 12.13 is minimized when bk = ak. We also observe the fol-lowing fundamental inequality. For ak the Fourier coefficients,∫

π

−π

∣∣∣∣∣ f (θ)− n

∑k=−n

akeikθ

∣∣∣∣∣2

dθ ≡∫

π

−π

| f (θ)−Sn f (θ)|2 dθ

=∫

π

−π

| f (θ)|2 dθ −2π

n

∑k=−n|ak|2 ≥ 0

so this yields Parseval’s inequality, an important inequality involving the Fourier coeffi-cients, 1

∫π

−π| f (θ)|2 dθ ≥ ∑

nk=−n |ak|2. This has proved most of the following approxi-

mation theorem.

Theorem 12.6.1 Let αn f (x) denote any linear combination of the functions θ →eikθ for −n≤ k ≤ n. Then∫

π

−π

| f −αn f |2 dx≥∫

π

−π

| f −Sn f |2 dx

Also,∫

π

−π|Sn f |2 dx≤

∫π

−π| f |2 dx.

288 CHAPTER 12. SERIES AND TRANSFORMSThen using this fact as needed, consider the following computation in which I will try tochoose b; to make[ \reas small as possible. Remember that \z| = zz whenever z is a complex number. Using thisand doing routine computations,_y by vito[\rek=—n= [. If (0)|? de — sel f (O)bpe!?dO + 20 y Ibe?~ oe —n k=—n2_y by elkk=—n(12.13)n-[ |f (0)/’d@—2(2n)Re )° ab, +20 y |by|"k=—-n k=—-nNote that if by; = ax, this would equal[7 \F@)Pao-202n) ¥ Jaul?+2" Y Jaul? = [" |e@Pao—20 YY la?k=—n k=—-n k=—nIn the general case, it follows from the Cauchy Schwarz inequality,: , 27, 1/2> | \f(@)/?40-22n) ( y i?) e mi?) +20 y |x|?k=—-n k=—n k=—-nIV[-ircoyeao—a( Y lal’t+ ¥ ri?) +20 Y° |bi?k=—n k=—n k=—n[\F@Pae-22 Y Ja?—a k=—-nTherefore, the expression in 12.13 is minimized when by = ax. We also observe the fol-lowing fundamental inequality. For a, the Fourier coefficients,[\F@- ¥ avek=—n2= | \ro)-s.f(@Pa0[ |f (0)|?d@—2n y lax’ >0k=-—nso this yields Parseval’s inequality, an important inequality involving the Fourier coeffi-cients, + [7 |f (0 0)|>de > Yn |ay|”. This has proved most of the following approxi-> Onmation theorem.Theorem 12.6.1 Lez On f (x) denote any linear combination of the functions 8 >elk@ for —n<k <n. Then[i ip-eupPax> [if sutP axAlso, [7 _|Snf | dx < [| f\" ax.