#import "@preview/physica:0.9.6": * #import "lib.typ": ccases = problem 1 == a) $ f(x) := sin(3 x) + 5 cos(2 x) $ is already in its fourier series form. the coefficients can be read as $ a_0 & = 0, \ a_n & = ccases(5, n = 2, 0) \ b_n & = ccases(1, n = 3, 0) $ == b) let $ g(x) = x^2, quad -pi <= x <= pi $ then we can compute the fourier series using the formulas for $2pi$-periodic functions $ a_0 & = 1/pi integral_(-pi)^pi x^2 dd(x) = 2/pi integral_0^pi x^2 dd(x) \ & = 2/(3 pi) [x^3]_0^pi = 2/3 pi^2 $ $ a_n & = 1/pi integral_(-pi)^pi x^2 cos(n x) dd(x) \ & = 1/pi [x^2/n sin(n x) + (2 x)/n^2 cos(n x) -2/n^3 sin(n x)]_(-pi)^pi \ & = 2/(n pi) [x^2 sin(n x) + (2 x)/n cos(n x) - 2/n^2 sin(n x)]_0^pi \ & = 2/(n pi) ((2 pi)/n cos(pi n)) \ & = 4/n^2 (-1)^n $ $ b_n & = 1/pi integral_(-pi)^pi x^2 sin(n x) dd(x) \ & = 1/pi [-x^2/n cos(n x) +(2 x)/n^2 sin(n x) + 2/n^3 cos(n x)]_(-pi)^pi \ & = 1/pi [(cancel(-pi^2/n cos(pi n)) + (2 pi)/n^2 sin(pi n) + cancel(2/n^3 cos(pi n))) \ & - (cancel(-pi^2/n cos(-pi n)) - (2 pi)/n^2 sin(-pi n) + cancel(2/n^3 cos(-pi n)))] \ & = 2/n^2 (sin(pi n) + sin(-pi n)) \ & = 2/n^2 (sin(pi n) - sin(pi n)) = 0 $ the fact that $b_n$ is zero is reassuring, since the oddness that it contributes with would be undesirable for a function like $x^2$. == c) show $ h(x) & = f(x) * g(x) = integral_(-pi)^pi g(y) f(x - y) dd(y) \ & = -(4 pi)/9 sin(3 x) + 5 pi cos(2 x) $ #line(length: 100%, stroke: (dash: "dotted")) $ h(x) = integral_(-pi)^pi y^2 [sin(3 (x - y)) + 5 cos(2 (x - y))] dd(y) $ this would be an awful lot of integration by parts, so let's be more clever about this. we can expand the trigonometric terms in $f(x - y)$ to obtain simpler terms that cancel out under $integral_(-pi)^pi g(y) sin(n y) dd(y) = 0$ - $ sin(3(x - y)) = sin(3x) cos(3y) - cos(3x) sin(3y) $ - $ cos(2(x - y)) = cos(2x) cos(2y) + sin(2x) sin(2y) $ thus we obtain $ h(x) & = sin(3x) integral_(-pi)^pi y^2 cos(3y) dd(y) \ & - cancel(cos(3 x) integral_(-pi)^pi y^2 sin(3 y) dd(y)) \ & + 5 cos(2x) integral_(-pi)^pi y^2 cos(2y) dd(y) \ & + cancel(5 sin(2x) integral_(-pi)^pi y^2 sin(2y) dd(y)) $ but these are simply the coefficients of $y^2$ from b), such that $ 1/pi integral_(-pi)^pi y^2 cos(2y) dd(y) = [4/n^2 (-1)^n]_(n=2) = 1 $ $ 1/pi integral_(-pi)^pi y^2 cos(3y) dd(y) = [4/n^2 (-1)^n]_(n=3) = -4/9 $ which gives $ h(x) = - (4 pi)/9 sin(3x) + 5 pi cos(2x) $ #h(1fr) #math.qed == d) $h(n)$ is also made up of basis vectors, such that the coefficients can be written as $ a_0 & = 0 \ a_n & = ccases(5pi, n = 2, 0) \ b_n & = ccases(-(4pi)/9, n = 3, 0) $ $ c_n (f) & = a_n (f) + b_n (f) = ccases(5, n = 2, 1, n = 3, 0) \ c_n (g) & = ccases(2/3 pi^2, n = 0, 4/n^2 (-1)^n,) \ c_n (h) & = ccases(5pi, n = 2, -(4 pi)/9, n = 3, 0) $ we can notice that $c_2 (h) = pi c_2 (f)$ and $c_3 (h) = pi c_3 (g)$. if we look for a deeper pattern, we can also notice that $c_2 (g) = 1$ and $c_3 (f) = 1$ such that we could say that $ c_2(h) = pi c_2(f) c_2(g) quad and quad c_3(h) = pi c_3(f) c_3(g) $ this is a profound result -- well, the generalized statement is -- and is related to the fourier transform. we are saying that convolving two functions is the same as taking their point-wise product at certain frequencies.