- Calculus, Volume 2: Multi-Variable Calculus and Linear Algebra with Applications to Differential Equations and Probability
- Tom M. Apostol
- Second Edition
- 1991
- 978-1-119-49676-2
1.13 Exercises
-
In the linear space of all real polynomials, define $(f, g) = \int_0^\infty e^{-t} f(t) g(t) \, dt.$
$\text{(a)}\quad$ Prove that this improper integral converges absolutely for all polynomials $f$ and $g.$
$\text{(b)}\quad$ If $x_n(t) = t^n$ for $n = 0, 1, 2, \ldots,$ prove that $(x_n, x_m) = (m + n)!$.
$\text{(c)}\quad$ Compute $(f, g)$ when $f(t) = (t + 1)^2$ and $g(t) = t^2 + 1.$
$\text{(d)}\quad$ Find all linear polynomials $g(t) = a + bt$ orthogonal to $f(t) = 1 + t.$
-
$\text{(a)}\quad$ For any choice of real polynomials $f(t)$ and $g(t),$ $f(t)g(t)$ is a real polynomial of degree $\leq n$ for some integer $n \geq 0.$ Accordingly, we can rewrite $f(t)g(t)$ as
\begin{align*}
f(t)g(t) &= \sum_{k=0}^n a_kt^k
\end{align*}
By extension of the triangle inequality, we have
\begin{align*}
|f(t)g(t)| &= \left|\sum_{k=0}^n a_kt^k\right|
\\
&\leq \sum_{k=0}^n |a_kt^k|
\end{align*}
And because $e^{-t} \gt 0$ for all real $t,$ we have $|e^{-t}| = e^{-t}.$ Hence,
\begin{align*}
|e^{-t}f(t)g(t)| &= e^{-t}\left|\sum_{k=0}^n a_kt^k\right|
\\
&\leq e^{-t} \sum_{k=0}^n |a_kt^k|
\\
\\
&\text{and}
\\
\\
\int_0^{\infty}|e^{-t}f(t)g(t)| &\leq \int_0^{\infty} e^{-t} \sum_{k=0}^n |a_kt^k|
\end{align*}
Integrating by parts with $u = \sum_{k=0}^n |a_kt^k|$ and $dv = e^{-t}\,dt,$ we get
\begin{align*}
\int_0^{\infty} e^{-t} \sum_{k=0}^n |a_kt^k| &= uv\, \biggr|_0^{\infty} - \int_0^{\infty}v\,du
\\
&= \left[-e^{-t} \sum_{k=0}^n |a_kt^k|\right]_0^{\infty} + \int_0^{\infty} e^{-t} \sum_{k=1}^n k\left|a_kt^{k-1}\right|
\\
&= |a_0| + \int_0^{\infty} e^{-t} \sum_{k=1}^n k\left|a_kt^{k-1}\right|
\end{align*}
Performing integration by parts on the rightmost term with $u = \sum_{k=1}^n k\left| a_kt^{k-1}\right|$ and $dv = e^{-t}\,dt,$ we get
\begin{align*}
\int_0^{\infty} e^{-t} \sum_{k=0}^n |a_kt^k| &= |a_0| + \int_0^{\infty} e^{-t} \sum_{k=1}^n k\left|a_k t^{k-1}\right|
\\
&= \left|a_0\right|(0!) + \left|a_1 \right|(1!) + \int_0^{\infty} e^{-t} \sum_{k=2}^n k(k-1)\left|a_kt^{k-2}\right|
\end{align*}
Continuing this process for $k = 2, \dots, n,$ we find that
\begin{align*}
\int_0^{\infty} e^{-t} \sum_{k=0}^n |a_kt^k| &= \sum_{k=0}^n \left|a_k\right|k!
\end{align*}
But this means that the improper integral $\int_0^{\infty}|e^{-t}f(t)g(t)|\,dt$ is bounded above by $\sum_{k=0}^n \left|a_k\right|k!,$ hence the improper integral $\int_0^{\infty}e^{-t}f(t)g(t)\,dt$ converges absolutely for any real polynomials $f(t)$ and $g(t).\quad \blacksquare$
$\text{(b)}\quad$ The inner product $(x_n, x_m)$ is given by:
\begin{align*}
(x_n, x_m) &= \int_0^{\infty}e^{-t}x_n(t)x_m(t)\,dt
\\
&= \int_0^{\infty}e^{-t}t^{m + n}\,dt
\end{align*}
As in part (a), we can integrate by parts with $u = t^{m + n}$ and $dv = e^{-t}\,dt$ to get:
\begin{align*}
\int_0^{\infty} e^{-t} t^{m + n}\,dt &= uv\, \biggr|_0^{\infty} - \int_0^{\infty}v\,du
\\
&= \left[-e^{-t}t^{m + n}\right]_0^{\infty} +
\int_0^{\infty}(m + n)e^{-t}t^{(m + n - 1)}\,dt
\\
&= (m + n)\int_0^{\infty}e^{-t}t^{(m + n - 1)}\,dt
\end{align*}
Integrating by parts once again with $u = t^{m + n - 1}$ and $dv = e^{-t}\,dt,$ we get:
\begin{align*}
(m + n)\int_0^{\infty}e^{-t}t^{(m + n - 1)}\,dt &=(m + n) \left[uv\, \biggr|_0^{\infty} - \int_0^{\infty}v\,du\right]
\\
&= (m + n)\left[-e^{-t}t^{m + n - 1}\right]_0^{\infty} +
(m + n)(m + n - 1)\int_0^{\infty}e^{-t}t^{(m + n - 2)}\,dt
\\
&= (m + n)(m + n - 1)\int_0^{\infty}e^{-t}t^{(m + n - 2)}\,dt
\end{align*}
If we repeat this process an additional $(m + n - 2)$ times, we reach the following integral for $(x_n, x_m):$
\begin{align*}
(x_n, x_m) &= (m + n)!\int_0^{\infty}e^{-t}\,dt
\\
&= (m + n)!\left[-e^{-t}\right]_{0}^{\infty}
\\
&= (m + n)!(0 + 1)
\\
&= (m + n)!
\quad
\blacksquare
\end{align*}
$\text{(c)}\quad$ When $f(t) = (1 + t)^2 = 1 + 2t + t^2$ and $g(t) = 1 + t^2,$ we have $f(t)g(t) = 1 + 2t + 2t^2 + 2t^3 + t^4.$ And because the coefficients of $f(t)g(t)$ are all positive, we can use the solution from part (a) directly to get:
\begin{align*}
(f, g) &= \int_0^{\infty}e^{-t}f(t)g(t)
\\
&= \sum_{k=0}^n a_k k!
\\
&= 1(0!) + 2(1!) + 2(2!) + 2(3!) + 1(4!)
\\
&= 43
\quad
\blacksquare
\end{align*}
$\text{(d)}\quad$ As we have shown in parts (a) through (c), the inner product $(f, g)$ given by
\begin{align*}
(f, g) &= \int_0^{\infty} e^{-t}f(t)g(t)\,dt
\end{align*}
can be written as the following sum:
\begin{align*}
(f, g) &= \sum_{k=0}^{n}a_k k!
\end{align*}
If $f(t) = 1 + t$ and $g(t) = a + bt,$ we have the product $f(t)g(t) = a + (a+b)t + bt^2.$ For $g$ to be orthogonal to $f,$ their inner product must be zero. In other words, we wish to find real scalars $a$ and $b$ that satisfy:
\begin{align*}
a(0!) + (a + b)(1!) + b(2!) &= 2a + 3b
\\
&= 0.
\end{align*}
This relation is satisfied when $b = -2a/3$ for any real $a$ As such, for any real $a,$ $$g(t) = \left(1 - \frac{2}{3}t\right)a$$
is orthogonal to $f(t) = 1 + t.\quad\blacksquare$