Lecture 29/30: Inner Product Spaces and Norms
Let \(V\) be. vector space over \(\mathbf{F} = \mathbf{R}\) or \(\mathbf{C}\).
The complex conjugate of \(z\) is \(\bar{z} = a - ib\).
- \(\langle x+z, y \rangle = \langle x , y \rangle + \langle z , y \rangle\)
- \(\langle cx, y \rangle = c \langle x , y \rangle\)
- \(\langle x, y \rangle = \overline{\langle y, x \rangle}\). Note if \(\mathbf{F} = \mathbf{R}\), then \(\langle x, y \rangle = \langle y, x \rangle\)
- \(\langle x, x \rangle > 0 \text{ if } x \neq \bar{0}\)
Example 1
The simplest example is \(V = \mathbf{R}^1\) (vector space over \(\mathbf{R}\)) where
This map satisfies the inner product properties. Notice that
- \(\langle x+z, y \rangle = (x+z)y = xy + zx = \langle x , y \rangle + \langle z , y \rangle \)
- \(\langle cx, y \rangle = cxy = c\langle x, y \rangle\)
- \(\langle x, y \rangle = \overline{\langle y, x \rangle} = yx = xy\).
- \(\langle x, x \rangle > 0 \text{ if } x \neq \bar{0}\)
Example 2
We can also define this inner product
which also satisfies the inner product properties (TODO)
Example 3: Dot Product
Another example is \(V = \mathbf{R}^n\) over \(\mathbf{R}\) where
which is commonly known as the dot product.
Example 4
Another example is \(V = \mathbf{R}^2\) over \(\mathbf{R}\) where
We claim that \(\langle \langle \ , \ \rangle \rangle\) is an inner product. To see this check each property in the definition. For example for property 4, we see that
Example 5
Define \(V = \mathbf{C}\) over \(\mathbf{C}\) where
Note here that if we defined the product as \(\langle z_1, z_2 \rangle = z_1z_2\). This will fail to satisfy the inner product conditions.
Example 6: Frobenius Inner Product
Define \(V = \mathbf{C}^0([0,1]) = \{f: [0,1] \rightarrow \mathbf{R} \ | \ f \text{ continuous}\}\)
is an inner product. Checking property (4)
Example 7
Define \(V = M_{n \times n}(\mathbf{R})\)
where
Does this map satisfy the inner product conditions?
- We want to show that \(\langle A + C, B \rangle = \langle A, B \rangle + \langle C, B \rangle\). Expand the inner product to see that
$$ \begin{align*} \langle A+C, B \rangle &= tr(B^t(A+C)) \\ &= tr(B^tA + B^tC) \\ &= tr(B^tA) + tr(B^tC) \\ &= \langle A, B \rangle + \langle C, B \rangle \end{align*} $$
- We want to show that \(\langle cA, B \rangle = c\langle A, B \rangle\):
$$ \begin{align*} \langle cA, B \rangle &= tr(B^t(cA)) \\ &= ctr(B^tA) \\ &= c\langle A, B \rangle \end{align*} $$
- We want to show that \(\langle A, B \rangle = \langle B, A \rangle\). (over \(\mathbf{R}\)). Note here that \(tr(C^t) = tr(C)\). Then
$$ \begin{align*} \langle A, B \rangle &= tr(B^tA) \\ &= tr((B^tA)^t) \\ &= tr(A^tB) \\ &= \langle B, A \rangle \end{align*} $$
- We need to show that \(\langle A, A \rangle > 0 \text{ if } A \neq \bar{0} \in M_{n \times n}(\mathbf{R})\)
$$ \begin{align*} \langle A, A \rangle &= tr(A^tA) \\ &= \sum_j (A^tA)_{jj} \\ &= \sum_j \sum_k (A^t)_{jk} (A)_{kj} \\ &= \sum_j \sum_k (A)_{kj} (A)_{kj} \\ &= \sum_j \sum_k A^2_{kj} \end{align*} $$Note here that \(A^2_{kj} > 0\) unless \(A_{1j},...A_{nj} = 0\). These are the entries of the \(j\)th column of \(A\). This means that this is zero unless \(A\) is the zero matrix.
In fact, \(tr(B^tA) = \sum_{ij} A_{ij}B_{ij}\). the definition of \(tr(B^tA)\) is for square matrices while the definition \(\sum_{ij} A_{ij}B_{ij}\) works for any matrices. This works for an inner product on matrices.
Inner Product Spaces
So far, we’ve seen that inner products are not unique. We can define many inner products on a given vector space. So we have the following definition to fix a specific inner product on a vector space
Example: (\(\mathbf{R}^2, \langle x,y \rangle = x_1y_1 + x_2y_2\)) is an inner product space different from \((\mathbf{R}^2, \langle x,y \rangle = 2x_1y_1 + x_1y_2 + x_2y_1 + x_2y_2)\)
The Norm of a Vector
For the rest of the lecture, we’re going to assume that we have a fixed inner product on \(V\).
Example \(V = \mathbf{C}^0([0,1])\), Let \(\Vert f \Vert: (\int_0^1 f(t)^2 dt)^{1/2}\). This is also called the \(L^2\)-norm.
- \(\Vert cx \Vert = |c|\Vert x \Vert \)
- \(\Vert x \Vert = 0 \leftrightarrow x = \bar{0}_V \)
- \(| \langle x , y \rangle | \leq \Vert x \Vert \Vert y \Vert \). Cauchy-Schwarz
- \(\Vert x + y \Vert \leq \Vert x \Vert \Vert y \Vert \). The Triangle Inequality
The proof for (a) and (b) follow easily. For (c). The motivation is from \(\mathbf{R}^2\) with the standard inner product.
But we know that \(| \cos(\theta)| \leq 1\). Therefore
But we want to prove this in general so:
Proof (c):
If \(y = \bar{0}_V\), then the inequality is true and we are done. So assume that \(y \neq \bar{0}_V\). If we multiply both sides by \(\frac{1}{\Vert y \Vert}\) which is a scalar, then this scalar can be factored out by property 1. Therefore, we can scale \(y\) by whatever factor we want and so let’s just assume that \(y\) has length 1 (\(\Vert y \Vert = 1\)). So it suffies to show that
To show this, take the project of vector \(x\) onto vector \(y\) which has length 1. The projection is a vector \(x - \langle x , y \rangle y\). Moreover,
Proof (d)
References
- Math416 by Ely Kerman