- (a)
- \(V = \RR ^2\), \(\left \langle \begin{bmatrix}x_1\\ y_1\end{bmatrix}, \begin{bmatrix}x_2\\ y_2\end{bmatrix} \right \rangle = x_1y_1x_2y_2\)
- (b)
- \(V = \RR ^3\),
\(\left \langle \begin{bmatrix}x_1\\ x_2\\ x_3\end{bmatrix}, \begin{bmatrix}y_1\\ y_2\\ y_3\end{bmatrix} \right \rangle = x_1y_1 - x_2y_2 + x_3y_3\)Click the arrow to see the answer.
Property (e) fails. - (c)
- \(V = \mathbb {C}\), \(\langle z, w \rangle = z\overline {w}\), where \(\overline {w}\) is complex conjugation
Click the arrow to see the answer.
Property (a) fails, as sometimes we get a complex number. However, we will return to this definition of \(\langle \ , \rangle \) in Complex Matrices – see Definition ?? - (d)
- \(V = \mathbb {P}^3\), \(\langle p(x), q(x) \rangle = p(1)q(1)\)
Click the arrow to see the answer.
Property (e) fails. - (e)
- \(V = \mathbb {M}_{22}\), \(\langle A, B \rangle = \mbox {det}(AB)\)
- (f)
- \(V = \mathcal {F}[0, 1]\), \(\langle f, g \rangle = f(1)g(0) + f(0)g(1)\)
Click the arrow to see the answer.
Property (e) fails.
Inner Product Spaces
We have used the dot product in \(\RR ^n\) to compute the length of vectors (Corollary cor:length_via_dotprod) and also the angle between vectors. The goal of this section is to define an inner product on an arbitrary vector space \(V\) over the real numbers. The dot product is an example an inner product for \(\RR ^n\).
- (a)
- \(\langle \vec {v}, \vec {w}\rangle \) is a real number for all \(\vec {v}\) and \(\vec {w}\) in \(V\).
- (b)
- \(\langle \vec {v}, \vec {w}\rangle = \langle \vec {w}, \vec {v}\rangle \) for all \(\vec {v}\) and \(\vec {w}\) in \(V\).
- (c)
- \(\langle \vec {v} + \vec {w}, \vec {u}\rangle = \langle \vec {v}, \vec {u}\rangle + \langle \vec {w}, \vec {u}\rangle \) for all \(\vec {u}\), \(\vec {v}\), and \(\vec {w}\) in \(V\).
- (d)
- \(\langle r\vec {v}, \vec {w}\rangle = r\langle \vec {v}, \vec {w}\rangle \) for all \(\vec {v}\) and \(\vec {w}\) in \(V\) and all \(r\) in \(\RR \).
- (e)
- \(\langle \vec {v}, \vec {v}\rangle > 0\) for all \(\vec {v} \neq \vec {0}\) in \(V\).
A real vector space \(V\) with an inner product \(\langle \ , \rangle \) will be called an inner product space. Note that every subspace of an inner product space is again an inner product space using the same inner product.
The next example is important in analysis.
If \(\vec {v}\) is any vector, then, using Property prop:inner_prod_3 of Definition def:innerproductspace, we get
and it follows that the number \(\langle \vec {0}, \vec {v}\rangle \) must be zero. This observation is recorded for reference in the following theorem, along with several other properties of inner products. The other proofs are left as Practice Problem ex:10_1_20.
- (a)
- \(\langle \vec {u}, \vec {v} + \vec {w}\rangle = \langle \vec {u}, \vec {v}\rangle + \langle \vec {u}, \vec {w}\rangle \)
- (b)
- \(\langle \vec {v}, r\vec {w}\rangle = r\langle \vec {v}, \vec {w}\rangle = \langle r\vec {v}, \vec {w}\rangle \)
- (c)
- \(\langle \vec {v}, \vec {0}\rangle = 0 = \langle \vec {0}, \vec {v} \rangle \)
- (d)
- \(\langle \vec {v}, \vec {v}\rangle = 0\) if and only if \(\vec {v} = \vec {0}\)
If \(\langle \ , \rangle \) is an inner product on a space \(V\), then, given \(\vec {u}\), \(\vec {v}\), and \(\vec {w}\) in \(V\),
for all \(r\) and \(s\) in \(\RR \) by Property prop:inner_prod_3 and Property prop:inner_prod_4 of Definition def:innerproductspace. Moreover, there is nothing special about the fact that there are two terms in the linear combination or that it is in the first component:
and
hold for all \(r_{i}\) and \(s_{i}\) in \(\RR \) and all \(\vec {v}\), \(\vec {w}\), \(\vec {v}_{i}\), and \(\vec {w}_{j}\) in \(V\). These results are described by saying that inner products “preserve” linear combinations. For example,
If \(A\) is a symmetric \(n \times n\) matrix and \(\vec {x}\) and \(\vec {y}\) are columns in \(\RR ^n\), we regard the \(1 \times 1\) matrix \(\vec {x}^{T}A\vec {y}\) as a number. If we write
then Properties prop:inner_prod_1 -prop:inner_prod_4 of Definition def:innerproductspace follow from matrix arithmetic (only Property prop:inner_prod_2 of Definition def:innerproductspace requires that \(A\) is symmetric). Property prop:inner_prod_5 of Definition def:innerproductspace reads
and this condition characterizes the positive definite matrices (Theorem thm:024830). This proves the first assertion in the next theorem.
Given an inner product \(\langle \ , \rangle \) on \(\RR ^n\), let \(\{\vec {e}_{1}, \vec {e}_{2}, \dots , \vec {e}_{n}\}\) be the standard basis of \(\RR ^n\). If \(\vec {x} = \displaystyle \sum _{i = 1}^{n} x_i\vec {e}_i\) and \(\vec {y} = \displaystyle \sum _{j = 1}^{n} y_j\vec {e}_j\) are two vectors in \(\RR ^n\), compute \(\langle \vec {x}, \vec {y}\rangle \) by adding the inner product of each term \(x_{i}\vec {e}_{i}\) to each term \(y_{j}\vec {e}_{j}\). The result is a double sum.
Thus, just as every linear operator \(\RR ^n \to \RR ^n\) corresponds to an \(n \times n\) matrix, every inner product on \(\RR ^n\) corresponds to a positive definite \(n \times n\) matrix. In particular, the dot product corresponds to the identity matrix \(I_{n}\).
for all \(\vec {x}\), so \(\langle \vec {x}, \vec {x}\rangle = 0\) implies \(\vec {x} = \vec {0}\). Hence \(\langle \ , \rangle \) is indeed an inner product, so \(A\) is positive definite.
Let \(\langle \ , \rangle \) be an inner product on \(\RR ^n\) given as in Theorem thm:030372 by a positive definite matrix \(A\). If \(\vec {x} = \left [ \begin{array}{cccc} x_1 & x_2 & \cdots & x_n \end{array} \right ]^T \), then \(\langle \vec {x}, \vec {x}\rangle = \vec {x}^{T}A\vec {x}\) is an expression in the variables \(x_{1}, x_{2}, \dots , x_{n}\) called a quadratic form. For more on quadratic forms, see Section 8.8 of [Nicholson], pp. 472–482.
Norm and Distance
Note that Property (e) of Definition 1 guarantees that \(\langle \vec {v}, \vec {v}\rangle \geq 0\), so \(\norm {\vec {v}}\) is a real number.
The norm of a continuous function \(f = f(x)\) in \(\mathcal {C}[a, b]\) (with the inner product from Example 5) is given by
A vector \(\vec {v}\) in an inner product space \(V\) is called a unit vector if \(\norm {\vec {v}} = 1\). The set of all unit vectors in \(V\) is called the unit ball in \(V\). For example, if \(V = \RR ^2\) (with the dot product) and \(\vec {v} = (x, y)\), then
Hence the unit ball in \(\RR ^2\) is the unit circle \(x^{2} + y^{2} = 1\) with centre at the origin and radius \(1\). However, the shape of the unit ball varies with the choice of inner product.
Let \(a > 0\) and \(b > 0\). If \(\vec {v} = (x, y)\) and \(\vec {w} = (x_{1}, y_{1})\), define an inner product on \(\RR ^2\) by
Example 20 graphically illustrates the fact that norms and distances in an inner product space \(V\) vary with the choice of inner product in \(V\).
The next theorem reveals an important and useful fact about the relationship between norms and inner products, extending the Cauchy inequality for \(\RR ^n\) (Theorem ??).
Proof of Cauchy-Schwarz Inequality Write \(\norm {\vec {v}} = a\) and \(\norm {\vec {w}} = b\). Using Theorem 8 we compute:
Conversely, if \(|\langle \vec {v}, \vec {w}\rangle | = \norm {\vec {v}} \norm { \vec {w} } = ab\) then \(\langle \vec {v}, \vec {w}\rangle = \pm ab\). Hence (1) shows that \(b\vec {v} - a\vec {w} = \vec {0}\) or \(b\vec {v} + a\vec {w} = \vec {0}\). It follows that one of \(\vec {v}\) and \(\vec {w}\) is a scalar multiple of the other, even if \(a = 0\) or \(b = 0\).
Another famous inequality, the so-called triangle inequality (See Triangle Inequality in the Appendix), also comes from the Cauchy-Schwarz inequality. It is included in the following list of basic properties of the norm of a vector.
- (a)
- \(\norm {\vec {v}} \geq 0\) for every vector \(\vec {v}\) in \(V\).
- (b)
- \(\norm {\vec {v}} = 0\) if and only if \(\vec {v} = \vec {0}\).
- (c)
- \(\norm { r \vec {v}} = |r|\norm {\vec {v}}\) for every \(\vec {v}\) in \(V\) and every \(r\) in \(\RR \).
- (d)
- \(\norm {\vec {v} + \vec {w}} \leq \norm {\vec {v}} + \norm {\vec {w}}\) for all \(\vec {v}\) and \(\vec {w}\) in \(V\) (triangle inequality).
Because \(\norm { \vec {v} } = \sqrt {\langle \vec {v}, \vec {v} \rangle }\), properties (a) and (b) follow immediately from ?? and ?? of Theorem 8. As to (c), compute
Hence (d) follows by taking positive square roots.
It is worth noting that the usual triangle inequality for absolute values,
is a special case of (d) where \(V = \RR = \RR ^1\) and the dot product \(\langle r, s \rangle = rs\) is used.
In many calculations in an inner product space, it is required to show that some vector \(\vec {v}\) is zero. This is often accomplished most easily by showing that its norm \(\norm {\vec {v}}\) is zero. Here is an example.
The norm properties in Theorem 26 translate to the following properties of distance familiar from geometry. The proof is Practice Problem .
- (a)
- \(\mbox {d}(\vec {v}, \vec {w}) \geq 0\) for all \(\vec {v}\), \(\vec {w}\) in \(V\).
- (b)
- \(\mbox {d}(\vec {v}, \vec {w}) = 0\) if and only if \(\vec {v} = \vec {w}\).
- (c)
- \(\mbox {d}(\vec {v}, \vec {w}) = \mbox {d}(\vec {w}, \vec {v})\) for all \(\vec {v}\) and \(\vec {w}\) in \(V\).
- (d)
- \(\mbox {d}(\vec {v}, \vec {w}) \leq \mbox {d}(\vec {v}, \vec {u}) + \mbox {d}(\vec {u}, \vec {w})\) for all \(\vec {v}\), \(\vec {u}\), and \(\vec {w}\) in \(V\).
Practice Problems
- (a)
- \(\vec {v} = f\) in \(\mathcal {C}[0, 1]\) where \(f(x) = x^2\)
\(\langle f, g \rangle \int _{0}^{1} f(x)g(x)dx\) - (b)
- \(\vec {v} = f\) in \(\mathcal {C}[-\pi , \pi ]\) where \(f(x) = \cos x\)
\(\langle f, g \rangle \int _{-\pi }^{\pi } f(x)g(x)dx\)Click the arrow to see the answer.
\(\frac {1}{\sqrt {\pi }}f\) - (c)
- \(\vec {v} = \left [ \begin{array}{r} 1 \\ 3 \end{array} \right ]\) in \(\RR ^2\) where \(\langle \vec {v}, \vec {w} \rangle = \vec {v}^T \left [ \begin{array}{rr} 1 & 1 \\ 1 & 2 \end{array} \right ] \vec {w}\)
- (d)
- \( \vec {v} = \left [ \begin{array}{r} 3 \\ -1 \end{array} \right ]\) in \(\RR ^2\), \(\langle \vec {v}, \vec {w} \rangle = \vec {v}^T \left [ \begin{array}{rr} 1 & -1 \\ -1 & 2 \end{array} \right ] \vec {w}\)
Click the arrow to see the answer.
\(\frac {1}{\sqrt {17}} \left [ \begin{array}{r} 3 \\ -1 \end{array} \right ]\)
- (a)
- \(\vec {u} = \begin{bmatrix}3\\ -1\\ 2\\ 0\end{bmatrix}\), \(\vec {v} = \begin{bmatrix}1\\ 1\\ 1\\ 3\end{bmatrix}; \langle \vec {u}, \vec {v} \rangle = \vec {u} \dotp \vec {v}\)
- (b)
- \(\vec {u} = \begin{bmatrix}1\\ 2\\ -1\\ 2\end{bmatrix}\), \(\vec {v} = \begin{bmatrix}2\\ 1\\ -1\\ 3\end{bmatrix}; \langle \vec {u}, \vec {v} \rangle = \vec {u} \dotp \vec {v}\)
Click the arrow to see the answer.
\(\sqrt {3}\) - (c)
- \(\vec {u} = f\), \(\vec {v} = g \) in \(\mathcal {C}[0, 1]\) where \(f(x) = x^2 \) and \(g(x) = 1 - x\); \(\langle f, g \rangle = \int _{0}^{1} f(x)g(x)dx\)
- (d)
- \(\vec {u} = f\), \(\vec {v} = g \) in \(\mathcal {C}[-\pi , \pi ]\) where \(f(x) = 1\) and \(g(x) = \cos x\); \(\langle f, g \rangle = \int _{-\pi }^{\pi } f(x)g(x)dx\)
Click the arrow to see the answer.
\(\sqrt {3\pi }\)
- (a)
- \(A = \left [ \begin{array}{rr} 2 & 1 \\ 1 & 1 \end{array} \right ]\)
- (b)
- \(A = \left [ \begin{array}{rr} 5 & -3 \\ -3 & 2 \end{array} \right ]\)
Click the arrow to see the answer.
\( \langle \vec {v}, \vec {v} \rangle = 5v_1^2 - 6v_1v_2 + 2v_2^2 = \frac {1}{5}[(5v_1 - 3v_2)^2 + v_2^2] \) - (c)
- \(A = \left [ \begin{array}{rr} 3 & 2 \\ 2 & 3 \end{array} \right ]\)
- (d)
- \(A = \left [ \begin{array}{rr} 3 & 4 \\ 4 & 6 \end{array} \right ]\)
Click the arrow to see the answer.
- (a)
- \( \left \langle \left [ \begin{array}{r} v_1 \\ v_2 \end{array} \right ], \left [ \begin{array}{r} w_1 \\ w_2 \end{array} \right ] \right \rangle = v_1w_1 + 2v_1w_2 + 2v_2w_1 + 5v_2w_2\)
- (b)
- \(\left \langle \left [ \begin{array}{r} v_1 \\ v_2 \end{array} \right ], \left [ \begin{array}{r} w_1 \\ w_2 \end{array} \right ] \right \rangle = v_1w_1 - v_1w_2 - v_2w_1 + 2v_2w_2\)
Click the arrow to see the answer.
\(\left [ \begin{array}{rr} 1 & -2 \\ -2 & 1 \end{array} \right ]\) - (c)
- \(\left \langle \left [ \begin{array}{r} v_1 \\ v_2 \\ v_3 \end{array} \right ], \left [ \begin{array}{r} w_1 \\ w_2 \\ w_3 \end{array} \right ] \right \rangle = 2v_1w_1 + v_2w_2 + v_3w_3 - v_1w_2 \\ -v_2w_1 + v_2w_3 + v_3w_2\)
- (d)
- \(\left \langle \left [ \begin{array}{r} v_1 \\ v_2 \\ v_3 \end{array} \right ], \left [ \begin{array}{r} w_1 \\ w_2 \\ w_3 \end{array} \right ] \right \rangle = v_1w_1 + 2v_2w_2 + 5v_3w_3 \\ - 2v_1w_3 - 2v_3w_1\)
Click the arrow to see the answer.
\(\left [ \begin{array}{rrr} 1 & 0 & -2 \\ 0 & 2 & 0 \\ -2 & 0 & 5 \end{array} \right ]\)
Click the arrow to see the answer.
- (a)
- \(\langle \vec {v} + \vec {w}, 2\vec {u} - \vec {v} \rangle \)
- (b)
- \(\langle \vec {u} - 2 \vec {v} - \vec {w}, 3\vec {w} - \vec {v} \rangle \)
\(\answer {-15}\)
- (a)
- Using Property (b): \( \langle \vec {u}, \vec {v} + \vec {w} \rangle = \langle \vec {v} + \vec {w}, \vec {u} \rangle = \langle \vec {v}, \vec {u} \rangle + \langle \vec {w}, \vec {u} \rangle = \langle \vec {u}, \vec {v} \rangle + \langle \vec {u}, \vec {w} \rangle \).
- (b)
- Using Property (b) and Property (d): \(\langle \vec {v}, r\vec {w} \rangle = \langle r\vec {w}, \vec {v} \rangle = r \langle \vec {w}, \vec {v} \rangle = r \langle \vec {v}, \vec {w} \rangle \).
- (c)
- Using Property (c): \(\langle \vec {0}, \vec {v} \rangle = \langle \vec {0} + \vec {0}, \vec {v} \rangle = \langle \vec {0}, \vec {v} \rangle + \langle \vec {0}, \vec {v} \rangle \), so \( \langle \vec {0}, \vec {v} \rangle = 0\). The rest is Property (b).
- (d)
- Assume that \(\langle \vec {v}, \vec {v} \rangle = 0\). If \(\vec {v} \neq \vec {0}\) this contradicts Property (e), so \(\vec {v} = \vec {0}\). Conversely, if \(\vec {v} = \vec {0}\), then \(\langle \vec {v}, \vec {v} \rangle = 0\) by Part 3 of this theorem.
- (a)
- Expand \(\langle 2\vec {u} - 7\vec {v}, 3\vec {u} + 5\vec {v} \rangle \).
- (b)
- Expand \(\langle 3\vec {u} - 4\vec {v}, 5\vec {u} + \vec {v} \rangle \).
Click the arrow to see the answer.
\(15\norm {\vec {u}}^{2} - 17 \langle \vec {u}, \vec {v} \rangle - 4\norm {\vec {v}}^{2}\) - (c)
- Show that \(\norm { \vec {u} + \vec {v} } ^2 = \norm { \vec {u} } ^2 + 2 \langle \vec {u}, \vec {v} \rangle + \norm { \vec {v} } ^2 \).
- (d)
- Show that \(\norm { \vec {u} - \vec {v} } ^2 = \norm { \vec {u} } ^2 - 2 \langle \vec {u}, \vec {v} \rangle + \norm { \vec {v} } ^2\).
Click the arrow to see the answer.
\(\norm {\vec {u} + \vec {v}}^{2} = \langle \vec {u} + \vec {v}, \vec {u} + \vec {v} \rangle = \norm {\vec {u}}^{2} + 2\langle \vec {u}, \vec {v}\rangle + \norm {\vec {v}}^{2}\)
\(\mbox {d}(\vec {v}, \vec {w}) = \mbox {d}(\vec {v} + \vec {u}, \vec {w} + \vec {u})\) for all \(\vec {v}\), \(\vec {w}\), and \(\vec {u}\) in \(V\).
- (a)
- Show that \(\langle \vec {u}, \vec {v} \rangle = \frac {1}{4}[\norm { \vec {u} + \vec {v} } ^2 - \norm { \vec {u} - \vec {v} } ^2]\) for all \(\vec {u}\), \(\vec {v}\) in an inner product space \(V\).
- (b)
- If \(\langle \ , \rangle \) and \(\langle \ , \rangle ^\prime \) are two inner products on \(V\) that have equal associated norm functions, show that \(\langle \vec {u}, \vec {v}\rangle = \langle \vec {u}, \vec {v}\rangle ^\prime \) holds for all \(\vec {u}\) and \(\vec {v}\).
- (a)
- Show that \(W = \{\vec {w} \mid \vec {w} \mbox { in } V, \langle \vec {v}, \vec {w} = 0\}\) is a subspace of \(V\).
- (b)
- Let \(W\) be as in (a). If \(V = \RR ^3\) with the dot product, and if \(\vec {v} = \begin{bmatrix}1\\ -1\\ 2\end{bmatrix}\), find a basis for \(W\).
Click the arrow to see the answer.
\(\left \{\begin{bmatrix}1\\ 1\\ 0\end{bmatrix}, \begin{bmatrix}0\\ 2\\ 1\end{bmatrix}\right \}\)
- (a)
- If \(\norm {\vec {u}} \leq 1\), then \(\langle \vec {u}, \vec {v}\rangle ^{2} \leq \norm {\vec {v}}^{2}\) for all \(\vec {v}\) in \(V\).
- (b)
- \((x \cos \theta + y \sin \theta )^{2} \leq x^{2} + y^{2}\) for all real \(x\), \(y\), and \(\theta \).
If \(\vec {u} = (\cos \theta , \sin \theta )\) in \(\RR ^2\) (with the dot product) then \(\norm {\vec {u}} = 1\). Use (a) with \(\vec {v} = \begin{bmatrix}x\\ y\end{bmatrix}\).
- (c)
- \(\norm { r_1\vec {v}_1 + \dots + r_n\vec {v}_n } ^2 \leq [r_1 \norm { \vec {v}_1 } + \dots + r_n \norm { \vec {v}_n } ]^2\) for all vectors \(\vec {v}_{i}\), and all \(r_{i} > 0\) in \(\RR \).
- (a)
- Show that \(AA^T = \left [ \begin{array}{rr} \norm { \vec {u} } ^2 & \vec {u} \dotp \vec {v} \\ \vec {u} \dotp \vec {v} & \norm { \vec {v} } ^2 \end{array} \right ]\).
- (b)
- Show that \(\mbox {det}(AA^{T}) \geq 0\).
- (a)
- If \(\vec {v}\) and \(\vec {w}\) are nonzero vectors in an inner product space \(V\), show that \(-1 \leq \frac {\langle \vec {v}, \vec {w} \rangle }{\norm { \vec {v} } \norm { \vec {w} }} \leq 1\), and
hence that a unique angle \(\theta \) exists such that
\(\frac {\langle \vec {v}, \vec {w} \rangle }{\norm { \vec {v} } \norm { \vec {w} }} = \cos \theta \) and \(0 \leq \theta \leq \pi \). This angle \(\theta \) is called the angle between \(\vec {v}\) and \(\vec {w}\). - (b)
- Find the angle between \(\vec {v} = \begin{bmatrix}1\\ 2\\ -1\\ 1\\ 3\end{bmatrix}\) and \(\vec {w} = \begin{bmatrix}2\\ 1\\ 0\\ 2\\ 0\end{bmatrix}\) in \(\RR ^5\) with the dot product.
- (c)
- If \(\theta \) is the angle between \(\vec {v}\) and \(\vec {w}\), show that the law of cosines is valid:
\begin{equation*} \norm { \vec {v} - \vec {w} } = \norm { \vec {v} } ^2 + \norm { \vec {w} } ^2 - 2\norm { \vec {v} } \norm { \vec {w} } \cos \theta . \end{equation*}
- (a)
- Show that \(\norm {\dotp }\) satisfies the conditions in Theorem 26.
- (b)
- Show that \(\norm {\dotp }\) does not arise from an inner product on \(\RR ^2\) given by a matrix \(A\). If it did, use Theorem 10 to find numbers \(a\), \(b\), and \(c\) such that \(\norm {\begin{bmatrix}x\\ y\end{bmatrix}}^{2} = ax^{2} + bxy + cy^{2}\) for all \(x\) and \(y\).
Text Source
This section was adapted from Section 10.1 of Keith Nicholson’s Linear Algebra with Applications. (CC-BY-NC-SA)
W. Keith Nicholson, Linear Algebra with Applications, Lyryx 2018, Open Edition, pp. 521–530.