Why Complex Angles Need a Phase-Invariant Definition
In a real inner product space, the normalized inner product of two nonzero vectors gives the cosine of their angle. Over \(\mathbb{C}\), an inner product can itself be complex, so it cannot directly be the cosine of a real angle. Its magnitude, however, is a real number, and the complex Cauchy–Schwarz inequality bounds that magnitude by the product of the vector norms. This gives a natural definition whose value is unchanged by complex phase.
We use the convention that a complex inner product is linear in its first slot and conjugate-linear in its second. Thus, for \(\mathbf{u},\mathbf{v}\in\mathbb{C}^n\), the standard inner product is \(\langle\mathbf{u},\mathbf{v}\rangle=\sum_{j=1}^n u_j\overline{v_j}\). It satisfies conjugate symmetry, \(\langle\mathbf{v},\mathbf{u}\rangle=\overline{\langle\mathbf{u},\mathbf{v}\rangle}\), and induces the norm \(\|\mathbf{u}\|=\sqrt{\langle\mathbf{u},\mathbf{u}\rangle}\). The same angle definition applies in any complex inner product space, not only \(\mathbb{C}^n\).
The range in the definition follows from the complex Cauchy–Schwarz inequality: for vectors in a complex inner product space, \(|\langle\mathbf{u},\mathbf{v}\rangle|\leq\|\mathbf{u}\|\|\mathbf{v}\|\). This is the Complex Cauchy–Schwarz Inequality established earlier in the course. For nonzero vectors, dividing by the positive product \(\|\mathbf{u}\|\|\mathbf{v}\|\) gives a number in \([0,1]\). The inverse cosine of a number in \([0,1]\) belongs to \([0,\pi/2]\), so the formula is real-valued and well-defined.
Computing the Angle in \(\mathbb{C}^n\)
Worked Example: A Nonzero Complex Inner Product
Find the angle between \(\mathbf{u}=(1,i)\) and \(\mathbf{v}=(1,1)\) in \(\mathbb{C}^2\), with the standard inner product.
Using linearity in the first slot and conjugation in the second,
The squared norms are \(\|\mathbf{u}\|^2=1\cdot\overline{1}+i\cdot\overline{i}=1+1=2\) and \(\|\mathbf{v}\|^2=1+1=2\). Hence \(\|\mathbf{u}\|=\|\mathbf{v}\|=\sqrt{2}\). Since \(|1+i|=\sqrt{1^2+1^2}=\sqrt{2}\), the normalized magnitude is
Therefore \(\theta(\mathbf{u},\mathbf{v})=\arccos(\sqrt{2}/2)=\pi/4\). As a check, \(\cos(\pi/4)=\sqrt{2}/2\), exactly the normalized magnitude calculated from the two vectors.
The absolute value is essential. The inner product records phase as well as size, whereas the angle is designed not to change just because a vector is multiplied by a complex number of magnitude one. Such multiplication is called a phase change. In fact, the angle is unchanged even if either vector is multiplied by any nonzero complex scalar.
Proof. By conjugate-linearity in the second slot, \(\langle\alpha\mathbf{u},\beta\mathbf{v}\rangle=\alpha\overline{\beta}\langle\mathbf{u},\mathbf{v}\rangle\), so its magnitude is \(|\alpha||\beta||\langle\mathbf{u},\mathbf{v}\rangle|\). The induced norm satisfies \(\|\alpha\mathbf{u}\|=|\alpha|\|\mathbf{u}\|\) and \(\|\beta\mathbf{v}\|=|\beta|\|\mathbf{v}\|\). Therefore the normalized magnitude for \(\alpha\mathbf{u},\beta\mathbf{v}\) is \[ \frac{|\alpha||\beta||\langle\mathbf{u},\mathbf{v}\rangle|}{|\alpha|\|\mathbf{u}\|\,|\beta|\|\mathbf{v}\|} =\frac{|\langle\mathbf{u},\mathbf{v}\rangle|}{\|\mathbf{u}\|\|\mathbf{v}\|}. \] Applying \(\arccos\) proves the first claim. For the second, conjugate symmetry gives \(|\langle\mathbf{v},\mathbf{u}\rangle|=|\overline{\langle\mathbf{u},\mathbf{v}\rangle}|=|\langle\mathbf{u},\mathbf{v}\rangle|\), and the norm product is unchanged when the vectors are interchanged. The angle formula therefore gives the same value in either order. \(\square\)
Worked Example: A Phase Change Does Not Change the Angle
Let \(\mathbf{u}=(1,2i)\) and let \(\mathbf{v}=i\mathbf{u}=(i,-2)\). Compute their angle directly.
First, \(\|\mathbf{u}\|^2=1+|2i|^2=5\) and \(\|\mathbf{v}\|^2=|i|^2+|-2|^2=5\), so both norms are \(\sqrt{5}\). The inner product, with conjugation in the second slot, is
Thus \(|\langle\mathbf{u},\mathbf{v}\rangle|=5\), and the normalized magnitude is \(5/(\sqrt{5}\sqrt{5})=1\). It follows that \(\theta(\mathbf{u},\mathbf{v})=\arccos(1)=0\). The computation checks the phase-invariance result: \(\mathbf{v}\) is \(\mathbf{u}\) multiplied by the phase \(i\), and the angle is zero.
What the Endpoint Values Mean
The angle can equal \(0\) or \(\pi/2\), but it cannot be obtuse. The zero endpoint occurs precisely when the vectors are complex scalar multiples of one another. In this setting, “scalar multiple” means a multiple by a complex scalar, not necessarily a real one. The other endpoint occurs precisely when the inner product is zero.
Proof. Write \(s=\|\mathbf{u}\|^2>0\) and \(a=\langle\mathbf{v},\mathbf{u}\rangle\), and set \(c=a/s\). Expanding the squared norm using linearity in the first slot and conjugate-linearity in the second gives \[ \|\mathbf{v}-c\mathbf{u}\|^2 =\|\mathbf{v}\|^2-\overline{c}\langle\mathbf{v},\mathbf{u}\rangle -c\langle\mathbf{u},\mathbf{v}\rangle+|c|^2\|\mathbf{u}\|^2 =\|\mathbf{v}\|^2-\frac{|a|^2}{s}. \] Here \(\langle\mathbf{u},\mathbf{v}\rangle=\overline{a}\), so each of the two middle terms contributes \(-|a|^2/s\), while the last term contributes \(+|a|^2/s\). If the angle is zero, the definition implies \(|\langle\mathbf{u},\mathbf{v}\rangle|=\|\mathbf{u}\|\|\mathbf{v}\|\). Since \(|a|=|\langle\mathbf{u},\mathbf{v}\rangle|\), the expansion becomes \(\|\mathbf{v}-c\mathbf{u}\|^2=\|\mathbf{v}\|^2-\|\mathbf{v}\|^2=0\). Positive definiteness of the inner product then gives \(\mathbf{v}=c\mathbf{u}\). Conversely, if \(\mathbf{v}=c\mathbf{u}\), then \(c\ne0\) because \(\mathbf{v}\ne\mathbf{0}\). Consequently \(|\langle\mathbf{u},\mathbf{v}\rangle|=|c|\|\mathbf{u}\|^2\) and \(\|\mathbf{v}\|=|c|\|\mathbf{u}\|\), making the normalized magnitude \(1\) and the angle zero. Finally, \(\arccos(x)=\pi/2\) for \(x\in[0,1]\) exactly when \(x=0\). The normalized magnitude is zero exactly when \(|\langle\mathbf{u},\mathbf{v}\rangle|=0\), which is equivalent to \(\langle\mathbf{u},\mathbf{v}\rangle=0\). \(\square\)
Worked Example: The Maximum Complex Angle
Let \(\mathbf{u}=(1,i)\) and \(\mathbf{v}=(1,-i)\) in \(\mathbb{C}^2\). Find their angle.
Their squared norms are \(\|\mathbf{u}\|^2=1+|i|^2=2\) and \(\|\mathbf{v}\|^2=1+|-i|^2=2\). Their inner product is
Thus the normalized magnitude is \(0/(\sqrt{2}\sqrt{2})=0\), so \(\theta(\mathbf{u},\mathbf{v})=\arccos(0)=\pi/2\). Substitution back into the formula confirms that \(\cos(\pi/2)=0\), matching the normalized magnitude.
Common Mistakes with Complex Inner Products
A common error is to replace the magnitude of the normalized inner product with its real part. This can make a phase change appear to alter the angle. For example, take \(\mathbf{u}=(1,0)\) and \(\mathbf{v}=(i,0)=i\mathbf{u}\). The correct inner product is \(\langle\mathbf{u},\mathbf{v}\rangle=-i\), so its magnitude is \(1\); both norms are \(1\), and the angle is \(\arccos(1)=0\). Using the real part instead would give \(\operatorname{Re}(-i)=0\) and incorrectly suggest an angle of \(\pi/2\). The phase \(i\) changes the inner product but not its magnitude.
Another frequent mistake is to omit conjugation in the complex inner product. With \(\mathbf{u}=(1,i)\), omitting conjugation and calculating \(1\cdot1+i\cdot i=0\) would wrongly suggest the largest possible angle between \(\mathbf{u}\) and itself. The correct calculation is \(\langle\mathbf{u},\mathbf{u}\rangle=1+i\overline{i}=2\), as required by the norm definition. Check the convention for the inner product before computing, then take the magnitude of the resulting complex number.
The phase-invariance theorem also gives a useful interpretation: the angle depends on the complex directions represented by the vectors, not on the particular nonzero scalar multiples chosen to represent them. In particular, a unit complex scalar can change the inner product’s phase while leaving the angle unchanged. The absolute value in the definition is exactly what removes that irrelevant phase information.
Check Your Understanding
Use the convention that the inner product is linear in its first slot. For each calculation, keep track of conjugation before taking the magnitude.
- Find the angle between \((1,0)\) and \((1,i)\) in \(\mathbb{C}^2\).
- Let \(\mathbf{v}\ne\mathbf{0}\) and \(\alpha\in\mathbb{C}\setminus\{0\}\). What does phase invariance say about the angle between \(\mathbf{v}\) and \(\alpha\mathbf{v}\)?
- What condition on \(\langle\mathbf{u},\mathbf{v}\rangle\) is equivalent to the angle being \(\pi/2\)?
- Explain why the angle formula must use \(|\langle\mathbf{u},\mathbf{v}\rangle|\), rather than the real part of the inner product.
- For \(\mathbf{u}=(1,i)\) and \(\mathbf{v}=(1,-i)\), compute the inner product and verify the angle from its normalized magnitude.
