Orthogonality In Complex Inner Product Spaces

Orthogonality When Scalars Are Complex

For real vectors, the inner product is a real number, so reversing the order of its arguments leaves its value unchanged. In a complex inner product space, an inner product can itself be complex, and reversing the arguments generally conjugates its value. This difference affects calculations even though the definition of orthogonality remains simple: the inner product must equal zero.

Throughout this tutorial, complex inner products are linear in the first argument and conjugate-linear in the second. Keeping that convention visible is essential. Some sources use the opposite convention, so a calculation involving a nonzero inner product or a complex scalar can look different there. The zero condition defining orthogonality, however, is unchanged.

Linearity and conjugate-linearity mean that, for vectors \(\mathbf{u},\mathbf{v},\mathbf{w}\in V\) and scalars \(a,b\in\mathbb{C}\), the rules are

In particular, a complex scalar in the first argument comes out unchanged, while a complex scalar in the second argument comes out conjugated. For real scalars these rules agree, since a real scalar is its own conjugate. This is why calculations in real inner product spaces do not reveal the distinction.

For \(\mathbb{C}^n\), the standard inner product with this convention is \(\langle\mathbf{u},\mathbf{v}\rangle=\sum_{j=1}^{n}u_j\overline{v_j}\). Equivalently, \(\langle\mathbf{u},\mathbf{v}\rangle=\mathbf{v}^{*}\mathbf{u}\), where \(\mathbf{v}^{*}\) is the conjugate transpose. The conjugation is on the entries of the second vector in the sum. That detail determines all the calculations that follow.

Testing Orthogonality by Computing the Inner Product

Complex coordinates can cancel through their phases. Orthogonality does not require each coordinate contribution to the inner product to vanish separately; it requires their sum to vanish. Nor does orthogonality mean that the vectors have disjoint nonzero coordinates. The first example had nonzero entries in both vectors in both coordinates, yet the two terms in the inner product cancelled.

Proof. By definition, \(\mathbf{u}\perp\mathbf{v}\) means \(\langle\mathbf{u},\mathbf{v}\rangle=0\). Conjugate symmetry gives \(\langle\mathbf{v},\mathbf{u}\rangle=\overline{\langle\mathbf{u},\mathbf{v}\rangle}\). If the first inner product is zero, its conjugate is zero, so \(\mathbf{v}\perp\mathbf{u}\). Conversely, if \(\langle\mathbf{v},\mathbf{u}\rangle=0\), conjugate symmetry gives \(\langle\mathbf{u},\mathbf{v}\rangle=\overline{0}=0\), so \(\mathbf{u}\perp\mathbf{v}\). \(\square\)

This symmetry concerns only whether the inner product is zero; it does not say that the two inner products are always equal. A nonzero complex value and its conjugate can differ.

What Happens When a Vector Is Scaled?

Let \(\alpha,\beta\in\mathbb{C}\). Applying linearity in the first argument and conjugate-linearity in the second gives the useful combined rule

The conjugate on \(\beta\) is required because \(\beta\) multiplies the vector in the second argument. If \(\mathbf{u}\) and \(\mathbf{v}\) are orthogonal, this formula shows that their scalar multiples are also orthogonal, including when one of the scalars is zero. When both scalars are nonzero, the converse holds as well: \(\alpha\mathbf{u}\perp\beta\mathbf{v}\) implies \(\mathbf{u}\perp\mathbf{v}\), because \(\alpha\overline{\beta}\ne0\).

Proof. By linearity in the first argument, \(\langle\alpha\mathbf{u},\beta\mathbf{v}\rangle=\alpha\langle\mathbf{u},\beta\mathbf{v}\rangle\). Conjugate-linearity in the second argument gives \(\langle\mathbf{u},\beta\mathbf{v}\rangle=\overline{\beta}\langle\mathbf{u},\mathbf{v}\rangle\). Combining these equalities proves the formula. If \(\langle\mathbf{u},\mathbf{v}\rangle=0\), the formula gives \(\langle\alpha\mathbf{u},\beta\mathbf{v}\rangle=\alpha\overline{\beta}\cdot0=0\). If \(\alpha\) and \(\beta\) are nonzero, then \(\alpha\overline{\beta}\ne0\); hence \(\alpha\overline{\beta}\langle\mathbf{u},\mathbf{v}\rangle=0\) holds exactly when \(\langle\mathbf{u},\mathbf{v}\rangle=0\). \(\square\)

Common Mistake: Using the Same Scalar Rule in Both Arguments

A frequent error is to treat the inner product as linear in each argument. Under the convention used here, that gives the wrong result whenever a non-real scalar is put in the second argument and the original inner product is nonzero. With \(\mathbf{u}\) and \(\mathbf{w}\) from the earlier example, \(\langle\mathbf{u},\mathbf{w}\rangle=3+i\). A mistaken calculation might write

The correction is to conjugate the scalar in the second argument. Since \(\overline{i}=-i\), the scalar rule gives

Direct computation confirms the correction. We have \(i\mathbf{w}=(-1,1+i)\), whose conjugate is \((-1,1-i)\). Therefore

The incorrect answer \(-1+3i\) is not a small rounding difference: it results from using the first-argument rule in the second argument. As a second check on the distinction, placing \(i\) in the first argument instead gives \(\langle i\mathbf{u},\mathbf{w}\rangle=i(3+i)=-1+3i\), not \(1-3i\). The location of the scalar matters.

Keep the Zero Test and the Convention Separate

To decide orthogonality in a complex inner product space, compute the inner product using the stated convention and test whether the result is exactly zero. Do not infer orthogonality from the appearance of the coordinates, and do not discard a nonzero imaginary part. For example, \(3+i\) is not zero, so the vectors in the second worked example are not orthogonal.

For the standard inner product on \(\mathbb{C}^n\), a reliable sequence is to write the coordinate formula, conjugate the second vector, multiply corresponding entries, and add. When scaling vectors, use \(\alpha\overline{\beta}\langle\mathbf{u},\mathbf{v}\rangle\), not \(\alpha\beta\langle\mathbf{u},\mathbf{v}\rangle\). These steps keep the definition, the complex arithmetic, and the chosen convention aligned.