Complex Conjugation In Inner Products

Why Complex Conjugation Appears

Over the real numbers, multiplying a scalar by itself gives a nonnegative result: \(a^2\ge 0\) for every real \(a\). Complex numbers do not have this property. For example, \(i^2=-1\), so a direct analogue of squaring a complex coordinate can produce a negative value. An inner product must still give a real, nonnegative value when a vector is paired with itself. Complex conjugation supplies the needed adjustment: if \(z=a+bi\), then \(z\overline{z}=a^2+b^2\), which is real and nonnegative.

Throughout this tutorial, the scalar field is \(\mathbb{C}\), and we use the course convention that an inner product is linear in its first argument and conjugate-linear in its second. Thus scaling the first argument by \(c\) scales the result by \(c\), while scaling the second argument by \(c\) scales the result by \(\overline{c}\). These rules explain how to calculate with complex inner products; they do not mean that conjugation is applied to an entire vector or pairing indiscriminately.

The conjugate changes the sign of the imaginary part, not the real part. It obeys \(\overline{z+w}=\overline{z}+\overline{w}\) and \(\overline{zw}=\overline{z}\,\overline{w}\). In particular, \(z\overline{z}\) is real for every \(z\), and it is zero exactly when \(z=0\): the sum \(a^2+b^2\) is zero exactly when both real numbers \(a\) and \(b\) are zero.

Conjugation in Inner-Product Calculations

A complex inner product on a complex vector space \(V\) is a pairing \(\langle\mathbf{u},\mathbf{v}\rangle\in\mathbb{C}\) with, among its defining properties, linearity in the first argument, conjugate-linearity in the second, conjugate symmetry, and positive definiteness. The calculation rules used here are \(\langle c\mathbf{u},\mathbf{v}\rangle=c\langle\mathbf{u},\mathbf{v}\rangle\) and \(\langle\mathbf{u},c\mathbf{v}\rangle=\overline{c}\langle\mathbf{u},\mathbf{v}\rangle\). Conjugate symmetry means \(\langle\mathbf{u},\mathbf{v}\rangle=\overline{\langle\mathbf{v},\mathbf{u}\rangle}\). Positive definiteness means \(\langle\mathbf{u},\mathbf{u}\rangle\) is real and nonnegative, and it is zero exactly when \(\mathbf{u}=\mathbf{0}\).

These properties distinguish a complex inner product from a pairing that is simply linear in both arguments. Conjugate symmetry makes each self-product equal to its own conjugate, so it is real. Conjugate-linearity makes the contribution from a complex scalar in the second slot its conjugate, which is what turns a scalar self-product into a squared modulus.

Proof. By conjugate symmetry, \(\langle\mathbf{x},\mathbf{x}\rangle=\overline{\langle\mathbf{x},\mathbf{x}\rangle}\). A complex number equal to its conjugate has zero imaginary part, so the self-product is real. For the scaling identity, first use linearity in the first argument and then conjugate-linearity in the second: \(\langle c\mathbf{x},c\mathbf{x}\rangle=c\langle\mathbf{x},c\mathbf{x}\rangle=c\overline{c}\langle\mathbf{x},\mathbf{x}\rangle=|c|^2\langle\mathbf{x},\mathbf{x}\rangle\). This calculation also applies when \(c=0\), since both sides are zero. \(\square\)

In one complex coordinate, the pairing \(\langle z,w\rangle=z\overline{w}\) illustrates these rules directly. Its first slot is linear, and its second slot is conjugate-linear. In particular, the self-product of a scalar \(z\) is \(z\overline{z}=|z|^2\), not \(z^2\).

Why Bilinearity in Both Arguments Fails

For a real vector space, an inner product is linear in both arguments. It is tempting to carry that rule unchanged to a complex vector space. The scalar \(i\) shows why that cannot preserve positive definiteness: if both slots were linear, putting \(i\mathbf{x}\) in both slots would multiply a self-product by \(i^2=-1\). A nonzero vector would then have a negative self-product, contrary to positive definiteness.

Proof. Suppose \(V\) contains a nonzero vector \(\mathbf{x}\), and suppose a pairing on \(V\) is bilinear in both arguments and positive definite. Bilinearity in the first and second arguments gives \(\langle i\mathbf{x},i\mathbf{x}\rangle=i\,i\,\langle\mathbf{x},\mathbf{x}\rangle=-\langle\mathbf{x},\mathbf{x}\rangle\). Because \(\mathbf{x}\ne\mathbf{0}\), positive definiteness gives \(\langle\mathbf{x},\mathbf{x}\rangle>0\). Thus the displayed self-product is a negative real number. But \(i\mathbf{x}\ne\mathbf{0}\), since multiplying by the nonzero scalar \(i\) cannot turn a nonzero vector into zero; positive definiteness therefore requires \(\langle i\mathbf{x},i\mathbf{x}\rangle>0\). The same number cannot be both negative and positive, a contradiction. The nonzero-space hypothesis matters: on the zero vector space, the zero pairing is bilinear and positive definite vacuously, because there is no nonzero vector to test. \(\square\)

Scaling and Squared Lengths

Positive definiteness gives a squared length through the self-product: the value \(\langle\mathbf{x},\mathbf{x}\rangle\) is real and nonnegative, and it vanishes only for \(\mathbf{x}=\mathbf{0}\). The scaling identity shows how that value changes when the vector is multiplied by a complex scalar. Because \(|c|^2\ge0\), multiplying a vector by \(c\) cannot reverse the sign of its squared length. If \(c\ne0\) and \(\mathbf{x}\ne\mathbf{0}\), then \(|c|^2>0\), so the scaled vector still has positive squared length.

Common Mistake: Squaring Instead of Conjugating

A common error is to calculate a complex self-product as though the same scalar appeared linearly in both slots. For \(z=1+2i\), that incorrect calculation is \(z^2=(1+2i)^2=1+4i+4i^2=-3+4i\). This is not even real, so it cannot be a positive squared length. The correction is to conjugate the scalar in the second slot: \((1+2i)\overline{(1+2i)}=(1+2i)(1-2i)=1-2i+2i-4i^2=5\). The corrected value is real and positive, and it equals \(1^2+2^2\).

The same issue appears in the smallest possible test: omitting conjugation gives \(i^2=-1\), while the corrected self-product is \(i\overline{i}=i(-i)=1\). When checking a complex inner-product calculation, verify both which argument is being scaled and whether its scalar must be conjugated. In particular, do not replace conjugation by taking a real part or by changing the sign of the whole expression.