Complex Matrix And Function Inner Products

Why Conjugation Appears

For real vectors, multiplying corresponding entries and adding gives a nonnegative result when a vector is paired with itself. Over the complex numbers, that same rule fails: \(i\cdot i=-1\). To obtain a nonnegative self-pairing, one factor must be complex-conjugated, since \(z\overline{z}=|z|^2\). This principle gives natural inner products for complex matrices and for complex-valued functions.

We use the convention that an inner product is linear in its first argument and conjugate-linear in its second. In particular, if \(\alpha,\beta\in\mathbb{C}\), then \(\langle \alpha u+\beta v,w\rangle=\alpha\langle u,w\rangle+\beta\langle v,w\rangle\), while \(\langle u,\alpha v+\beta w\rangle=\overline{\alpha}\langle u,v\rangle+\overline{\beta}\langle u,w\rangle\). The conjugate-linearity in the second argument follows from the first-argument linearity and conjugate symmetry.

The Frobenius Inner Product for Complex Matrices

For \(\mathbf{A},\mathbf{B}\in\mathbb{C}^{m\times n}\), define their Frobenius inner product by multiplying corresponding entries of \(\mathbf{A}\) by the conjugates of the corresponding entries of \(\mathbf{B}\), then summing. This is the complex version of the entrywise matrix inner product.

Theorem. For positive integers \(m,n\), this formula is an inner product on the complex vector space \(\mathbb{C}^{m\times n}\), which is finite-dimensional.

Proof. Let \(\mathbf{A},\mathbf{B},\mathbf{C}\in\mathbb{C}^{m\times n}\) and \(\alpha,\beta\in\mathbb{C}\). In each entry, \((\alpha a_{ij}+\beta c_{ij})\overline{b_{ij}}=\alpha a_{ij}\overline{b_{ij}}+\beta c_{ij}\overline{b_{ij}}\). Summing over \(i,j\) proves linearity in the first argument. Also, \(\overline{\langle\mathbf{B},\mathbf{A}\rangle_F}=\overline{\sum_{i,j}b_{ij}\overline{a_{ij}}}=\sum_{i,j}\overline{b_{ij}}a_{ij}=\langle\mathbf{A},\mathbf{B}\rangle_F\), proving conjugate symmetry. Finally, \(\langle\mathbf{A},\mathbf{A}\rangle_F=\sum_{i,j}a_{ij}\overline{a_{ij}}=\sum_{i,j}|a_{ij}|^2\), a real nonnegative sum. It is zero exactly when each \(|a_{ij}|^2=0\), which is exactly when every entry \(a_{ij}=0\), or \(\mathbf{A}=\mathbf{O}\). Thus the formula is positive definite. The matrices with a \(1\) in one position and \(0\) elsewhere form a basis, since every matrix is the sum of its entries times these matrices; hence the space is finite-dimensional. \(\square\)

The same inner product can be written using the conjugate transpose \(\mathbf{B}^{*}\) and the trace, which is the sum of the diagonal entries of a square matrix. Indeed, the \(k\)-th diagonal entry of \(\mathbf{A}\mathbf{B}^{*}\) is \(\sum_{j=1}^{n}a_{kj}\overline{b_{kj}}\). Summing these diagonal entries gives \(\langle\mathbf{A},\mathbf{B}\rangle_F=\operatorname{tr}(\mathbf{A}\mathbf{B}^{*})\). The product is \(m\times m\), so its trace is defined even when \(\mathbf{A}\) and \(\mathbf{B}\) are not square.

Inner Products on Continuous Complex Functions

There is a parallel construction for functions. Fix real numbers \(a<b\), and let \(C([a,b],\mathbb{C})\) be the complex vector space of continuous functions from \([a,b]\) to \(\mathbb{C}\). For two such functions \(f,g\), the product \(f(x)\overline{g(x)}\) is continuous, so its integral on the closed bounded interval exists. Define the pairing by integrating that product.

Theorem. For real \(a<b\), this formula is an inner product on \(C([a,b],\mathbb{C})\).

Proof. Let \(f,g,h\in C([a,b],\mathbb{C})\) and \(\alpha,\beta\in\mathbb{C}\). Linearity of the integral gives \(\langle\alpha f+\beta h,g\rangle=\int_a^b(\alpha f(x)+\beta h(x))\overline{g(x)}\,dx=\alpha\langle f,g\rangle+\beta\langle h,g\rangle\). Conjugating the integral gives \(\overline{\langle g,f\rangle}=\overline{\int_a^b g(x)\overline{f(x)}\,dx}=\int_a^b\overline{g(x)}f(x)\,dx=\langle f,g\rangle\), so conjugate symmetry holds. For any \(f\), \(\langle f,f\rangle=\int_a^b|f(x)|^2\,dx\), which is real and nonnegative. If \(f\) is identically zero, this integral is zero. Conversely, suppose \(f\) is not identically zero. Then there is \(x_0\in[a,b]\) with \(|f(x_0)|>0\). By continuity, there is a \(\delta>0\) such that \(|f(x)-f(x_0)|<|f(x_0)|/2\) for \(x\in[a,b]\) with \(|x-x_0|<\delta\). The triangle inequality then gives \(|f(x)|>|f(x_0)|/2\) on that part of the interval. Because \(a<b\), this neighborhood contains a subinterval of positive length, including when \(x_0\) is an endpoint. The integral of \(|f(x)|^2\) over that subinterval is at least its length times \(|f(x_0)|^2/4\), so it is strictly positive. Therefore \(\langle f,f\rangle=0\) holds exactly for the zero function. \(\square\)

Why the Function Space Matters

Continuity is important to the positive-definiteness proof: a continuous function that is nonzero at one point must remain bounded away from zero on some interval of positive length. Without a suitable restriction on the functions, an integral can miss a nonzero value at an isolated point.

Common mistake: assuming the integral detects every nonzero function. On \([0,1]\), define \(q(0)=1\) and \(q(x)=0\) for every \(x>0\). This function is not the zero function, but it is nonzero at just one point, so \(\int_0^1|q(x)|^2\,dx=0\). Thus the proposed integral pairing is not positive definite on the space of all functions on \([0,1]\). The correction is to specify a function space for which the integral has the required property, such as \(C([0,1],\mathbb{C})\). In that space, the function \(q\) is excluded because it is not continuous.

Common mistake: omitting conjugation. For the one-entry matrix \(\mathbf{A}=[i]\), the incorrect rule that multiplies entries without conjugation gives \(i\cdot i=-1\) when pairing \(\mathbf{A}\) with itself. The Frobenius rule instead gives \(i\overline{i}=i(-i)=1\), a positive real number. The conjugate in the second factor is what makes the self-pairing a sum of squared moduli rather than a potentially negative or nonreal quantity.

Both constructions use the same principle: multiply one entry or function value by the conjugate of the other, then sum or integrate. Their positive-definiteness checks depend on the underlying objects: a matrix is zero exactly when all its entries are zero, while for continuous functions a zero integral of the squared modulus forces the function to vanish throughout the interval.