Frobenius Norms Of Matrices

Measuring the Size of a Matrix

A matrix has many entries, so comparing two matrices entry by entry can be awkward. The Frobenius norm packages the sizes of all the entries into one nonnegative number: square their absolute values, add the results, and take the square root. It is a useful way to quantify the overall size of a matrix, and its connection to the Frobenius inner product lets us understand how that size changes under addition.

For a real matrix, the absolute value of an entry is just its ordinary absolute value. For a complex matrix, the absolute value is the complex modulus. If \(\mathbf{A}\) and \(\mathbf{B}\) have the same dimensions, their sum is defined entry by entry. We can compare their Frobenius norms as numbers; a larger norm means a larger Frobenius size, not that every entry of one matrix is larger than the corresponding entry of the other.

This formula comes from the Frobenius inner product: with the convention that the inner product is linear in its first slot, \(\langle\mathbf{A},\mathbf{B}\rangle=\sum_{i=1}^{m}\sum_{j=1}^{n}a_{ij}\overline{b_{ij}}\). Thus \(\|\mathbf{A}\|_F=\sqrt{\langle\mathbf{A},\mathbf{A}\rangle}\). In particular, \(\|\mathbf{A}\|_F\geq0\), and \(\|\mathbf{A}\|_F=0\) exactly when every entry is zero, that is, when \(\mathbf{A}=\mathbf{O}\). The Frobenius norm is the norm induced by this inner product.

Calculating and Comparing Norms

The calculation treats every entry in the same way, regardless of its row or column. Therefore, rearranging the entries without changing any of their values does not change the sum of their squared absolute values. Transposition is one such rearrangement.

Proof. The entries of \(\mathbf{A}^{\top}\) are the entries of \(\mathbf{A}\) with their row and column positions exchanged. The double sum of their squared absolute values therefore has exactly the same terms: \(\|\mathbf{A}^{\top}\|_F^2=\sum_{i=1}^{n}\sum_{j=1}^{m}|a_{ji}|^2=\sum_{i=1}^{m}\sum_{j=1}^{n}|a_{ij}|^2=\|\mathbf{A}\|_F^2\). Both norms are nonnegative, so equality of their squares gives equality of the norms. For scalar multiplication, every entry of \(c\mathbf{A}\) is \(ca_{ij}\), so \(\|c\mathbf{A}\|_F^2=\sum_{i=1}^{m}\sum_{j=1}^{n}|ca_{ij}|^2=|c|^2\sum_{i=1}^{m}\sum_{j=1}^{n}|a_{ij}|^2=|c|^2\|\mathbf{A}\|_F^2\). Taking nonnegative square roots gives \(\|c\mathbf{A}\|_F=|c|\|\mathbf{A}\|_F\). This also covers \(c=0\), since both sides are then zero. \(\square\)

What Happens to the Norm of a Sum?

For a sum, the squared norm is not generally the sum of the squared norms. The Frobenius inner product measures how the entries of the two matrices align: positive contributions can increase the squared norm of the sum, and negative contributions can decrease it. The following identity makes this exact. It applies to matrices of the same dimensions, so that both their sum and their inner product are defined.

Proof. By the definition of the induced norm and the linearity of the inner product in its first slot, \(\|\mathbf{A}+\mathbf{B}\|_F^2=\langle\mathbf{A}+\mathbf{B},\mathbf{A}+\mathbf{B}\rangle=\langle\mathbf{A},\mathbf{A}\rangle+\langle\mathbf{A},\mathbf{B}\rangle+\langle\mathbf{B},\mathbf{A}\rangle+\langle\mathbf{B},\mathbf{B}\rangle\). The inner product satisfies \(\langle\mathbf{B},\mathbf{A}\rangle=\overline{\langle\mathbf{A},\mathbf{B}\rangle}\). For any complex number \(z\), \(z+\overline z=2\operatorname{Re}z\). The two diagonal terms are \(\|\mathbf{A}\|_F^2\) and \(\|\mathbf{B}\|_F^2\), so collecting the cross terms proves the identity. Over \(\mathbb{R}\), taking the real part leaves the inner product unchanged. \(\square\)

Common Pitfall: Adding Norms as Though They Were Squares

A tempting but incorrect calculation is to claim that the norm of a sum always equals the sum of the norms, or that the squared norm always equals the sum of the squared norms. In the example above, the first claim would give \(\|\mathbf{P}+\mathbf{Q}\|_F=3+\sqrt{6}\), even though the entries of the sum give the value \(3\). The second claim would give \(9+6=15\) for the squared norm, instead of the correct value \(9\). Both errors ignore the cross term \(2\langle\mathbf{P},\mathbf{Q}\rangle=-6\).

To correct the calculation, first check the dimensions, then either add the matrices entry by entry and compute the norm from the resulting entries, or use the squared-norm identity and include the inner-product term. The two methods must agree. For a scalar multiple, use the absolute value of the scalar: multiplying a matrix by a negative real number still multiplies its norm by a nonnegative factor. For a transpose, the positions change but the entries do not, so the sum of their squared absolute values stays the same.

These calculations give a consistent way to compare matrix sizes without confusing a matrix with any one of its entries. The entry formula is especially practical for exact examples, while the inner-product identity explains why adding matrices can either increase or decrease the resulting size.