The parallelogram law is an identity stating that the sum of the squares of the lengths of a parallelogram’s diagonals equals the sum of the squares of the lengths of its four sides. In a vector space equipped with a norm, its standard form is
It holds for every pair of vectors whenever the norm is induced by an inner product. Conversely, a real or complex norm satisfying this identity for every pair of vectors is induced by a unique inner product. This characterization connects elementary Euclidean geometry with the structure of abstract normed spaces.
Geometric meaning
Let and represent two adjacent sides of a parallelogram in Euclidean space. Its diagonals are represented by and , up to orientation. If the adjacent side lengths are and , and the diagonal lengths are and , the identity becomes
The result does not depend on the angle between the adjacent sides. As that angle varies while the side lengths remain fixed, one diagonal can lengthen while the other shortens, but the sum of their squared lengths remains constant.
One elementary derivation uses the law of cosines. If the angle between the adjacent sides is , then
Adding the equations eliminates the angle-dependent terms.
For a rectangle, the diagonals have equal length, so the identity reduces to
the Pythagorean theorem. The parallelogram law therefore extends a familiar relation for right triangles to arbitrary parallelograms.
Derivation from an inner product
An inner product induces the norm
In a real inner product space,
and
Their sum is the parallelogram identity.
For a complex inner product space, the corresponding expansions contain the real part of the inner product:
The mixed terms again cancel.
Thus the law holds in every inner product space, regardless of dimension or completeness. Examples include coordinate spaces with the usual dot product, spaces of square-integrable functions, and all Hilbert spaces.
Characterization of inner product norms
The converse is the Jordan–von Neumann theorem, established by Pascual Jordan and John von Neumann in 1935:
A real or complex normed vector space has a norm induced by an inner product if and only if its norm satisfies the parallelogram law for all pairs of vectors.
The quantifier “for all pairs” is essential. An equality that happens to hold for particular vectors does not characterize the norm of the whole space.
The theorem does not require completeness. If a Banach space—a complete normed vector space—satisfies the identity, the resulting inner product makes it a Hilbert space.
Recovering the inner product
The polarization identity reconstructs the inner product from the norm. Over the real numbers,
Over the complex numbers, using the convention that the inner product is linear in its first argument,
For the convention that the inner product is linear in its second argument, the signs of the imaginary terms are reversed.
These formulas show uniqueness: two inner products inducing exactly the same norm must agree. They do not, however, turn an arbitrary norm into an inner product. The parallelogram law supplies the algebraic compatibility needed for linearity.
Outline of the real converse
Suppose a real normed space satisfies the parallelogram law. Set
Symmetry and follow immediately. Applying the parallelogram identity to the terms in this definition gives
Taking yields
Substitution then gives additivity:
Additivity implies homogeneity for rational scalars. Continuity, inherited from the norm, extends this to all real scalars. Symmetry gives linearity in the other argument, and
gives positive definiteness. Hence is the required inner product.
Norms that fail the identity
The parallelogram law is stronger than the triangle inequality and is not an axiom of a general normed vector space.
On , consider the norm
For and ,
whereas
Thus this norm cannot be induced by an inner product.
For the maximum norm,
the same vectors give a left-hand side of and a right-hand side of . Failure can therefore occur in either direction.
More generally, for the usual finite-dimensional spaces of dimension at least two, with , these vectors give
Equality with occurs precisely when . The maximum-norm example handles .
The identity tests the particular norm, not whether the underlying vector space can support some other inner product. For example, with its norm fails the law, although the same vector space with its Euclidean norm satisfies it.
Consequences and applications
Midpoint identities and approximation
An equivalent form is
Replacing and by and gives
This equation measures exactly how taking a midpoint reduces average squared distance. In a Hilbert space, it is central to proving the existence and uniqueness of the nearest point to in a nonempty closed convex set. For a closed linear subspace, this nearest-point construction is orthogonal projection.
Strict convexity
If and are distinct unit vectors, the midpoint identity yields
Thus an inner product norm is strictly convex: the midpoint of two distinct unit vectors lies inside the unit ball.
Strict convexity alone does not imply the parallelogram law. For example, the usual norms for are strictly convex, but in dimension at least two they satisfy the parallelogram law only for .
Squared-error decompositions
The law extends to finite collections of vectors. If
then
For real-valued data, this is the decomposition underlying the least-squares property of the sample mean and familiar identities for variance. For vectors, it explains why the arithmetic mean minimizes the sum of squared inner-product distances.
Terminological distinction
“Parallelogram law” can also refer to the geometric rule for adding vectors: when two vectors form adjacent sides of a parallelogram, their sum is its diagonal from the common starting point. That construction is distinct from the squared-norm identity. Vector addition remains meaningful in spaces whose norms do not satisfy the parallelogram identity.