A geometric question
Every real inner product defines a norm by . Expanding two squares shows that this norm obeys the parallelogram identity. Does every norm arise this way? No. We work over ; complex polarization requires a different formula.
Definition 9.1.1
(Parallelogram identity)
A norm on a real vector space satisfies the parallelogram identity if for every pair ,
Theorem 9.1.2
(Inner-product norms are exactly those satisfying the parallelogram identity)
A norm on a real vector space comes from a real inner product if and only if it satisfies the parallelogram identity for every pair of vectors. In that case the inner product is unique and is given by the polarization formula
Proof
(Polarization, additivity, then continuity)
If the norm comes from an inner product, expanding the squares gives the identity and also forces the formula for , so any inducing inner product equals . Conversely, suppose the norm satisfies the identity and define by the formula. It is symmetric because , and , so for . The nontrivial task is to prove linearity rather than assume that a difference of squared norms is bilinear.
Apply the parallelogram identity to the pairs and , then subtract. After dividing by 4, the result is
Putting gives , since . Given , take and . The displayed identity becomes
Thus is additive in the first argument. Additivity implies rational homogeneity: for rational .
For fixed , the function is continuous: the triangle inequality implies , and sums, differences and squares preserve continuity. Given a real , choose rationals and pass to the limit in . Hence . Symmetry gives linearity in the second argument too. Therefore is a real inner product and induces the original norm because .
Proof X-ray
Which step uses the identity, and which uses the norm
The proof has three layers, and each draws on a different property. The parallelogram identity is spent exactly once, in the additivity step: applying it to and and subtracting turns four squared norms into the relation . The substitutions and , are the non-obvious moves: the first gives doubling, the second converts the relation into plain additivity. Additivity only yields homogeneity for rational scalars. The last layer uses a different hypothesis, the triangle inequality for the norm, which makes continuous so that rational approximations reach every real scalar. Without continuity, additivity alone does not force for all real . The argument never uses finite dimension.
Worked examples
Worked example 9.1.3
(Polarizing a weighted norm)
On , define . This is a norm: its square is , so it is the Euclidean norm after an invertible linear change of coordinates. Its inducing inner product is
For and , the squared lengths of and are 7 and 3. Polarization gives , matching the cross term in the displayed inner product.
Worked example 9.1.4
(Checking the identity on one pair)
Let , which is positive for . For and : has squared norm , and has squared norm , so the left side is . Also and , so the right side is . Polarization gives , so .
Worked example 9.1.5
(The one-dimensional case)
On with , polarization gives , the usual product. Here the parallelogram identity reads , which holds for all real .
Counterexample mode
A norm can fail the test
A tempting claim is that every norm satisfies the parallelogram identity. For , take as above: , so the left side of the identity is , while the right side is . Therefore this norm is not induced by any real inner product. It remains a valid norm; only the extra inner-product structure fails.
Common mistake
(Checking the identity for a single pair proves nothing)
Every norm satisfies the identity for : . In particular the max norm passes this test, yet it fails at . The identity must hold for every pair, and a single failing pair rules out every inducing inner product.
Exercises
Quick check
(For the weighted example, compute and .)
Substitute into .
Solution · Answer
and .
Quick check
(In any norm, what is under the displayed polarization formula?)
Note that .
Solution · Answer
.
Quick check
(Why does failure of the parallelogram identity at one pair rule out every inducing inner product?)
What does every inner-product norm satisfy?
Solution · Answer
Every inner-product norm satisfies the identity for every pair of vectors, so one failing pair is incompatible with the norm coming from an inner product.
Quick check
(Does additivity by itself justify real homogeneity in the proof?)
Which scalars does additivity reach?
Solution · Answer
No. Additivity gives rational homogeneity; continuity of , which comes from the triangle inequality, extends it to all real scalars.