The following is from the book "Tensor Methods in Statistics", which could be downloaded there


I have a question regarding section 0.3.1, titled "Duality and dual spaces", there it is said

Let $V$ be a vector space with basis $\{ e_1, \ldots, e_n \}$. In the usual expression for $v \in V$ as a linear combination of the basis vectors, we write $v = v^i e_i$. The notation exhibits a certain formal symmetry between the components $v^i$ and the basis vectors $e_i$, but the interpretation, as usually given, is quite asymmetric: $v^i$ is a 'scalar', whereas $e_i$ is a 'vector' in $V$. However $\{v^i, \ldots, v^n\}$, as linear functionals, form a basis in $V^*$.

Here comes my first question, how could the scalar $v^i$ be regarded as a linear functional, and furthermore, how could $\{ v^i, \ldots, v^n\}$ be interpreted as a basis for $V^*$ when its just a set of scalars?

He continues

Thus, the expression $v^i e_i$ could equally well be interpreted as a point in $V^*$ with components $(e_1, \ldots, e_n)$ relative to the dual basis.

Now he interprets the basis vectors $e_i$ as components in the dual basis, but I thought linear functionals, i.e. element of $V^*$ could be represented equally by $n$-tuples of numbers.

  • $\begingroup$ Dear Stefan, This discussion is somewhat related. Regards, $\endgroup$ – Matt E Aug 7 '12 at 11:31

When the author says that $v^i$ can be thought of as a functional, they are not talking about the scalar $v^i$ attached to any particlar vector $v$. Rather, they mean the scalar valued function $v \mapsto v^i$ (sometimes called "projection onto the $i$th component").

This is a scalar-valued function on the set of vectors $\{v \, | \, v = (v^1,\ldots, v^n)\},$ i.e. it is a functional.

Although it is a little confusing to denote this functional by $v^i$, this is similar to the kind of confusion that can arise in calculus, where we might use $x$ to denote both a particular number and a variable, and hopefully won't be hard to get used to.

The resulting basis $v^i$ of $V^*$ is called the dual basis, since the value of $v^i$ on $e_j$ is $\delta_{ij}$. (This is just a fancy way of saying that the $i$th component of $e_j$ is $\delta_{ij}$.)

The expresssion $\sum_i v^i e_i $ then has a symmetry: we can interpret it as originally intended: the $e_i$ are basis vectors for $V$, and the $v^i$ are scalars giving the coordinates of a vector in terms of this basis --- or we can think of the $v^i$ as denoting the dual basis of $V^*$, and the $e_i$ as being coordinates of a vector in $V^*$ in terms of this basis.

How are the $e_i$ coordinates on a vector of $v^*$? Well, we just have to reverse engineer the above process of interpreting the scalar valued function $v^i$ as a vector, namely: given a functional $v^*$, we can evaluate it on the vector $e_i$ to get a scalar, and so we can think of this evaluation process as being a functional on $V^*$. In this way the vector $e_i \in V$ becomes thought of as a functional on $V^*$. (This identification of vectors in $V$ with functionals in $V^*$, via evaluation, is often called double duality.)

It's easy to check that in this way, $e_i$ just becomes the functional on $V^*$ which gives the $i$th coordinate of a functional $v^*$ in terms of the basis $(v^i)$. This explains the author's new interpretation of the expression $\sum_i v^i e_i$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.