Tensor algebras

Motivation

Given two R-modules M and N, the R-module MβŠ—RN was created to satisfy the universal property that captures the idea of being able to multiply elements in M with elements in N. In the special case M=N, the module MβŠ—RM gives us a way to compute products of elements in M. It doesn't quite give us a ring structure on M, however, since it doesn't allow us to multiply together more than two elements in M. Can we amplify our tensor product construction to obtain a module with a full ring structure, capturing the desire of layering a multiplicative operation on the R-module M?

The desired universal property

With the above motivation in mind, we are looking for a functor from the category of R-modules to the category of R-algebras that is "free" in the same sense as the free R-module construction on a set S; i.e., that is left adjoint to the corresponding forgetful functor.

(Primary) Universal property of the tensor algebra

Let U:Rβˆ’Algβ†’Rβˆ’Mod be the forgetful functor from the category of R-algebras to the category of R-modules. Then there is functor T:Rβˆ’Modβ†’Rβˆ’Alg together with a natural bijection

HomRβˆ’Alg(T(M),A)β†’βˆΌHomRβˆ’Mod(M,U(A)).

In other words, the functor T is a left adjoint of the forgetful functor U.

Even without having seen the construction yet, such a property gives us a way to think about T(M):

The construction

Definition of tensor algebra

Suppose R is a commutative ring (with unity) and M is an R-module. Set T0(M)=R and for each positive integer k define the kth tensor power of M to be the R-module

Tk(M)=MβŠ—RMβŠ—Rβ‹―βŠ—RM⏟kΒ times.

The elements of Tk(M) are called k-tensors.

Then define the tensor algebra of M to be the R-module

T(M)=⨁k=0∞Tk(M).

Every element of T(M) is a finite formal linear combination of k-tensors.

As the name implies, the R-module T(M) has a (natural) structure of an R-algebra. The multiplication on simple tensors is defined by concatenation of tensors:

(m1βŠ—β‹―βŠ—mi)β‹…(m1β€²βŠ—β‹―βŠ—mjβ€²):=m1βŠ—β‹―βŠ—miβŠ—miβ€²βŠ—β‹―βŠ—mjβ€².

The multiplication on sums of tensors is defined via the distributive laws. Note that this multiplication satisfies Ti(M)Tj(M)βŠ†Ti+j(M). In other words, the tensor algebra T(M) has the structure of a graded ring.

It's worth noting that, as a ring, the tensor algebra T(M) is generated by the elements of T0(M)=R and T1(M)=M.

At some point we should verify that this construction satisfies the desired universal property. For now, however, let's look at some examples.

Examples

  1. Let R=Z and M=Q/Z. One can verify that Q/ZβŠ—ZQ/Z≃0, and so the tensor algebra is simply

    T(Q/Z)≃ZβŠ•(Q/Z).

    We can also unwind the multiplication rule. Let's look at a specific example. Two elements in ZβŠ•(Q/Z) are 2βŠ•[13] and 3βŠ•[45]. When we add these two elements we get the element

    (2+3)βŠ•[13+45]=5βŠ•[1715]=5βŠ•[215].

    Notice that this is not what we would get conventionally by adding the rational numbers (2+13)+(3+45)=6+215.

    When we multiply these two elements we get the element

    (2βŠ•[13])βŠ—(3βŠ•[45])=(2β‹…3)βŠ•(2β‹…[45]+3β‹…[13])βŠ•([13]βŠ—[45])=6βŠ•[135]βŠ•0=6βŠ•[35].
  2. Let R=Z and M=Zn. One can verify that ZnβŠ—ZZn≃Zn, and so the tensor algebra is

    T(Zn)≃ZβŠ•ZnβŠ•ZnβŠ•β‹―β‰ƒZ[x]/(nx).

    For that final isomorphism, the correspondence is that each finite formal sum c0βŠ•[c1]βŠ•[c2]βŠ•β‹― maps to the coset represented by the polynomial c0+c1x+c2x2+β‹―.

  3. Suppose R=F is a field and V is a finite-dimensional F-vector space. Let B={v1,…,vn} be a basis for V as an F-vector space. Then a basis for the F-vector space Tk(V) is

    {vi1βŠ—vi2βŠ—β‹―βŠ—vik∣vij∈B}.

    (When k=0, the basis is simply {1}.) In particular, Tk(V) is an F-vector space of dimension nk.