# Math Help - Tensor product constructs

1. ## Tensor product constructs

Hi all.
I have already done linear interpolation in four variables and it works but the accuracy is not good, that's why I need to do non-linear interpolation for the same four variables.

Now, I have since learnt that one of the "easiest" ways to do it is by use of tensor product constructs.
Now, my question is, given, say, 4 univariate cubic polynomials (each of the 4 is interpolating its respective variable ), how does one multiply all four together ? (That is the definition of a tensor product construct in this context).
Thus, given the cubic polynomials,
var1 = aw^3 + bw^2 + bw + d
var2 = ex^3 + fx^2 + gx + h
var3 = iy^3 + jy^2 + ky + l
var4 = mz^3 + nz^2 + oz + p

F(w,x,y,z) = tensor product construct. How does one do that ?
Now please, no references to wikipedia articles -- or any other articles for that matter. I have read and re-read lots of articles online and at this point, they wont help --- what I need now is for someone who knows how to do a tensor product construct of the 4 univariate cases above to produce a multivariate case in 4 dimensions.
Thanks.

Frank.

2. Hi,

I guess you don't know what a tensor product is.

If you want tensor products you first need vector spaces.

The idea is that you create a new space, called the tensor space. This space fullfilled all vector space axioms (abelian group for the inner composition, outer operation for the scalar multiplication and the structure like a monoid (but now real monoid) and a structure which looks like distributive laws but like the monoid no real distributive laws. People often says that a vector space has distributive laws but this is not true.) So a tensor is only an element of a tensor space. If you have polynomials, you have to interpretate them as vectors in the spaces of polynomials.

If you take $n$ vectorspaces in the cartesian product (thats the reason why one say tensor to the $n$-th) and map them into a new space with the map $T$ (like tensor).

1.) $T$ has to be multi linear, which means linear in each argument. Think about the determinant, which is an alternating multilinear form. Its the same thing

2.) $T$ is universal. This means, that you have one unity map to an aribratary space such you can build $T$ as the composition of this.