$\renewcommand\vec[1]{{\boldsymbol #1}}$If $V$ is a vector space and $T,S \in \mathfrak{T}^0_{\;\;2}(V)$ are covariant tensors of type $(0,2)$, then we define their Kulkarni-Nomizu$\newcommand\KN{\bigcirc \kern-2.5ex\wedge \;}$ product $T \KN S \in \mathfrak{T}^0_{\;\;4}(V)$ by $$(T \KN S)(\vec{x},\vec{y},\vec{z},\vec{w}) \doteq T(\vec{x},\vec{z})S(\vec{y},\vec{w}) + T(\vec{y},\vec{w})S(\vec{x},\vec{z})-T(\vec{x},\vec{w})S(\vec{y},\vec{z}) - T(\vec{y},\vec{z})S(\vec{x},\vec{w}).$$

I understand that $T \KN S$ is a curvature like tensor, in the sense that it satisfies the following symmetries:

  1. $(T \KN S)(\vec{x},\vec{y},\vec{z},\vec{w}) = -(T\KN S)(\vec{y},\vec{x},\vec{z},\vec{w}) = -(T \KN S)(\vec{x},\vec{y},\vec{w},\vec{z})$;
  2. $(T \KN S)(\vec{x},\vec{y},\vec{z},\vec{w}) = (T \KN S)(\vec{z},\vec{w},\vec{x},\vec{y})$;
  3. $(T \KN S)(\vec{x},\vec{y},\vec{z},\cdot)+(T \KN S)(\vec{y},\vec{z},\vec{x},\cdot)+(T \KN S)(\vec{z},\vec{x},\vec{y},\cdot)=0$, if $T$ and $S$ are symmetric.

We also have the bonus property that $T \KN S = S \KN T$.

And although this quantity appearing frequently in certain calculations is enough justificative for giving it a name and notation, this seems artificial to me so far.

I do not have any intuition whatsoever for that formula, nor can I think of a way to write it neatly as $\sum_{\sigma \in S_4}{\rm something}$. We can write $$(T\KN S)(\vec{x},\vec{y},\vec{z},\vec{w}) = \begin{vmatrix} T(\vec{x},\vec{z}) & S(\vec{y},\vec{z}) \\ T(\vec{x},\vec{w}) & S(\vec{y},\vec{w})\end{vmatrix}-\begin{vmatrix} T(\vec{y},\vec{z}) & S(\vec{x},\vec{z}) \\ T(\vec{y},\vec{w}) & S(\vec{x},\vec{w})\end{vmatrix},$$but I'm failing to interpret this either.

I'd like some insight on this definition.

If I didn't mess up, we have the more "symmetric" expression

$$(T\KN S)(\vec{x},\vec{y},\vec{z},\vec{w}) = \begin{vmatrix} T(\vec{x},\vec{z}) & S(\vec{x}+\vec{y},\vec{z}) \\ T(\vec{x},\vec{w}) & S(\vec{x}+\vec{y},\vec{w})\end{vmatrix}+\begin{vmatrix} S(\vec{x},\vec{z}) & T(\vec{x}+\vec{y},\vec{z}) \\ S(\vec{x},\vec{w}) & T(\vec{x}+\vec{y},\vec{w})\end{vmatrix}.$$Still not happy.


$\newcommand\KN{\bigcirc\kern-2.5ex\wedge \;}$Your symmetries $1-2$ are equivalent to requiring that $T \KN S \in S^2(\Lambda^2 V)$; so what we're really trying to do is to take bilinear forms on $V$ (in practice these are usually built from the metric and the Ricci tensor) and turn them into bilinear forms on $\Lambda^2 V$.

Any symmetric bilinear form $T \in S^2 V$ naturally induces a symmetric bilinear form on two-forms $\Lambda^2 T \in S^2 (\Lambda^2 V)$ via the formula $$(\Lambda^2 T)(v \wedge w, x \wedge y) = T(v,x)T(w,y) - T(v,y) T(x,w).$$ If you're unconvinced about the naturality of this construction, note that it corresponds via raising an index to the endomorphism $\Lambda^2 T^\sharp : \Lambda^2 V \to \Lambda^2 V$ induced on two-forms by an endomorphism $T^\sharp : V \to V$, which may be more familiar.

Thinking of $\Lambda^2$ as a function $S^2V \to S^2(\Lambda^2 V)$, we see that it is in fact a (vector-valued) quadratic form. By polarizing this quadratic form (a natural thing to do!) we get a symmetric bilinear form on $S^2(V)$. This bilinear form is (up to a factor of $\frac12$) the Kulkarni-Nomizu product. That is, the K-N product is the unique symmetric bilinear mapping $S^2(V) \times S^2(V) \to S^2(\Lambda^2 V)$ such that $T \KN T = 2\Lambda^2 T$. (I think the factor of $2$ is an unfortunate historical accident, but it does have the convenient property of making the curvature tensor of the sphere equal to $g \KN g$.)

That's the best justification (other than the fact it pops up so often) I have for its existence - hopefully that makes it seem a bit less artificial. For real intuition I think you just have to work with it for a while - I found its role in the irreducible decomposition of curvature tensors made it click for me. I learnt this stuff from chapter 11 of this freely available book.

  • $\begingroup$ This is beautiful. Exactly the sort of thing I was looking for, thanks a lot $\endgroup$ – Ivo Terek Sep 5 '17 at 17:19

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.