11
$\begingroup$

It is possible that on a sphere $S^n$ there is a natural Riemannian metric in $R^(n+1)$. But it is not always possible for pseudo Riemann metric since the sum of two symmetric matrix which are not positive definite but may have rank different from the two matrix. So I wonder what is the sufficient and necessary condition for the dimension of a sphere which can be endowed with a Lorentz metric.

$\endgroup$
8
  • 3
    $\begingroup$ Manifolds have Lorentz metrics if and only if they have a section of the corresponding projective unit tangent bundle. If you want details, please post to the math.stackexchange website listed in the FAQ as this is more of an upper-level undergraduate mathematics issue. $\endgroup$ Commented Nov 4, 2010 at 18:52
  • $\begingroup$ If you don't want to post your question on math.stackexchange, take a look in the nearest manifolds textbook at the proof that abstract manifolds admit Riemann metrics (i.e. avoid proofs that use embeddings in euclidean space). This is a paracompactness/bump function/partition of unity argument. Now consider how you would perturb that proof for Lorentz metrics. $\endgroup$ Commented Nov 4, 2010 at 21:16
  • $\begingroup$ @Ryan: should one understand that with a partition of unity argument you can show the existence of Lorenz metrics on every (paracompact) manifold? I have serious doubts about that... $\endgroup$ Commented Nov 26, 2010 at 18:00
  • $\begingroup$ @Andrei: it appears you did not read what I said in my first comment "Manifolds have Lorentz metrics if and only if they have a section of the corresponding projective unit tangent bundle". $\endgroup$ Commented Nov 26, 2010 at 18:43
  • $\begingroup$ Andrei, I think Ryan is suggesting that when you try to adapt the proof for a Riemannian metric to a Lorentzian one, you run into anecessary condition for the Lorentzian metric to exist. Then you can check that the necessary condition is also sufficient. $\endgroup$
    – Deane Yang
    Commented Nov 26, 2010 at 18:44

2 Answers 2

31
$\begingroup$

A compact simply connected manifold carriez a Lorenz metric iff its Euler characteristic vanishes.

Proof: If $\chi(M)=0$, $M$ carries a nowhere vanishing vector field $X$. Pick up a Riemannian metric $g$ on $M$ (using a partition of unity argument) and denote by $\eta$ the 1-form dual to $X$: $\eta(Y):=g(X,Y)$ for all $Y\in TM$. Then
$$g-2\frac{\eta\otimes\eta}{g(X,X)}$$ is a Lorenz metric.

Conversely, if $M$ has a Lorenz metric $h$ of signature $(n-1,1)$, pick again a Riemannian metric $g$ and consider the symmetric endomorphism $A$ of $TM$ defined by $h(.,.)=g(A.,.)$. The eigenspaces of $A$ corresponding to the unique negative eigenvalue define a line sub-bundle of $TM$ which is trivial if $M$ is simply connected, so $\chi(M)=0$.

Therefore, the answer to your question is: $S^n$ carries a Lorenz metric iff $n$ is odd.

$\endgroup$
1
  • $\begingroup$ Does this theorem have a name? Thanks! $\endgroup$
    – 346699
    Commented Apr 22, 2014 at 12:59
21
$\begingroup$

I would just like to point out that a slightly more general version of the statement in Moroianu's answer holds:

A compact orientable manifold $M$ admits a Lorentzian metric if and only if $\chi(M)=0$.

Clearly, simply-connected manifolds are orientable, hence the above statement is more general. But, e.g., tori clearly admit product Lorentzian metrics.

For the proof of the above statement, recall $M$ admits a semi-Riemannian metric of index $k$ if and only if the tangent bundle $TM$ admits a sub-bundle, or distribution, of rank $k$. Thus, $M$ admits a Lorentzian metric if and only if it admits a line bundle (which is also equivalent to $M$ admitting a vector field that never vanishes). The obstruction to the existence of a line bundle on $M$ is given by the Euler class of $M$ (see e.g. Davis and Kirk "Lecture notes in Algebraic Topology"). By the Gauss-Bonnet-Chern Theorem, the Euler characteristic is the Euler class evaluated on the fundamental class of $M$, so that, for compact orientable manifolds, the obstruction is precisely $\chi(M)$.

$\endgroup$
0

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .