1: \begin{abstract}  Let $\zz[1/p]$ denote the ring of integers with the prime $p$ inverted.  There  is a canonical homomorphism 

2: $\Psi\co \hskip-.07in \oplus  \Theta^3_{\zz[1/p]}  \to \Theta^3_{\qq}$, where   $\Theta^3_{R}$ denotes the  three-dimensional smooth $R$--homology cobordism group of $R$--homology spheres and the direct sum is over all prime integers.   Gauge theoretic  methods prove the kernel is infinitely generated.  Here we prove that $\Psi$ is not surjective, with cokernel infinitely generated.  

3: As a basic example we show that for $p$ and $q$ distinct primes,  there is no rational homology cobordism from the lens space  $L(pq,1) $ to any  $M_p \cs M_q$, where $H_1(M_p) = \zz_{p}$ and $H_1(M_q) = \zz_{q}$.  More subtle examples  include cases in which a cobordism to  such a connected sum exists topologically but not smoothly.  (Conjecturally, such a splitting always exists topologically.)  Further examples can be chosen to represent 2--torsion in $\Theta^3_{\qq}$.   

4: 

5: Let $\calk$ denote the kernel  of $\Theta^3_\qq \to \widehat{\Theta}^{3}_\qq$, where  $\widehat{\Theta}^{3}_\qq$ denotes the topological homology cobordism group.  Freedman proved that $\Theta^3_\zz \subset \calk$.  A corollary of results here is that $\calk / \Theta^3_\zz $ is infinitely generated.  We also  demonstrate the failure in dimension three of splitting theorems that apply to higher dimensional knot concordance groups. \end{abstract}

6: