Mdtryhht

For a subring $R⊂ mathbb Q$, a knot $K⊂S^3$ is called $R$-slice if there exists an embedded disk $D$ in an $R$-homology $4$-ball $B$ such that $∂(B,D) = (S^3,K)$, see [Definition 1.3, KW16]. We say $K$ is rationally (resp. integrally) slice if $R= mathbb Q$ (resp. $= mathbb Z$).

In terms of crossing, the minimal example of rationally slice knot seems (probably is) figure-eight knot $4_1$, see [Theorem 4.16, Cha07].

A knot $K ⊂ S^3$ is slice if it bounds a smoothly embedded disk $D^2$ in the $4$-ball $B^4$. Again in terms of crossing, the minimal example of slice knot is unknot $0_1$.

My question is that is there any minimal example of integrally slice knot?

Both $6_1$ and $3_1 # m(3_1)$ are smoothly slice (as is the unknot), and I claim that all other knots of at most seven crossings are not integrally slice. This will follow from two claims: first, if $K$ is integrally slice then $fracDelta_K''(1)2$ is even; and second, if $K$ is alternating and rationally slice then its signature is zero. (Note that $8_3$ satisfies both of these but is not smoothly slice; I don't know whether it is integrally slice.)

The key observation is that if $K$ is integrally slice, then $S^3_1(K)$ must be integrally homology cobordant to $S^3$. To see this, we take a slice disk bounded by $K$ in a homology ball, and we remove a ball about some point on the disk to get a concordance from $U$ to $K$ inside a homology cobordism from $S^3$ to itself. Performing a 1-surgery along this cylinder gives us the desired homology cobordism.

From here, filling in the $S^3$ end with a ball gives us a smooth homology ball bounded by $S^3_1(K)$. Thus $S^3_1(K)$ has vanishing Rohklin invariant, or equivalently its Casson invariant is even, and the surgery formula for the latter says that $fracDelta_K''(1)2$ must be even.

For the second claim, $S^3_1(K)$ is rationally homology cobordant to $S^3$, so its Heegaard Floer d-invariant must be zero. When $K$ is alternating, this was computed to be $2min(0,-lceil -sigma(K)/4rceil)$ for alternating $K$ by Ozsváth and Szabó (arXiv:0209149, corollary 1.5), so we must have $sigma(K) geq 0$. But the same argument applies to the mirror $m(K)$, with signature $-sigma(K)$, so in fact $K$ must have signature zero.