We show improved fine-grained hardness of two key lattice problems in the
ℓp norm: Bounded Distance Decoding to within an α factor of the
minimum distance (BDDp,α) and the (decisional)
γ-approximate Shortest Vector Problem (SVPp,γ),
assuming variants of the Gap (Strong) Exponential Time Hypothesis (Gap-(S)ETH).
Specifically, we show:
1. For all p∈[1,∞), there is no 2o(n)-time algorithm for
BDDp,α for any constant α>αkn,
where αkn=2−ckn<0.98491 and ckn
is the ℓ2 kissing-number constant, unless non-uniform Gap-ETH is false.
2. For all p∈[1,∞), there is no 2o(n)-time algorithm for
BDDp,α for any constant α>αp‡, where
αp‡ is explicit and satisfies αp‡=1 for 1≤p≤2, αp‡<1 for all p>2, and αp‡→1/2 as p→∞, unless randomized Gap-ETH is false.
3. For all p∈[1,∞)∖2Z and all C>1, there
is no 2n/C-time algorithm for BDDp,α for any constant
α>αp,C†, where αp,C† is explicit and
satisfies αp,C†→1 as C→∞ for any fixed p∈[1,∞), unless non-uniform Gap-SETH is false.
4. For all p>p0≈2.1397, p∈/2Z, and all C>Cp, there is no 2n/C-time algorithm for SVPp,γ for
some constant γ>1, where Cp>1 is explicit and satisfies Cp→1 as p→∞, unless randomized Gap-SETH is false.