# Problem unsolvable in $2^{o(n)}$ on inputs with $n$ bits, assuming ETH?

If we assume the Exponential-Time Hypothesis, then there is no $2^{o(n)}$ algorithm for $n$-variable 3-SAT, and many other natural problems, such as 3-COLORING on graphs with $n$ vertices. Notice though that, in general, encoding the input for $n$-variable 3-SAT or $n$-vertex 3-COLORING takes something like $O(n\log n)$ bits. For example, to describe a sparse graph as input to 3-COLORING, for each edge we would have to list its endpoints. So the lower bound is not exponential in the length of the input. Therefore, my question is the following:

Is there a problem for which no $2^{o(n)}$ algorithm exists for inputs of length $n$ bits (assuming ETH)?

Ideally, the problem would be in NP (no cheating with succinct NEXP-hard problems!) and be reasonably natural, but I won't be picky.

Let me also note that after digging around I found that there are efficient ways to encode planar graphs with $O(n)$ bits. So, if one could find a problem that takes time exponential in the number of vertices even for planar graphs, the question would be settled. However, because planar graphs have treewidth $O(\sqrt{n})$, most natural problems have sub-exponential algorithms in this case.

• If you are fine with dropping the condition that the problem is in NP, then you can use any undecidable problem and you do not even need the exponential time hypothesis. So probably you should be picky. – Tsuyoshi Ito Jan 20 '13 at 23:18
• A problem in NP for which our present algorithmic knowledge is consistent with such a ETH result is edge chromatic number in a dense graph. AFAIK, we do not yet know of any $2^{o(|E|)}$ time algorithms... – Andreas Björklund Jan 21 '13 at 6:33
• It is probably not possible to compress $n$ variable SAT instances to $O(n \log n)$ as stated in the question (Dell, van Malkebeek, "Satisfibility Allows No Nontrivial Sparsification Unless The Polynomial-Time Hierarchy Collapses"). However, 3-SAT has no $2^{o(m)}$ algorithm where $m$ is the number of clauses, as shown by Impagliazzo and Paturi (assuming ETH). Since the hard cases have $O(m)$ variables we have $b=O(m \log m)$, so $m=\Omega(b/\log b)$, so 3-SAT has no $2^{o(b/\log b)}$ algorithm where $b$ is the input size. – Colin McQuillan Jan 21 '13 at 13:36
• Can you point to a reference where I can find efficient ways to encode planar graphs with O(n) bits? (I assume that n is the number of vertices). If so, there are NP-Complete problems for planar graphs and maybe the problem can also be presented in O(n) bits using these methods. Thanks. – Avi Tal Aug 16 '18 at 15:22
• @AviTal This paper comes to mind arxiv.org/abs/cs/0102005 . The usual keywords for this line of research seem to be "compact" and "succinct" encodings or representations. (succinct is stronger in that it is optimal up to an additive term instead of a multiplicative factor for compact). – A.N. Sep 22 '18 at 13:26