[Paper Review] Trace reconstruction with $\exp( O( n^{1/3} ) )$ samples

Fëdor Nazarov, Yuval Peres|arXiv (Cornell University)|Dec 12, 2016

DNA and Biological Computing4 references3 citations

TL;DR

This paper presents a novel trace reconstruction algorithm that reconstructs an unknown binary string of length $ n $ from $ \exp(O(n^{1/3})) $ i.i.d. samples of the deletion channel, using statistics of individual output bits. The method leverages complex analysis and generating functions to achieve this exponential improvement over the prior $ \exp(\widetilde{O}(n^{1/2})) $ bound, and proves the bound is tight under the same statistical model.

ABSTRACT

In the trace reconstruction problem, an unknown bit string $x \in \{0,1\}^n$ is observed through the deletion channel, which deletes each bit of $x$ with some constant probability $q$, yielding a contracted string $\widetilde{x}$. How many independent copies of $\widetilde{x}$ are needed to reconstruct $x$ with high probability? Prior to this work, the best upper bound, due to Holenstein, Mitzenmacher, Panigrahy, and Wieder (2008), was $\exp(\widetilde{O}(n^{1/2}))$. We improve this bound to $\exp(O(n^{1/3}))$ using statistics of individual bits in the output and show that this bound is sharp in the restricted model where this is the only information used. Our method, that uses elementary complex analysis, can also handle insertions.

Motivation & Objective

• To reduce the number of i.i.d. traces required for reliable reconstruction of an unknown binary string from the deletion channel.
• To improve upon the prior upper bound of $ \exp(\widetilde{O}(n^{1/2})) $ traces established by Holenstein et al.
• To analyze the information-theoretic limits of reconstruction when only individual bit statistics from the output are used.
• To extend the method to handle insertion and substitution channels, showing similar bounds hold.

Proposed method

• Uses the generating function $ \mathbb{E}\left[\sum_{j\geq 0}(\widetilde{X}_j - \widetilde{Y}_j)w^j\right] = p \sum_{k=0}^{n-1}(x_k - y_k)(pw + q)^k $ to relate the difference in original strings to the difference in output bit statistics.
• Applies complex analysis by choosing $ w $ such that $ pw + q $ lies on the unit circle, enabling the use of roots of unity and symmetry arguments.
• Employs a symmetry argument and padding of output strings with zeros to ensure that the generating function captures full information about the original string.
• Uses the fact that the expected difference in output bit distributions is a low-degree polynomial in $ pw + q $, which can be recovered via sampling.
• Adapts the method to insertion and substitution channels by modifying the generating function transformation: $ w \mapsto \alpha w / (1 - \beta w) $ for insertions and scaling for substitutions.
• Proves the bound is tight in the restricted model by constructing two distinct strings whose output bit statistics are indistinguishable with fewer than $ \exp(\Omega(n^{1/3})) $ traces.

Experimental results

Research questions

• RQ1 Can the number of traces required for trace reconstruction be reduced below $ \exp(\widetilde{O}(n^{1/2})) $ for the deletion channel?
• RQ2 Is $ \exp(O(n^{1/3})) $ traces sufficient for reconstruction using only individual bit statistics from the output?
• RQ3 What is the information-theoretic limit of trace reconstruction when only marginal statistics of output bits are available?
• RQ4 Can the method be extended to handle insertion and substitution noise in addition to deletions?
• RQ5 Is the $ \exp(O(n^{1/3})) $ bound tight under the constraint of using only bit-level statistics?

Key findings

• The paper establishes that $ \exp(O(n^{1/3})) $ i.i.d. traces suffice to reconstruct any unknown binary string $ x \in \{0,1\}^n $ from the deletion channel with high probability.
• The bound is tight in the restricted model where only individual bit statistics from the output are used, as shown by constructing two distinct strings that are indistinguishable with fewer than $ \exp(\Omega(n^{1/3})) $ traces.
• The method relies on complex analysis and generating functions to recover the original string from the statistical differences in output bit distributions.
• The same bound extends to insertion channels with i.i.d. geometric insertions, showing $ \exp(O(n^{1/3})) $ traces suffice for reconstruction.
• For substitution channels, the bound also holds, as the linear transformation of the generating function remains invertible and preserves the required structure.
• The analysis shows that the $ \exp(O(n^{1/3})) $ bound is optimal under the constraint of using only marginal bit statistics from the output traces.