← Latest papers
⚛️ quantum physics

Postselection on Halting and Reductions to the Halting Problem

This paper demonstrates that probabilistic Turing machines with postselection on halting, where the correct answer is merely more likely than the wrong one, exactly characterize the class of problems solvable via adaptive queries to the halting problem, while also quantifying how specific probability bounds on halting and correctness determine the number of non-adaptive queries that can be replaced.

Original author: Saman Farhat, Farzad Didehvar

Published 2026-11-17
📖 5 min read🧠 Deep dive

Original author: Saman Farhat, Farzad Didehvar

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer

In the vast landscape of computer science, there is a fundamental boundary known as the halting problem. It is a simple question with a devastating answer: given a computer program and an input, can we always predict whether the program will eventually stop running or if it will get stuck in an infinite loop forever? Mathematicians have proven that no single algorithm can answer this question for every possible program. This limitation defines the edge of what is computable. But what happens if we allow a computer to be probabilistic, meaning it can flip a coin to make decisions, and what if we are willing to ignore the runs where the computer never stops? This is the territory explored by researchers Saman Farhat and Farzad Didehvar. They investigate a specific type of machine that runs on coin flips, discards any attempt that runs forever, and only counts the results from the attempts that do finish. The central question they address is how much power this "postselection" on halting gives a computer, and exactly how the rules of the game change the kinds of problems the machine can solve.

The researchers began by examining a scenario where the machine must be quite confident in its answer. If the correct answer is more than twice as likely to appear as the wrong one among the finished runs, the machine can solve a specific set of difficult problems. These are problems that can be answered by asking a fixed list of questions about the halting problem all at once, without waiting for the answers to the first questions to decide the next ones. This is a known limit. However, the authors asked what would happen if they relaxed that rule. What if the machine only needed to be slightly more likely to be right than wrong, without the strict "twice as likely" requirement? They discovered that this small change in confidence unlocks a much larger class of problems. When the machine is allowed to have an unbounded error margin—meaning the correct answer just needs to be the most probable one, even if only by a tiny amount—it gains the ability to solve problems that require asking questions about the halting problem one after another, where each new question depends on the answer to the previous one.

This finding connects two seemingly different concepts: the probability of a machine stopping and the number of times a guess changes its mind. The researchers showed that probability acts like a budget. Every time a machine changes its mind about the answer, it must spend a portion of its total probability of stopping. If the machine is required to stop with a certain minimum frequency, and its correct answer must be more likely than the wrong one by a specific margin, there is a strict mathematical limit on how many times it can change its mind. This limit, in turn, dictates exactly how many questions it can ask about the halting problem in a fixed, non-adaptive list. The authors calculated this trade-off precisely. They found that for any given requirement on how often the machine must stop and how sure it must be of its answer, there is an exact number of pre-planned questions it can replace. If the requirements are too strict, the machine can only solve simple, everyday problems. If the requirements are looser, it can solve increasingly complex problems, up to the point where it can solve any problem that can be solved by asking questions one by one.

The study also explored what happens when these machines are stacked inside one another, or when they are given access to a "black box" that can answer specific questions instantly. They found that the same rules apply at every level of complexity. A machine nested inside another, or one using an oracle, follows the same logic: the probability budget and the confidence margin determine the depth of the questions it can ask. Interestingly, the researchers showed that quantum computers, which use the strange properties of quantum bits instead of simple coin flips, do not gain any extra power in this specific context. They decide exactly the same set of problems as their classical, coin-flipping counterparts. This suggests that the power comes not from the type of randomness used, but from the fundamental way the machine filters its own history, keeping only the successful paths and discarding the rest.

Ultimately, this work provides a complete map of what these probabilistic machines can achieve. It clarifies that the ability to solve hard problems is not just about having a powerful processor, but about how the machine manages its own uncertainty. By treating probability as a resource that is spent every time a guess is revised, the researchers turned an abstract question about computer limits into a concrete accounting problem. They proved that the boundary between what is solvable and what is not is determined by a simple balance: how often the machine is willing to stop, and how much more likely it must be to be right than wrong. This balance dictates whether the machine can solve a problem with a single glance, a fixed list of questions, or a dynamic, step-by-step inquiry. The results confirm that while postselection on halting is a powerful tool, it is not magic; it is a precise mechanism with strict, calculable limits.

Drowning in papers in your field?

Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.

Try Digest →