ResearchPod Summary
This paper introduces a novel approach to the integer factorization problem on a quantum computer. While Shor’s original algorithm achieves polynomial-time factorization, it inherently relies on probabilistic outcomes with a non-zero failure rate. In contrast, this study constructs an exact quantum circuit family that provides the complete prime factorization of any integer N ≥ 2 with probability one. The construction remains within the bounds of polynomial time, qubit count, and gate count relative to the input bit length.
The construction relies on two primary innovations. First, the authors develop an order-finding trial where the success probability is an explicit dyadic rational—a number with a power-of-two denominator. This allows for precise control over success probabilities using a fixed finite set of gates (NOT, CNOT, Toffoli, Hadamard, and specific phase gates). By replacing the standard Fourier transform with a piecewise-linear phase function, the authors bypass the need for arbitrary-precision rotations, which are typically required in approximate algorithms.
Second, the paper employs a recursive verification technique for auxiliary data. As the algorithm factors an integer, it concurrently generates and validates the prime factorizations of auxiliary values like p-1 (where p is a prime factor of the input). By using these verified factors to retrospectively calculate the exact probability of successful transitions, the authors implement an "exact amplification" scheme. This allows the system to force success with certainty.
This result demonstrates that factoring is in the class of problems that can be solved exactly by a quantum computer. The primary contribution is theoretical, addressing how to manage success probabilities and exact arithmetic in quantum circuits without relying on approximation. The construction confirms that integer factorization does not require variable-angle rotations or unbounded repetitions to achieve absolute certainty, provided the overhead of the exact gate model is accepted.
[[RP_SECTION:problem-with-exact-probability|Problem with exact probability]]
Sam: [plain, direct] Shor's factoring algorithm almost always works, but almost always is not always. A new proof claims to close that gap: a quantum circuit that factors every integer with probability exactly one. It runs in polynomial time and uses only a small, fixed menu of gates.
Alex: [curious, measured] Before the how, place it for me. Whose work is this, and what kind of document are we talking about?
Sam: [matter-of-fact] It's a theory paper, credited to OpenAI and dated September 2026. There are no experiments. The contribution is one theorem and the construction that proves it.
Alex: [probing] Most people I know treat Shor's high success probability as good enough. You just check the factors and rerun. Why should a non-specialist care about pushing that to exactly one?
Sam: [steady] Because rerunning until success has no polynomial guarantee in the worst case. And making rotations ever more precise doesn't give exact correctness with a fixed gate set either. This paper gets certainty and a worst-case polynomial bound together. So we'll cover what earlier exact methods needed, the trick that makes certainty possible, and how far to trust it.
Alex: [inviting] Start with the earlier work. There must have been exact quantum algorithms before this. [[RP_SECTION:quantum-algorithm-history|Quantum algorithm history]]
Sam: [explaining] There were, but in different models. Mosca and Zalka built exact Fourier transforms using rotations computed during the run. Mosca's exact factoring needed that same freedom, gate parameters chosen on the fly. Imran's exact order finding needed someone to hand it a multiple of the unknown answer. The paper calls these material differences in model and input. Here the gate set is fixed, the Hadamard, a quarter-turn phase gate, and a few reversible logic gates. The only input is the number itself.
Alex: [connecting] And the newer factoring algorithms, like Regev's?
Sam: [precise] Regev's construction, and Pilatte's unconditional versions of it, still keep a nonzero failure probability. They address resource costs, which is a different goal.
Alex: [direct] So what's the one idea that turns high probability into certainty? [[RP_SECTION:achieving-exact-certainty|Achieving exact certainty]]
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.
Sam: [unhurried] It's an old trick used in a new place. Suppose you could build a preparation that succeeds with probability exactly one quarter. Not roughly, exactly. Then two reflections finish the job. One flips the sign of the successful part. The other reflects about the starting state. With exactly a quarter, those two moves rotate the state entirely onto success. That's the simplest case of amplitude amplification, the method that came out of quantum search.
Alex: [skeptical] That moves the whole problem into the word "exactly." How do you know the success probability to infinite precision when it depends on the very number you're factoring?
Sam: [warm] That's the heart of the paper, and it splits into two jobs. The first is arithmetic: make a useful event have a probability you can compute exactly. The second is bookkeeping: recover the information needed to compute it.
Alex: [focused] Take the arithmetic first.
Sam: [explaining] Factoring reduces to finding an order. That's the smallest power of a number that leaves remainder one modulo m. Shor finds it with a quantum Fourier transform, which needs very fine phase rotations. This paper swaps those out. It uses a phase function that traces the unit circle in straight lines between four points, the quarter turns. Those four phases are reachable with the Hadamard and quarter-turn gates alone.
Alex: [summarizing] So you trade a smooth circle for a square, and the gates you need become exact.
Sam: [steady] Yes, and you pay with a messier probability distribution. But it's messy in a useful way. On each piece, the probabilities are polynomials of low degree. You can sum them exactly using power sums, without listing every case. Everything comes out as a dyadic rational, a fraction whose denominator is a power of two.
Alex: [curious] Why does a power-of-two denominator matter so much?
Sam: [plain] Because Hadamards give you fair coin flips. Flip enough fair coins and compare the result to a numerator, and you hit any dyadic probability exactly. Then two small classical correction branches top up the mass. The result is that the trial returns the true order with a probability given by an explicit formula in that order. Any candidate it does return is guaranteed to be a multiple of the true order.
Alex: [pushing back] But you still don't know the order while running. A formula in an unknown doesn't give you a known number.
Sam: [measured] Right, and that's the second job. While factoring, the run also factors one less than each odd prime factor it finds, and repeats that recursively. Take 21 as an example. Its prime factors are 3 and 7. So the run also factors 2 and 6. And since 3 divides 6, it factors 2 again underneath. Those extra factorizations let you compute every order exactly, and so every transition's exact probability.
Alex: [connecting] That recursive shape sounds like Pratt's primality certificates.
Sam: [precise] The paper notes the resemblance and says the role is different. Primes are checked by a deterministic primality test in the appendix, adapted from Agrawal, Kayal and Saxena. The extra factorizations exist to supply exact probabilities.
Alex: [skeptical] Here's my worry. You only have this data after the run finishes. Testing earlier steps with data you got from finishing sounds like conditioning on success, or quietly smuggling in the answer. [[RP_SECTION:verification-and-success|Verification and success]]
Sam: [careful] The authors take that objection head on. Each step is mixed with a rare uniform guess of its output. After completion, a filter keeps good outputs with a rounded-down probability. It keeps a guess only if it matches the canonical correct output. That tops each step up to exactly the same passing probability. Then comes the key argument. Imagine running the same tests with the true data and the same coins. Passing those tests forces every output to be good. And all-good outputs force the run to produce the true data. Conversely, any data that passes verification is the true data, by unique factorization. So the two descriptions mark exactly the same runs. No advice, and no conditioning on eventual completion.
Alex: [summarizing] So every step passes with the same known probability, and the product across steps is known in advance.
Sam: [steady] Yes. The run is padded to n to the tenth steps, where n is the bit length of the input. The overall success probability is a known dyadic, comfortably above one half. One last mixing step, with a branch that guesses the whole factor list, brings it to exactly one quarter. Then the two reflections finish it. [[RP_SECTION:trust-and-limitations|Trust and limitations]]
Alex: [evaluative] How much should I trust this? There's no data, just a proof.
Sam: [sober] Then trust rests on the proof checking out. My view is that it's checkable. The construction is explicit, the constants are stated, and each lemma is self-contained. But we're working from an excerpt, some steps are omitted, and no venue or review is mentioned. Treat it as a claimed theorem awaiting careful scrutiny.
Alex: [practical] And the main limitation?
Sam: [plain] Cost. Inputs are padded to at least 128 bits, and the polynomials are huge. The authors say outright these parameters are not meant as practical improvements. They also claim nothing about classical factoring. And, as my own reading, exactness here lives in an idealized model with perfect gates. It says nothing about noisy hardware.
Alex: [wrapping up] So who should sit down with the full paper, and where should they start?
Sam: [decisive] Anyone in quantum complexity or exact quantum algorithms. Start with Section 5, on equalizing each step and identifying the deferred tests with the true-data tests. That argument is the new part. Then read Section 3 for the four-phase transform and its exact sums.
Alex: [reflective] And for everyone else, the one line to keep?
Sam: [memorable] Certainty didn't come from finer rotations. It came from knowing the odds exactly, then one reflection.
Alex: [calm] Exact odds, then a single step.