Microsoft Researchers and AI Prove Polynomial MIMO Detection Algorithm, Solving 25-Year Problem

GPT-5.6 and Fable 5 collaborated to solve a mathematical problem that had remained open for 25 years, concerning MIMO detection in wireless communications. Microsoft Research principal researcher Dimitris Papailiopoulos participated in proving the result, which demonstrates a polynomial-time algorithm capable of exactly recovering transmitted signals when the signal-to-noise ratio reaches the maximum likelihood threshold. The algorithm carries a complexity of O(N^3) operations, marking a qualitative advance over the exponential complexity of traditional exhaustive search.

MIMO detection is a fundamental problem in wireless communications: the receiver must reconstruct the original bits from noise-corrupted signals. In 2001, Hassibi and Vikalo proposed the sphere decoding algorithm, but in 2005 Jaldén and Ottersten proved that its expected complexity is exponential. Since then, a range of approximation methods, including semidefinite relaxation and bit flipping, have failed to reach the theoretical threshold. In 2020, the box relaxation method attained a signal-to-noise ratio of 4logN, yet the threshold stands at 2logN, leaving the gap unresolved.

The new algorithm proceeds in two steps. The first is LMMSE rounding, which generates a continuous estimate via linear minimum mean square error estimation and rounds it, with proof that the proportion of incorrectly guessed bits approaches zero. The second is greedy bit-by-bit flipping, which at each round flips the single bit that most reduces the cost function until no further improvement is possible. The paper shows that the algorithm achieves exact recovery when the signal-to-noise ratio equals 2logN, with a total complexity of O(N^3) operations, of which the greedy step takes O(NlogN) iterations.

The paper further proves a bidirectional result: above the threshold, the signal can be recovered exactly; below it, maximum likelihood detection itself fails. Dimitris verified the proof, and the result resolves a theoretical question that had gone unsolved for 25 years, providing fresh theoretical support for the field of wireless communications.