Haah's 3D cubic code was introduced as a candidate for a self-correcting quantum memory in three dimensions, using the absence of string logical operators to obstruct thermal errors. It has been a long-standing open question whether it is a self-correcting quantum memory. Even though the memory time was explored numerically, no analytical upper bound was known. Here, we settle this question rigorously and show that the cubic code thermalizes rapidly at every positive temperature. Our proof relies on the framework of [Commun. Math. Phys. 407, 195 (2026)] to turn a decay-of-correlations condition into a modified logarithmic Sobolev inequality for the Davies generator. To the best of our knowledge, this is the first analytic proof of rapid mixing for all positive temperatures for any fracton code.
A big open question in the quantum information theory concerns the feasibility of a self-correcting quantum memory. A quantum state recorded in such memory can be stored reliably for a macroscopic time without need for active error correction, if the memory is in contact with a cold enough thermal bath. Here we report analytic and numerical evidence for self-correcting behavior in the quantum spin lattice model known as the 3D cubic code. We prove that its memory time is at least L(cβ), where L is the lattice size, β is the inverse temperature of the bath, and c>0 is a constant coefficient. However, this bound applies only if the lattice size L does not exceed a critical value which grows exponentially with β. In that sense, the model can be called a partially self-correcting memory. We also report a Monte Carlo simulation indicating that our analytic bounds on the memory time are tight up to constant coefficients. To model the readout step we introduce a new decoding algorithm, which can be implemented efficiently for any topological stabilizer code. A longer version of this work can be found in Bravyi and Haah, arXiv:1112.3252.
Quantum error-correcting codes (QECs) are essential components of quantum computation and have deep connections to quantum phases of matter. A key obstruction to passive self-correcting QECs is the presence of string logical operators, which can generate logical errors through constant-energy-barrier processes. Haah's Codes (fracton codes) showed that three-dimensional stabilizer codes can forbid such string logical operators, but their translation-invariant structure supports self-similar fractal logical operators with a logarithmic energy barrier. We introduce the qutrit random cubic codes, a family of local qutrit Calderbank-Shor-Steane stabilizer Hamiltonians with similar cube-check structure as Haah's Code 1 but built from spatially varying stabilizers. We prove that these models retain the no-string property and numerically observe that they have properties distinct from translation-invariant fracton codes: the smallest ground-state degeneracy exponent is
k=2 for odd
L and
k=4 for even
L; noncontractible plane-logical operators span the entire logical space; and charge-push diagnostics show that the self-similar fractal operators are absent. These results demonstrate that constrained randomness can fundamentally change the nature of stabilizer codes and improve their self-correction properties. They further point to broader families of quantum error-correcting codes and quantum phases beyond canonical topological and fracton orders.
Cubic codes were proposed by Haah as candidates for self-error correction in three dimensions (3D). While these codes are not self-correcting, Bravyi and Haah showed that they are partially self-correcting. In this paper we are interested in generalizations of the cubic code to prime alphabet. Kim initiated the study of such codes over prime alphabet. Haah also proposed a framework based on modules that enables the study of cubic codes over higher alphabet. However, there are many open questions remaining, especially those pertaining to the performance of nonbinary cubic codes. Building on Bravyi and Haah's decoder, we study the performance of nonbinary cubic codes over the quantum erasure and depolarizing channels. This is the first such study of nonbinary cubic codes.
A self-correcting quantum memory is a type of quantum error correcting code that can correct errors passively through cooling. A major open question in the field is whether self-correcting quantum memories can exist in 3D. In this work, we propose two candidate constructions for 3D self-correcting quantum memories. The first construction is an extension of Haah's code, which retains translation invariance. The second construction is based on fractals with greater flexibility in its design. Additionally, we review existing 3D quantum codes and suggest that they are not self-correcting.
Abstract Self-correcting quantum memories would store logical information without a real-time decoder, and Haah's cubic code is the best studied three-dimensional candidate, but how its memory time depends on the environment beyond a flat Markovian bath has not been mapped. Here we simulate the thermal syndrome dynamics of the cubic code at
β=4.5--
5 for up to
13718 qubits under detailed-balance rate filters on each of its three energy channels, under an Ornstein--Uhlenbeck modulation that gives the bath a correlation time, and at rescaled code energies. The zero-energy diffusion channel controls the memory time: suppressing it by
103 lengthens the memory time
80- to
120-fold for
L=7--
13, the gain enters as a prefactor rather than a steeper size law, and it outlives the flat-bath optimum, where the memory time peaks between
L=13 and
17 and falls by
39% at
L=19 while the filtered bath still grows to
L=17 and gains a factor
45 at
L=19. The pair-creation channel is invisible over three decades; a bath correlation time adds a factor of two to eleven, consistent with annealed and quenched hazard averages; and a uniform
12% energy rescaling raises the size exponent from
2.7 to
4.3 along the Bravyi--Haah line, which no gapped pseudomode we solve provides. These results form a channel-resolved finite-size benchmark for bath engineering of a passive three-dimensional memory; they establish neither asymptotic scaling nor an advantage over active decoding.
We construct self-correcting 3D surface codes over
Zq with local dimension
q growing with system size. The model preserves the local geometry and stabilizer support of the ordinary 3D surface code and remains translation invariant in the bulk, while modifying only the stabilizer Pauli powers. We use the Lee metric to characterize error weight and consider local Metropolis dynamics with single-qudit
X±1 and
Z±1 updates, each changing the energy by at most a constant independent of
q and system size
N. We prove that, below the critical temperature, when the local dimension scales as
q=Θ(N), the memory times in both error sectors grow at least as
exp[Ω(N1/3)]. Our numerical results further support these analytical results. Together, these results establish growing local Hilbert-space dimension as a new resource for designing self-correcting quantum memories while retaining simple local geometry and local thermal dynamics.