Finite-Length Quantum Hamming Bound

• Finite-Length Quantum Hamming Bound is a non-asymptotic sphere-packing limit that ensures the code dimension times correctable error patterns fit within the ambient Hilbert space.
• The analysis employs advanced linear programming and Krawtchouk polynomial techniques to validate the bound for both nondegenerate and degenerate quantum subspace codes.
• Recent breakthroughs have established the bound across binary and nonbinary codes, resolving longstanding finite-length challenges in exact quantum error correction.

The finite-length quantum Hamming bound is the non-asymptotic sphere-packing constraint for exact quantum error correction. For a quantum subspace code of length $n$, dimension $K=q^k$, and distance $d$, with $t=\left\lfloor \frac{d-1}{2}\right\rfloor$, it requires that the code dimension times the number of correctable local error patterns fit inside the ambient Hilbert space. In its standard $q$-ary form, the bound is

$K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$

or equivalently

$$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$

For nondegenerate codes this follows from disjoint error spheres; for degenerate codes the same inequality was for decades a central open finite-length question because distinct correctable errors can act identically on the code subspace, invalidating the naive counting argument. The modern literature shows a progression from partial finite-length and distance-restricted results to unconditional theorems for binary subspace codes and then for arbitrary local dimension (Dallas et al., 2022, Zhang et al., 14 Jun 2026, Zhang et al., 21 Jun 2026).

1. Basic formulation and sphere-packing interpretation

A quantum code is commonly written as $((n,K,d))_q$ or, when $K=q^k$, as $[[n,k,d]]_q$. It is a $K=q^k$0-dimensional subspace $K=q^k$1 with minimum distance $K=q^k$2, so that it exactly corrects all errors of weight at most

$K=q^k$3

Exact correctability is characterized by the Knill–Laflamme condition

$K=q^k$4

with $K=q^k$5 the projector onto the code subspace. In the qubit Pauli picture, an error is an $K=q^k$6-fold tensor product of $K=q^k$7, and there are $K=q^k$8 distinct Pauli errors of weight $K=q^k$9 (Dallas et al., 2022, Zhang et al., 14 Jun 2026).

For binary nondegenerate subspace codes, the finite-length Hamming bound is

$d$0

Equivalently, with

$d$1

one has

$d$2

The interpretation is a finite-length sphere-packing statement in Pauli Hamming space: each correctable error of weight at most $d$3 maps the code to an orthogonal syndrome sector, and the total number of such sectors times their dimension must not exceed the full physical dimension (Zhang et al., 14 Jun 2026). For small $d$4, this specializes to

$d$5

and

$d$6

(Zhang et al., 14 Jun 2026).

The same structure holds for arbitrary local dimension. Writing $d$7, the $d$8-ary finite-length quantum Hamming bound is

$d$9

which is the $t=\left\lfloor \frac{d-1}{2}\right\rfloor$0-ary Hamming-ball volume in the Hamming scheme associated with the phase-free Weyl error basis (Dallas et al., 2022, Zhang et al., 21 Jun 2026).

2. Degeneracy and the historical finite-length obstruction

The decisive subtlety is degeneracy. A code is degenerate if there exist distinct correctable errors $t=\left\lfloor \frac{d-1}{2}\right\rfloor$1 of weight at most $t=\left\lfloor \frac{d-1}{2}\right\rfloor$2 such that

$t=\left\lfloor \frac{d-1}{2}\right\rfloor$3

or equivalently if different physical errors map the code space into overlapping or identical corrupted subspaces (Zhang et al., 14 Jun 2026). In the notation of Knill–Laflamme, degeneracy corresponds to rank deficiency in the error-overlap matrix. In that case the usual disjoint-sphere argument breaks down, because the number of correctable error operators can exceed the number of distinct syndrome subspaces (Dallas et al., 2022).

This created a long-standing finite-length question: can degeneracy allow exact quantum subspace codes to violate the Hamming count? In binary notation, that would mean

$t=\left\lfloor \frac{d-1}{2}\right\rfloor$4

The literature repeatedly emphasized that no construction was known, but no general impossibility proof existed for exact degenerate codes (Dallas et al., 2022, Zhang et al., 14 Jun 2026).

A standard way to formalize the problem is through Krawtchouk-polynomial linear programming. For qubits, the Krawtchouk polynomials are

$t=\left\lfloor \frac{d-1}{2}\right\rfloor$5

and Li–Xing proved a quantum LP bound of the form

$t=\left\lfloor \frac{d-1}{2}\right\rfloor$6

for suitable polynomial choices obeying positivity and sign constraints (Dallas et al., 2022, Zhang et al., 14 Jun 2026). This framework became the basis for nearly all later finite-length progress.

The technical obstacle was pointwise rather than asymptotic. Earlier structural and LP results showed that large regions of parameter space were inaccessible to violations, but the existence of a single exceptional finite pair $t=\left\lfloor \frac{d-1}{2}\right\rfloor$7 for a degenerate exact code remained unresolved until 2026 in the binary case and shortly thereafter in general local dimension (Zhang et al., 14 Jun 2026, Zhang et al., 21 Jun 2026).

3. Finite-length partial results before the general theorems

Several results established the Hamming bound in restricted regimes and sharpened it in specialized settings.

Aly proved that the standard Hamming bound holds for all $t=\left\lfloor \frac{d-1}{2}\right\rfloor$8 stabilizer codes with $t=\left\lfloor \frac{d-1}{2}\right\rfloor$9, irrespective of degeneracy, and also recalled that distance-$q$0 nonbinary stabilizer codes obey the bound (0711.4603). In the same paper, the finite-length Hamming expressions for $q$1 and $q$2 were written as

$q$3

and

$q$4

respectively, and these were used together with the Singleton bound to limit the possible lengths of single- and double-error-correcting quantum MDS stabilizer codes (0711.4603).

Ashikhmin and Litsyn’s style of quantum LP argument was later sharpened into explicitly stronger finite-length inequalities. The paper “Strengthened quantum Hamming bound” combined the quantum Hamming and Singleton bounds into a single family

$q$5

with $q$6, and then strengthened this further through a quantum Lloyd-theoretic correction term. For $q$7, the strengthened bound was shown to hold even for impure codes, and for stabilizer codes the paper identified an infinite family of lengths with a $q$8-logical-qudit improvement over the standard Hamming bound (Yu et al., 2010).

A different direction was developed for degenerate stabilizer codes. “A Hamming-Like Bound for Degenerate Stabilizer Codes” proved that for each $q$9 there exists a positive integer $K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$0 such that every degenerate $K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$1 stabilizer code with $K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$2 satisfies

$K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$3

which is stricter than the standard binary Hamming bound for large enough $K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$4 (Nemec et al., 2023). For $K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$5, the paper obtained the all-length inequality

$K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$6

and concluded that all but a sparse set of optimal distance-$K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$7 stabilizer-code lengths force optimal codes to be nondegenerate (Nemec et al., 2023).

These results did not settle the full finite-length problem for arbitrary exact subspace codes, but they narrowed the search space sharply and supplied the key tools later used in the complete theorems.

4. The 2022 finite-length exclusion up to distance $K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$8

A major finite-length advance came from Dallas, Andreadakis, and Lidar, who combined a universal structural lower bound on $K \sum_{j=0}^{t} (q^2-1)^j \binom{n}{j} \le q^n,$9 with Li–Xing’s finite-length LP threshold argument to prove that no qubit code of distance $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$0 can violate the quantum Hamming bound, even allowing degeneracy and nonadditivity (Dallas et al., 2022).

The first ingredient is Rains’ shadow-enumerator theorem, which implies for all qubit codes with $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$1 that

$$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$2

This bound is structural and does not assume nondegeneracy (Dallas et al., 2022).

The second ingredient is the Li–Xing implication that for each fixed distance $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$3, there exists an integer $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$4 such that every $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$5 code with $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$6 satisfies the quantum Hamming bound. Dallas, Andreadakis, and Lidar numerically computed these thresholds for odd distances up to $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$7, extending earlier calculations that had reached only $$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$8. Their numerical definition was

$$K \, V_t^{(q^2)}(n) \le q^n,\qquad V_t^{(Q)}(n)=\sum_{j=0}^t (Q-1)^j\binom{n}{j},\quad Q=q^2.$$9

with

$((n,K,d))_q$0

(Dallas et al., 2022).

Operationally, the computation looped over $((n,K,d))_q$1, recursively generated binomial coefficients using

$((n,K,d))_q$2

evaluated Krawtchouk polynomials, formed the relevant ratios, and stopped once the maximizing shell became $((n,K,d))_q$3. The implementation used dynamic programming, and the measured runtime scaled approximately as $((n,K,d))_q$4, with a log-log slope $((n,K,d))_q$5 and correlation coefficient $((n,K,d))_q$6. The thresholds were computed up to $((n,K,d))_q$7 using USC’s CARC cluster (Dallas et al., 2022).

The decisive observation was that for every odd $((n,K,d))_q$8,

$((n,K,d))_q$9

Hence any code of distance $K=q^k$0 must satisfy $K=q^k$1, but every such $K=q^k$2 is already in the Li–Xing regime where the Hamming bound is automatic. This yields the theorem that for qubit codes there is no $K=q^k$3 code with $K=q^k$4 that violates the quantum Hamming bound, regardless of degeneracy or additivity (Dallas et al., 2022).

This result did not prove the bound for all distances, but it effectively closed the moderate-distance regime and converted the general problem into a residual high-distance finite-length question.

5. Complete binary resolution: degeneracy cannot violate the bound

The binary finite-length problem was settled in 2026 by the theorem that every exact binary quantum subspace code with $K=q^k$5 obeys the Hamming bound, without assuming nondegeneracy or additivity (Zhang et al., 14 Jun 2026). In its basic form, for any exact binary $K=q^k$6 code correcting all Pauli errors of weight at most $K=q^k$7,

$K=q^k$8

The proof recast the Li–Xing LP polynomial as an exact intersection count for quaternary Hamming balls. Let $K=q^k$9, let

$[[n,k,d]]_q$0

and for a separation vector $[[n,k,d]]_q$1 of weight $[[n,k,d]]_q$2, define the two-ball intersection

$[[n,k,d]]_q$3

With the quaternary Lloyd polynomial

$[[n,k,d]]_q$4

the core Fourier identity is

$[[n,k,d]]_q$5

This turns the LP ratio into

$[[n,k,d]]_q$6

To recover the Hamming bound, it suffices to show

$[[n,k,d]]_q$7

(Zhang et al., 14 Jun 2026).

The proof then established two monotonicity reductions. First, $[[n,k,d]]_q$8 is nonincreasing in the block length $[[n,k,d]]_q$9, using the recurrences

$K=q^k$00

$K=q^k$01

$K=q^k$02

together with inequalities $K=q^k$03 among normalized ratios derived from these recurrences (Zhang et al., 14 Jun 2026). Second, Rains’ shadow bound implies $K=q^k$04 for binary subspace codes with $K=q^k$05, so it is enough to prove the desired inequality at the minimal admissible length $K=q^k$06.

At that shortest length, the remaining task was reduced to monotonicity in the center separation $K=q^k$07. The paper introduced recursions

$K=q^k$08

and derived the criterion

$K=q^k$09

The technical core was then a node–edge decomposition of the two-center intersection counts: $K=q^k$10 followed by an edge-charging argument based on relations such as

$K=q^k$11

This yielded uniform node-load bounds and ultimately

$K=q^k$12

which by length monotonicity implies $K=q^k$13 for every admissible $K=q^k$14 (Zhang et al., 14 Jun 2026).

The conclusion is exact and unconditional within its scope: degeneracy can merge correctable error sectors, but cannot enlarge the finite-length binary Hamming bound (Zhang et al., 14 Jun 2026).

6. Arbitrary local dimension, related extensions, and nearby notions

Within a week of the binary endpoint theorem, Zhang and Chen proved the finite-length quantum Hamming bound for every exact $K=q^k$15-ary subspace code with $K=q^k$16, thereby closing the nonbinary finite-length gap. Combined with the binary theorem, this established the bound in arbitrary local dimension (Zhang et al., 21 Jun 2026).

The nonbinary proof again uses Li–Xing LP normalization, but now in the $K=q^k$17 Hamming scheme. For the raw Lloyd square,

$K=q^k$18

the obstruction is the two-center ratio

$K=q^k$19

The target is

$K=q^k$20

For $K=q^k$21 the paper proved a uniform half-gap

$K=q^k$22

after reducing by monotonicity to the critical length $K=q^k$23 and analyzing a Jacobi-matrix model for the normalized intersection ratios. Qutrits, $K=q^k$24, formed a boundary case: the half-gap disappears, so the paper split the analysis into a short Singleton–entropy regime, a long regime where the raw Lloyd square remains sufficient, and a bridge window handled by a quadratic-filtered Lloyd square, coefficient-certificate reduction, and a Stein–tangent positivity argument (Zhang et al., 21 Jun 2026).

The final theorem is that for every $K=q^k$25, every exact quantum subspace code $K=q^k$26 satisfies

$K=q^k$27

regardless of degeneracy (Zhang et al., 21 Jun 2026).

A concise chronology is as follows:

Result	Scope	Main statement
(0711.4603)	Nonbinary stabilizer, $K=q^k$28, $K=q^k$29	Hamming bound holds for all $K=q^k$30 stabilizer codes
(Dallas et al., 2022)	Qubit subspace codes, $K=q^k$31	No $K=q^k$32 code can violate the QHB
(Zhang et al., 14 Jun 2026)	Binary exact subspace codes, $K=q^k$33	Degeneracy cannot violate the finite-length binary Hamming bound
(Zhang et al., 21 Jun 2026)	Exact subspace codes, all $K=q^k$34	Finite-length quantum Hamming bound holds in arbitrary local dimension

The exact finite-length theorem does not cover subsystem codes, approximate correction, or hybrid continuous-variable/discrete-variable architectures. Those nearby regimes behave differently in the current literature. Haar-random codes can approximately correct a set of $K=q^k$35 unitary errors whenever $K=q^k$36, which approximately attains the Hamming scaling but only in diamond norm and only for approximate QEC (Ma et al., 8 Oct 2025). By contrast, a concatenated GKP–QLDPC scheme was reported to surpass the CSS Hamming bound $K=q^k$37 for certain code families when analog GKP information and a sequential min-sum decoder are used, but that benchmark applies to nondegenerate CSS codes and does not contradict the exact subspace-code theorem (Raveendran et al., 2021).

The finite-length quantum Hamming bound is therefore now best understood in a sharply delimited sense. For exact quantum subspace codes, the standard Hamming count is universal at finite block length and in arbitrary local dimension. Degeneracy is a genuine structural feature of QECCs, but not a loophole in finite-length sphere packing (Zhang et al., 14 Jun 2026, Zhang et al., 21 Jun 2026).