A Symplectic Proof of the Quantum Singleton Bound Frederick Dehmel, Shilun Li Abstract We present a symplectic linear algebraic proof of the Quantum Singleton Bound for stabiliser quantum error–correcting codes, together with a complete Lean4 formalisation of the linear–algebraic argument. The argument is formulated in terms of a finite dimensional symplectic vector space model of Pauli operators, distance based erasure correctability, and the cleaning lemma. 1 Motivation The Quantum Singleton Bound is a basic constraint on the parameters of a stabiliser code, but it is often proved using entropy inequalities and other information–theoretic tools. For stabiliser codes, however, the underlying structure is symplectic linear algebra: Paulis form a finite-dimensional symplectic space, commutation is bilinear, and erasure correctability is captured by the inclusion S^{\perp}\cap V_{E}\subseteq S . This viewpoint yields a proof that is conceptually aligned with the stabiliser formalism and isolates the algebraic ingredients (distance \Rightarrow correctable erasures, plus the cleaning lemma). A further benefit is mechanisation: the linear-algebraic argument is well-suited to formal verification, and we have formalised it in Lean4 . In contrast, entropy-based proofs typically require substantially heavier analytic infrastructure, making them harder to machine-check in a lightweight way. 2 Contributions Concretely, we: (i) give a symplectic linear–algebraic proof of the Quantum Singleton Bound for [[n,k,d]] stabiliser codes, deriving k+2(d-1)\leq n from distance-based erasure correctability and the cleaning lemma via a dimension argument; (ii) accompany the proof with a Lean4 formalisation of the main linear–algebraic development. 3 Introduction The Quantum Singleton Bound is a fundamental constraint on the parameters of a quantum error– correcting code. For an [[n,k,d]] stabiliser code it states k+2(d-1)\leq n. We prove this bound using the symplectic stabiliser formalism together with two key structural inputs: (i) distance implies correctability of erasures of size at most d-1 ; (ii) the cleaning lemma and its dimension identity g(M)+g(M^{c})=2k . 4 Conventions and symplectic model Remark 1 (Field convention) . Throughout, \mathbb{F} denotes a prime field \mathbb{F}_{p} . We work in the symplectic vector space V:=\mathbb{F}^{n}\times\mathbb{F}^{n}\cong\mathbb{F}^{2n} with elements written v=(v_{X},v_{Z}) . Definition 1 (Standard symplectic form) . For u=(u_{X},u_{Z}) and v=(v_{X},v_{Z}) in V , define \langle u,v\rangle:=\sum_{i=1}^{n}\bigl(u_{X}(i)v_{Z}(i)-u_{Z}(i)v_{X}(i)\bigr)\in\mathbb{F}. Lemma 1 (Nondegeneracy) . If u\in V satisfies \langle u,v\rangle=0 for all v\in V , then u=0 . Proof. Fix i . Pairing with (0,e_{i}) forces u_{X}(i)=0 , and pairing with (e_{i},0) forces u_{Z}(i)=0 . Hence all coordinates vanish. ∎ 5 Support, restriction, and distance Let [n]:=\{1,\dots,n\} . Definition 2 (Support and weight) . For v\in V , define \mathrm{supp}(v):=\{i\in[n]:(v_{X}(i),v_{Z}(i))\neq(0,0)\},\qquad\mathrm{wt}(v):=|\mathrm{supp}(v)|. Definition 3 (Support subspace) . For C\subseteq[n] , define V_{C}:=\{v\in V:v_{X}(i)=v_{Z}(i)=0\text{ for all }i\notin C\}. Lemma 2 (Dimension of a support subspace) . \dim(V_{C})=2|C| . Proof. Restriction to C yields an isomorphism V_{C}\cong\mathbb{F}^{|C|}\times\mathbb{F}^{|C|} . ∎ Definition 4 (Restriction map) . For E\subseteq[n] , define r_{E}:V\to V_{E} by zeroing coordinates outside E . Definition 5 (stabiliser subspace and distance) . A stabiliser code Q is specified by an isotropic subspace S\subseteq V with \dim(S)=n-k . Define the (Pauli) distance d:=\min\{\mathrm{wt}(v):v\in S^{\perp}\setminus S,\ v\neq 0\}. Lemma 3 (Logical space dimension) . If \dim(S)=n-k , then \dim(S^{\perp}/S)=2k . Proof. Nondegeneracy gives \dim(S)+\dim(S^{\perp})=2n , hence \dim(S^{\perp})=n+k and \dim(S^{\perp}/S)=2k . ∎ 6 Erasure correctability Definition 6 (Correctable erasure (commutant form)) . A subset E\subseteq[n] is correctable if S^{\perp}\cap V_{E}\subseteq S. In the stabiliser setting this condition is equivalent to the standard Knill-Laflamme error-correction condition for erasures supported on E [ 1 ] . Lemma 4 (Distance implies erasure correctability) . If |E|\leq d-1 , then E is correctable. Proof. If v\in S^{\perp}\cap V_{E} and v\notin S , then v\in S^{\perp}\setminus S with \mathrm{wt}(v)\leq|E|\leq d-1 , contradicting the definition of d . ∎ 7 Cleaning lemma and the dimension identity Definition 7 (Supportable logical operators) . For M\subseteq[n] , define g(M):=\dim\bigl((S^{\perp}\cap V_{M})/(S\cap V_{M})\bigr). This counts the number of independent logical Pauli operators (modulo stabilisers) that can be supported on M . Lemma 5 (Cleaning dimension identity) . For any M\subseteq[n] , g(M)+g(M^{c})=2k. Proof. The following argument is a linear–algebraic reformulation of the dimension-counting proof appearing in Preskill’s notes [ 2 ] . Let P\cong V denote the abelianised Pauli group viewed as a 2n –dimensional \mathbb{F} –vector space with the same symplectic form. Fix M\subseteq[n] and write M^{c} for its complement. Let P_{M}:=V_{M} and P_{M^{c}}:=V_{M^{c}} . Define the stabiliser–restriction subspaces S_{M}:=S\cap P_{M},\qquad S_{M^{c}}:=S\cap P_{M^{c}}. Choose a complementary subspace S_{0} such that S=S_{M}\oplus S_{M^{c}}\oplus S_{0}. Consider the restriction map r_{M} to M . We claim that the restriction of S_{M}\oplus S_{0} to M is injective: \dim\bigl(S_{M}\oplus r_{M}(S_{0})\bigr)=\dim(S_{M})+\dim(S_{0}). Indeed, if a nontrivial linear combination of elements of S_{M}\oplus S_{0} restricts to zero on M , then it is supported on M^{c} , hence lies in S\cap P_{M^{c}}=S_{M^{c}} , which is impossible unless the combination was trivial. Now compute \dim(S^{\perp}\cap P_{M}) . Since P_{M} is orthogonal to S_{M^{c}} (disjoint support), the constraints on commuting with S inside P_{M} are exactly orthogonality to r_{M}(S_{M}\oplus S_{0})=S_{M}\oplus r_{M}(S_{0}) . Therefore, \dim(S^{\perp}\cap P_{M})=\dim(P_{M})-\dim(S_{M})-\dim(S_{0})=2|M|-\dim(S_{M})-\dim(S_{0}). Similarly, \dim(S^{\perp}\cap P_{M^{c}})=2|M^{c}|-\dim(S_{M^{c}})-\dim(S_{0}). By definition, g(M)=2|M|-2\dim(S_{M})-\dim(S_{0}),\qquad g(M^{c})=2|M^{c}|-2\dim(S_{M^{c}})-\dim(S_{0}). Adding and using |M|+|M^{c}|=n and \dim(S)=\dim(S_{M})+\dim(S_{M^{c}})+\dim(S_{0})=n-k gives g(M)+g(M^{c})=2k. ∎ Remark 2 . It is worthwhile noting that the identity g(M)+g(M^{c})=2k does not imply that each logical Pauli operator is supportable on exactly one of M or M^{c} . Lemma 6 (Cleaning lemma (support on the complement)) . If M is correctable, then every logical class [L]\in S^{\perp}/S has a representative supported on M^{c} . Proof. If M is correctable then S^{\perp}\cap V_{M}\subseteq S , hence g(M)=0 . By the dimension identity, g(M^{c})=2k=\dim(S^{\perp}/S) . ∎ 8 A symplectic consequence of correctable erasure Lemma 7 (Two disjoint correctable sets bound logical dimension) . Partition the physical qudits into A\,B\,C with A\cap B=\varnothing and C=[n]\setminus(A\cup B) . If erasures of both A and B are correctable, then k\leq|C|. Proof. Set D:=A^{c}=B\cup C . Since A is correctable we have g(A)=0 , hence by the cleaning dimension identity g(D)=g(A^{c})=2k. By definition, g(D)=\dim\Bigl((S^{\perp}\cap V_{D})/(S\cap V_{D})\Bigr), so the quotient space L:=(S^{\perp}\cap V_{D})/(S\cap V_{D}) has dimension \dim(L)=2k . Consider the restriction map r_{C}:V_{D}\to V_{C} (zeroing coordinates outside C ). Let W:=S\cap V_{D} . Then r_{C}(W)\leq V_{C} is a subspace, and r_{C} induces a linear map \bar{r}_{C}:L\longrightarrow V_{C}/r_{C}(W),\qquad[v]\longmapsto r_{C}(v)\ \bmod\ r_{C}(W). This is well-defined because replacing v by v+w with w\in W changes r_{C}(v) by an element of r_{C}(W) . We claim \bar{r}_{C} is injective. Suppose \bar{r}_{C}([v])=0 . Then r_{C}(v)\in r_{C}(W) , so there exists w\in W with r_{C}(v)=r_{C}(w) . Hence r_{C}(v-w)=0 , so v-w is supported entirely on B , i.e. v-w\in V_{B} . Also v-w\in S^{\perp} (since v\in S^{\perp} and w\in S ). Thus v-w\in S^{\perp}\cap V_{B} , and because B is correctable we get v-w\in S . Since w\in S as well, it follows that v\in S , hence [v]=0 in L . Therefore, 2k=\dim(L)\leq\dim\bigl(V_{C}/r_{C}(W)\bigr)\leq\dim(V_{C})=2|C|, so k\leq|C| . ∎ 9 Quantum Singleton bound Theorem 1 (Quantum Singleton bound) . For any [[n,k,d]] stabiliser code, k+2(d-1)\leq n. Proof. Choose disjoint A,B\subseteq[n] with |A|=|B|=d-1 and let C=[n]\setminus(A\cup B) . By distance–correctability, erasures of A and B are correctable. Lemma 7 yields k\leq|C|=n-2(d-1), which is equivalent to the stated bound. ∎ Remark 3 (Existence of disjoint correctable sets) . In the proof above we choose disjoint subsets A,B\subseteq[n] with |A|=|B|=d-1 . Such a choice is possible whenever n\geq 2(d-1) . If instead n<2(d-1) , then no such disjoint subsets exist. In this regime the inequality k\leq n-2(d-1) has a non-positive right-hand side, and since k\geq 0 for any stabiliser code, the bound is automatically satisfied. Hence the argument above covers all cases. 10 Conclusion We derived the Quantum Singleton Bound from (i) symplectic stabiliser linear algebra, (ii) distance–based erasure correctability, and (iii) a dimension–counting proof of the cleaning lemma following Preskill. The final inequality follows from a purely symplectic dimension argument using correctability of two disjoint erasures. Appendix A Lean4 formalisation and code index A.1 Repository and permalinks The formalisation is in the tcslib repository at TCSlib/CodingTheory/QuantumSingleton.lean . A.2 Pointers to core definitions Mathematical object (paper) Lean declaration (GitHub link) Prime field \mathbb{F}=\mathbb{F}_{p} abbrev F Ambient space V\cong\mathbb{F}^{2n} abbrev V Standard symplectic form \langle\cdot,\cdot\rangle def sym_form Bilinear form packaging (as LinearMap.BilinForm ) noncomputable def symB Support and weight \mathrm{supp}(\cdot),\ \mathrm{wt}(\cdot) def supp , def wt Support subspace V_{C} (vectors supported on C ) def V_sub Dimension of a support subspace \dim(V_{C})=2|C| lemma dim_V_sub Restriction map r_{E} def r_E Symplectic orthogonal complement S^{\perp} abbrev sym_orth Isotropic subspace predicate def IsIsotropic Code distance (as an infimum over weights) noncomputable def code_dist Correctable erasure (commutant form) def correctable Supportable logical operator count g(M) noncomputable def g Logical dimension parameter k=n-\dim(S) def code_k A.3 Pointers to key lemmas and the main theorem Result (paper) Lean lemma/theorem (GitHub link) Nondegeneracy of the symplectic form lemma sym_form_nondegenerate Distance implies erasure correctability lemma dist_implies_correctable Cleaning dimension identity (formal analogue of g(M)+g(M^{c})=2k ) lemma cleaning_dimension_identity Two disjoint correctable sets bound logical dimension ( k\leq|C| ) lemma two_disjoint_correctable_sets _bound_logical_dimension Existence of disjoint sets of prescribed size (helper for the final step) lemma exists_disjoint_finsets_card Quantum Singleton bound theorem quantum_singleton_bound