Introduction
What makes a quantum computation hard to simulate classically, and what makes it costly on a fault-tolerant machine? For qubits, both questions lead to magic, or nonstabilizerness. Clifford circuits acting on stabilizer states are simulated efficiently with the stabilizer formalism [2]. In addition, magic is also the expensive ingredient of fault-tolerant quantum computation. Due to the Eastin–Knill theorem, no quantum error-correcting code admits a universal set of transversal logical gates [3], and fault-tolerant architectures usually implement the missing non-Clifford gates by consuming magic states, which are prepared in dedicated distillation factories [4, 5]. Magic thus measures both the cost of classical simulation by stabilizer methods and the cost of fault-tolerant execution. Free-fermion (matchgate) circuits are classically easy to simulate by different methods [6–9], even though their states generally carry magic. The two classes (matchgates and stabilizer circuits) become universal when combined [6, 10, 11] leading to the conclusion that: computational/circuit classes that are easy separately can be hard together. Fermionic states are also among the main targets of quantum computation. Free-fermion (Gaussian), a special class of fermionic states, include the Hartree–Fock and BCS (mean-field) states that serve as reference states in quantum chemistry and condensed matter, as initial states of quantum algorithms [12, 13], as building blocks of classical methods for interacting fermions [14–16], and as the native states of fermionic quantum processors [7, 17], while the correlated ground states of interacting models are what such algorithms aim to prepare. Where the magic of fermionic states resides, and how much of it must be carried by the entanglement between two parts of a system, is the question we address here.
Consider a pure fermionic state shared by two regions A A A and B B B . Part of its magic is local, since unitaries acting on A A A and B B B separately can remove it. What remains is the nonlocal magic, the magic stored in the entanglement between the regions [18, 19]. A two-party stabilizer state is a collection of Bell pairs and unentangled qubits once each party has chosen a suitable basis, so a state has no nonlocal magic exactly when local unitaries can turn it into such a state. Nonlocal magic therefore measures how far the entanglement of the state is from that of Bell pairs, and it depends only on the Schmidt spectrum across the cut (for fermions, together with the parities of the Schmidt vectors). It is also the part of the magic that unitary operations within the regions cannot remove: it bounds from below the number of non-Clifford (T T T ) gates, and hence of magic states, needed to prepare the state up to local unitaries. For fermions the local unitaries must commute with the fermion parity of each region (parity superselection), and we allow every such unitary, Gaussian or not. Our general results hold for every pure fermionic state of definite parity, free or interacting : an exact result for the largest fidelity with a stabilizer state that parity-even local unitaries can reach, the statement that parity superselection costs at most one bit, and a bound on the nonlocal magic by the capacity of entanglement C E C_E C E , the variance of the entanglement Hamiltonian. Written as qubits, free-fermion states typically carry extensive magic [20, 21]. Across any cut, they are, up to local Gaussian unitaries, a product of m m m entangled (BCS) pairs: one mode in A A A and one in B B B that are both empty or both occupied [Figure 1 (a)]. How much of their magic is nonlocal depends on which local operations are allowed. If the local unitaries are restricted to Gaussian ones, which map free-fermion states to free-fermion states, the nonlocal magic grows linearly with the number m m m of partially entangled pairs [22, 23]. General local unitaries may be non-Gaussian, but for fermions they are still constrained by superselection: they must commute with the fermion parity of each region [Figure 1 (b)].
Figure 1. Free-fermion entanglement conversion into Bell pairs. In panels (a,b) we take m = 4 m=4 m = 4 , k = 2 k=2 k = 2 ; panel (c) is schematic. (a) Across a cut, a free-fermion (Gaussian) state is, in suitably chosen modes, a product of m m m entangled (BCS) pairs q j ∣ 00 ⟩ + p j ∣ 11 ⟩ \sqrt{q_j}|00\rangle+\sqrt{p_j}|11\rangle q j ∣00 ⟩ + p j ∣11 ⟩ [1]. (b) The two parties apply local parity-respecting unitaries, without measuring or communicating, and aim at Bell pairs, the stabilizer states shared by two parties up to local unitaries. Best Gaussian operation (top box): for partially entangled pairs, with occupations bounded away from 0 0 0 , 1 2 \frac{1}{2} 2 1 , and 1 1 1 , the fidelity with every stabilizer state is exponentially small in m m m . Best operation overall (bottom box): after the Gaussian rotation to BCS form, a Fock-state permutation (non-Gaussian) reaches k k k Bell pairs plus empty modes, with fidelity at least of order 1 / m 1/\sqrt{m} 1/ m (Sec. IV). (c) The permutation places the 2 k − 1 2^{k-1} 2 k − 1 largest Schmidt values of each parity (blue: even, red: odd) on the target. Parity forbids mixing the colors, so the fourth-largest value, which is even, is skipped in favor of a smaller odd one; this is the cost of parity superselection. The fidelity is Φ k = 2 − k ( ∑ k e p t λ i ) 2 \Phi_k=2^{-k}\left(\sum_{\mathrm{kept}}\sqrt{\lambda_i}\right)^2 Φ k = 2 − k ( ∑ kept λ i ) 2 , maximized over k k k [Eq. (7) ].
Non-Gaussian local unitaries can only lower the Gaussian-restricted value, and for subsystems of at most two modes it was shown that they cannot [22]. General bounds for qubits show that the nonlocal magic of any state is at most logarithmic in the system size [24], and Aditya, Sierant, and Turkesi found that for free fermions it is logarithmic in the Gaussian-restricted value under mild conditions [25] (Sec. II). Two questions remain: (1) what the nonlocal magic of free fermions actually is, rather than how large it can be, and (2) what changes when the local operations must respect parity superselection but may be non-Gaussian. In this paper, we answer these and some further questions. Our main results are as follows.
Parity-even local unitaries including non-Gaussian unitaries i.e., a Gaussian mode rotation followed by a permutation of Fock states [Figure 1 (b,Figure 1 c)], bring m m m partially entangled pairs to fermionic Bell pairs with fidelity of order 1 / m 1/\sqrt{m} 1/ m , an order that cannot be improved. Due to this, both measures of magic (Sec. II) drop from Θ ( m ) \Theta(m) Θ ( m ) to Θ ( log m ) \Theta(\log m) Θ ( log m ) when the local unitaries are allowed to be parity-preserving non-Gaussian unitaries. For up to three pairs, however, local Gaussian unitaries already minimize the SRE and the distinction between Gaussian and non-Gaussian parity preserving unitaries is not essential.
The slow 1 / m 1/\sqrt{m} 1/ m decay of fidelity with stabilizer state has a statistical origin. We read the Schmidt weights λ i \lambda_i λ i as probabilities of the entanglement energies X i = − log 2 λ i X_i=-\log_2\lambda_i X i = − log 2 λ i , measured in bits. Bell pairs have a flat spectrum, with all X i X_i X i equal, so the best Bell-pair fidelity is essentially the probability that X X X falls into one window. This probability can be small only if X X X is spread over many bits, that is, only if its variance C E / ln 2 2 C_E/\ln^2 2 C E / ln 2 2 is large, where C E C_E C E is the capacity of entanglement. This gives a bound for every state (free or interacting): every pure fermionic state of definite parity can be brought by parity-even local unitaries to a stabilizer state with fidelity at least 0.478 / C E + 5 0.478/\sqrt{C_E+5} 0.478/ C E + 5 . For free fermions, X X X is a sum of independent contributions, one per pair. Each pair adds at most c ∗ ≈ 0.44 c_* \approx0.44 c ∗ ≈ 0.44 to C E C_E C E , so X X X spreads over only about m \sqrt{m} m bits and the fidelity decays only as 1 / m 1/\sqrt{m} 1/ m (Sec. IV).
When the distribution of X X X becomes Gaussian as C E → ∞ C_E\to\infty C E → ∞ , the nonlocal magic follows capacity laws: 1 2 log 2 C E \frac{1}{2}\log_2 C_E 2 1 log 2 C E for the fidelity-based measure and
3 2 log 2 C E \frac{3}{2}\log_2 C_E 2 3 log 2 C E for the SRE. For a critical block of ℓ \ell ℓ sites it should then grow only as log log ℓ \log\log\ell log log ℓ (Sec. IV E).
The plan of the paper is as follows. Section II defines the measures of magic and collects the results on which we build. Sections III–IV develop our results, and § 5 discusses their consequences and limitations. The Supplemental Material (SM) contains the proofs and further remarks.
Measures of nonlocal magic and prior results
In this section, we define the measures of magic used in this paper and collect the results on which we build. In the context of stabilizer states, the magic of a state ∣ ψ ⟩ \lvert\psi\rangle ∣ ψ ⟩ of N N N qubits is quantified by the stabilizer Rényi entropy (SRE) [26],
M α ( ∣ ψ ⟩ ) = 1 1 − α log 2 ( 2 − N ∑ P ⟨ ψ ∣ P ∣ ψ ⟩ 2 α ) (1) M_{\alpha}(\lvert\psi\rangle)=\frac{1}{1-\alpha}\log_2\left(2^{-N}\sum_{P}\langle\psi\lvert P\lvert\psi\rangle^{2\alpha}\right) \tag*{(1)} M α (∣ ψ ⟩) = 1 − α 1 log 2 ( 2 − N P ∑ ⟨ ψ ∣ P ∣ ψ ⟩ 2 α ) (1)
typically using α = 2 \alpha=2 α = 2 , with the sum over all 4 N 4^N 4 N Pauli strings P P P . M α ( ψ ) M_{\alpha}(\psi) M α ( ψ ) vanishes if and only if ∣ ψ ⟩ \lvert\psi\rangle ∣ ψ ⟩ is a stabilizer state. For a state shared by two regions A A A and B B B , part of the magic is local: unitaries acting on A A A and B B B separately can remove it. What remains is the nonlocal magic [18, 19],
M α N L ( ∣ ψ ⟩ ) = min U A ⊗ U B M α ( U A ⊗ U B ∣ ψ ⟩ ) (2) M_{\alpha}^{\mathrm{NL}}(\lvert\psi\rangle)=\min_{U_A\otimes U_B}M_{\alpha}(U_A\otimes U_B\lvert\psi\rangle) \tag*{(2)} M α NL (∣ ψ ⟩) = U A ⊗ U B min M α ( U A ⊗ U B ∣ ψ ⟩) (2)
i.e., the magic stored in the correlations between A A A and B B B . The minimum runs over all local unitaries, so M α N L M_{\alpha}^{\mathrm{NL}} M α NL is a property of ∣ ψ ⟩ \lvert\psi\rangle ∣ ψ ⟩ alone; for a free-fermion state ∣ ψ ⟩ \lvert\psi\rangle ∣ ψ ⟩ the minimizing unitaries need not be Gaussian. A second measure of nonlocal magic is built from fidelities instead of Pauli moments. It is often easier to compute, since for qubits it is given by an explicit formula in the entanglement spectrum (Sec. III), and it has a direct operational meaning. Let STAB denote the set of pure stabilizer states of the N N N qubits. The largest fidelity of ∣ ψ ⟩ \lvert\psi\rangle ∣ ψ ⟩ with a stabilizer state defines the min-relative entropy of magic, D min ( ∣ ψ ⟩ ) = − log 2 max ∣ s ⟩ ∈ S T A B ∣ ⟨ s ∣ ψ ⟩ ∣ 2 D_{\min}(\lvert\psi\rangle)=-\log_2\max_{\lvert s\rangle\in\mathrm{STAB}}\lvert\langle s\lvert\psi\rangle\rvert^2 D m i n (∣ ψ ⟩) = − log 2 max ∣ s ⟩ ∈ STAB ∣⟨ s ∣ ψ ⟩ ∣ 2 , which, like M α M_{\alpha} M α , vanishes exactly on stabilizer states. Maximizing in addition over local unitaries gives the nonlocal stabilizer fidelity Φ q \Phi^{\mathrm{q}} Φ q and the nonlocal min-relative entropy of magic (NMRE) D min N L D_{\min}^{\mathrm{NL}} D m i n NL [24],
Φ q ( ∣ ψ ⟩ ) = max U A ⊗ U B max ∣ s ⟩ ∈ S T A B ∣ ⟨ s ∣ U A ⊗ U B ∣ ψ ⟩ ∣ 2 , \Phi^{\mathrm{q}}(\lvert\psi\rangle)=\max_{U_A\otimes U_B}\max_{\lvert s\rangle\in\mathrm{STAB}}\lvert\langle s\lvert U_A\otimes U_B\lvert\psi\rangle\rvert^2, Φ q (∣ ψ ⟩) = U A ⊗ U B max ∣ s ⟩ ∈ STAB max ∣⟨ s ∣ U A ⊗ U B ∣ ψ ⟩ ∣ 2 ,
D min N L ( ∣ ψ ⟩ ) = − log 2 Φ q ( ∣ ψ ⟩ ) (3) D_{\min}^{\mathrm{NL}}(\lvert\psi\rangle)=-\log_2\Phi^{\mathrm{q}}(\lvert\psi\rangle) \tag*{(3)} D m i n NL (∣ ψ ⟩) = − log 2 Φ q (∣ ψ ⟩) (3)
where the superscript q indicates that U A U_A U A and U B U_B U B are arbitrary unitaries on the qubits of A A A and B B B . Since a two-party stabilizer state is, up to local unitaries, a collection of Bell pairs and unentangled qubits, Φ q \Phi^{\mathrm{q}} Φ q measures how well A A A and B B B can turn their state into Bell pairs by local unitaries alone, without measuring or communicating.
For qubits it depends only on the entanglement spectrum (ES) [24, 27], and it is at most logarithmic for every state: D min N L ≤ log 2 ( ν + 1 ) D_{\min}^{\mathrm{NL}}\le\log_2(\nu+1) D m i n NL ≤ log 2 ( ν + 1 ) and D min N L ≤ log 2 ( 2 S 1 + 3 ) D_{\min}^{\mathrm{NL}}\le\log_2(2S_1+3) D m i n NL ≤ log 2 ( 2 S 1 + 3 ) , with ν \nu ν the number of qubits of the smaller side and S 1 S_1 S 1 the entanglement entropy in bits, and M 2 N L ≤ 4 D min N L M_2^{\mathrm{NL}}\le4D_{\min}^{\mathrm{NL}} M 2 NL ≤ 4 D m i n NL [24]; see also Refs. [30 ] [25, 28, 29]. For fermions, parity superselection restricts U A U_A U A and U B U_B U B in Eq. (2) to parity-even unitaries, which defines the fermionic nonlocal SRE M α f , N L M_{\alpha}^{\mathrm{f,NL}} M α f , NL and its fidelity-based counterpart D min f , N L D_{\min}^{\mathrm{f,NL}} D m i n f , NL (Sec. III).
Gaussian states are fixed by their two-point correlations and evolve efficiently under quadratic Hamiltonians, even at arbitrarily long times [31 ] [7]. Written as qubits, however, they typically carry extensive magic [20, 21]. This does not contradict their simulability, because magic measures the cost of a state for the stabilizer formalism, not classical hardness in general; it does force a Clifford+T T T preparation to use a number of T T T gates at least proportional to N N N [32 ] , although polynomially many suffice to prepare it to any fixed accuracy [33 , 34 ] . Across any cut, a pure Gaussian state is, up to local Gaussian unitaries, a product of M BCS pairs 2 ^2 2 q j ∣ 00 ⟩ + p j ∣ 11 ⟩ \sqrt{q_j}\lvert00\rangle+\sqrt{p_j}\lvert11\rangle q j ∣ 00 ⟩ + p j ∣ 11 ⟩ , each made of one mode in A A A and one in B B B that are both empty or both occupied, with occupation p j p_j p j and q j = 1 − p j q_j=1-p_j q j = 1 − p j [Figure 1 (a)]. Minimizing the SRE over local Gaussian unitaries (denoted M \mathcal{M} M to differentiate from M 2 N L M_2^{\mathrm{NL}} M 2 NL ) alone gives [22, 23] (Sec. S1 of the SM), for every pure Gaussian state [Theorem 1 of Ref. [22]],
M 2 ( p ) = − ∑ j log 2 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] . (4) \mathcal{M}_2(p)=-\sum_j\log_2\left[1-4p_jq_j(1-2p_j)^2\right]. \tag*{(4)} M 2 ( p ) = − j ∑ log 2 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] . (4)
With p j = sin 2 θ j p_j=\sin^2\theta_j p j = sin 2 θ j , each term is − log 2 ( 1 − 1 4 sin 2 4 θ j ) -\log_2(1-\frac{1}{4}\sin^2 4\theta_j) − log 2 ( 1 − 4 1 sin 2 4 θ j ) . It vanishes only for p j ∈ { 0 , 1 2 , 1 } p_j\in\{0,\frac{1}{2},1\} p j ∈ { 0 , 2 1 , 1 } , where the pair is a product state or a Bell pair, and it is largest, log 2 3 2 \log_2\frac{3}{2} log 2 2 3 , at θ j = π / 8 \theta_j=\pi/8 θ j = π /8 (and 3 π / 8 3\pi/8 3 π /8 ). We call the pairs partially entangled if their occupations stay bounded away from these values, p j ∈ [ 1 2 − δ ] ∪ [ 1 2 + δ , 1 − δ ] p_j\in[\frac{1}{2}-\delta]\cup[\frac{1}{2}+\delta,1-\delta] p j ∈ [ 2 1 − δ ] ∪ [ 2 1 + δ , 1 − δ ] for a fixed δ > 0 \delta>0 δ > 0 . For such partially entangled pairs the Gaussian-restricted nonlocal magic is extensive, M 2 = Θ ( m ) \mathcal{M}_2=\Theta(m) M 2 = Θ ( m ) .
The capacity of entanglement. One number bounds the nonlocal magic in this paper. Write the state in Schmidt form, ∣ ψ ⟩ = ∑ i λ i ∣ u i ⟩ A ∣ v i ⟩ B \lvert\psi\rangle=\sum_i\sqrt{\lambda_i}\lvert u_i\rangle_A\lvert v_i\rangle_B ∣ ψ ⟩ = ∑ i λ i ∣ u i ⟩ A ∣ v i ⟩ B ; the Schmidt values λ i \lambda_i λ i , the ES, are the eigenvalues of the reduced density matrix ρ A \rho_A ρ A . Writing ρ A = e − H E \rho_A=e^{-H_E} ρ A = e − H E defines the entanglement Hamiltonian H E = − ln ρ A H_E=-\ln\rho_A H E = − ln ρ A [35 ] : ρ A \rho_A ρ A is a thermal (Gibbs) state of H E H_E H E at unit temperature, whose energy levels, the entanglement energies − ln λ i = X i ln 2 -\ln\lambda_i=X_i\ln2 − ln λ i = X i ln 2 , with X i = − log 2 λ i X_i=-\log_2\lambda_i X i = − log 2 λ i the same energies in bits, are occupied with probabilities λ i \lambda_i λ i . The mean energy is the entanglement entropy, in natural units, and the variance of the entanglement energy is the capacity of entanglement [36 , 37 ] ,
C E = Var H E = ln 2 2 Var X = ∑ i λ i ln 2 λ i − ( ∑ i λ i ln λ i ) 2 (5) C_E = \operatorname{Var} H_E = \ln^2 2 \operatorname{Var} X
= \sum_i \lambda_i \ln^2 \lambda_i - \left(\sum_i \lambda_i \ln\lambda_i\right)^2 \tag*{(5)} C E = Var H E = ln 2 2 Var X = i ∑ λ i ln 2 λ i − ( i ∑ λ i ln λ i ) 2 (5)
the heat capacity of H E H_E H E . It measures the spread of the entanglement energies, not their number: it vanishes for a flat ES, such as that of Bell pairs or of a product state, however large the entanglement entropy, and it is large only when Schmidt values that differ by many orders of magnitude all carry appreciable weight; a dominant value with a faint tail, λ = ( 0.999 , 0.001 ) \lambda=(0.999,0.001) λ = ( 0.999 , 0.001 ) , has C E ≈ 0.05 C_E \approx0.05 C E ≈ 0.05 . For one BCS pair, whose Schmidt values are q q q and p p p , it is p q ln 2 ( q / p ) pq\ln^2(q/p) pq ln 2 ( q / p ) , which vanishes for a product pair and for a Bell pair; for independent pairs the capacities add (Sec. IV B).
Relation to prior work. For qubits, the exact nonlocal stabilizer fidelity is due to Refs. [24, 27], the bound result M 2 ≤ 4 D min M_2 \le4D_{\min} M 2 ≤ 4 D m i n is due to Ref. [38 ] , and its nonlocal form and the universal bounds above to Ref. [24], the dyadic-shell characterization of the nonlocal SRE to Ref. [28], and the doubly logarithmic growth of nonlocal magic at criticality to Refs. [30 ] [28, 29], with the slope 2 3 \frac{2}{3} 3 2 of the SRE derived for the ordered reference state in Ref. [29] and, via the dyadic-shell theorem, for the Ising chain in Ref. [28]. Reference [25] connected nonlocal magic to C E C_E C E , bounded it by 1 2 log 2 ( 1 + C E / ln 2 2 ) + log 2 160 3 \frac{1}{2}\log_2(1+C_E/\ln^2 2)+\log_2\frac{160}{3} 2 1 log 2 ( 1 + C E / ln 2 2 ) + log 2 3 160 , derived the capacity law for spectra that are smooth on the scale of one bit, and, through its eq:B3 , implies the order 1 2 log 2 m \frac{1}{2}\log_2 m 2 1 log 2 m for partially entangled free-fermion pairs of qubits.
In this paper, the new results are as follows — (a) the treatment of parity superselection: the exact fidelity, its cost of at most one bit, and the contrast with particle-number superselection; (b) an exact kernel identity and, from it, a capacity bound that, like that of Ref. [25], holds at every size, but is sharper; (c) explicit constants at every size for the free-fermion order 1 / m 1/\sqrt{m} 1/ m , including superselection, against 1 / m 1/m 1/ m for general states; (d) the capacity laws which we expect critical hopping chains to satisfy, with a lower bound on the asymptotic offset of the SRE; and (e) the sharp crossover at small sizes: extending the two-mode statement of Ref. [22] by a proof that local Gaussian unitaries are optimal among parity-even ones for up to three pairs , and from four pairs on they are not always optimal [Figure 3 (d)] (§ 10 of the SM).
Figure 3. Nonlocal magic of free and interacting fermions. (a, b) Blocks of ℓ ≤ 8192 \ell\le8192 ℓ ≤ 8192 sites of critical XX and Ising chains, L × L L \times L L × L blocks (L ≤ 96 L \le96 L ≤ 96 ) of a 2D Fermi sea, and Haar-random Gaussian states of N ≤ 4096 N \le4096 N ≤ 4096 modes cut in half. (a) Upper bounds on the nonlocal SRE M 2 f , N L \mathcal{M}_{2}^{f,\mathrm{NL}} M 2 f , NL , in bits: the Gaussian-restricted M 2 \mathcal{M}_{2} M 2 of Eq. (4) (solid) grows linearly with the entanglement entropy S 1 S_{1} S 1 , the certificate 4 D min f , N L 4D_{\min}^{f,\mathrm{NL}} 4 D m i n f , NL (dashed; Sec. III B) logarithmically. (b) D min f , N L D_{\min}^{f,\mathrm{NL}} D m i n f , NL against C E C_{E} C E , with the capacity law of Eq. (18) (red dashed, asymptotic) and the capacity bound of Eq. (11) (magenta, every state). (c) Offset δ M = M 2 − 3 2 log 2 C E \delta M=M_{2}-\frac{3}{2}\log_{2}C_{E} δ M = M 2 − 2 3 log 2 C E of reference states, which bound the nonlocal SRE from above (§ 10 A of the SM), for random BCS states (m ≤ 14 m \le14 m ≤ 14 ) and DMRG ground states of t t t –V V V and Hubbard chains near half filling (open markers: V = 0 V=0 V = 0 , U = 0 U=0 U = 0 ). Graded labelings bound M 2 f , N L \mathcal{M}_{2}^{f,\mathrm{NL}} M 2 f , NL , ordered ones only the qubit value M 2 N L \mathcal{M}_{2}^{\mathrm{NL}} M 2 NL . Lines: asymptotic lower bound of Eq. (19) (− 4.91 -4.91 − 4.91 ) and extrapolated reference-state limit i.e., − 2.39 -2.39 − 2.39 of the SRE offset. (d) Identical pairs, p j = 1 4 p_j=\frac{1}{4} p j = 4 1 . Aligned reference states put the largest Schmidt values of each parity on the stabilizer target that attains Φ S N L \Phi_{S}^{\mathrm{NL}} Φ S NL ; from m = 4 m=4 m = 4 on they lie below M 2 \mathcal{M}_{2} M 2 . The gray shaded region shows m ≤ 3 m \le3 m ≤ 3 , where local Gaussian unitaries are optimal (Sec. IV C).
Bell pairs from fermionic entanglement
The exact fidelity under superselection
In the Schmidt form ∣ ψ ⟩ = ∑ i λ i ∣ u i ⟩ A ∣ v i ⟩ B |\psi\rangle=\sum_i\sqrt{\lambda_i}|u_i\rangle_A|v_i\rangle_B ∣ ψ ⟩ = ∑ i λ i ∣ u i ⟩ A ∣ v i ⟩ B of Sec. II, local unitaries rotate the Schmidt vectors but leave the ES λ \lambda λ unchanged, and any two states with the same ES are related by local unitaries, so every nonlocal quantity is a function of λ \lambda λ . In particular, the largest fidelity that local unitaries can reach between two states with spectra λ \lambda λ and μ \mu μ is ( ∑ i λ i μ i ) 2 (\sum_i\sqrt{\lambda_i\mu_i})^2 ( ∑ i λ i μ i ) 2 , with both spectra sorted in decreasing order. The ES of k k k Bell pairs is flat, 2 k 2^k 2 k values equal to 2 − k 2^{-k} 2 − k . For qubits, setting μ i = 2 − k \mu_i=2^{-k} μ i = 2 − k for i ≤ 2 k i\le2^k i ≤ 2 k gives the fidelity of the best conversion into Bell pairs [24, 27],
Φ q = max k 2 − k ( ∑ i ≤ 2 k λ i ) 2 , \Phi^q=\max_k 2^{-k}\left(\sum_{i\le2^k}\sqrt{\lambda_i}\right)^2, Φ q = k max 2 − k i ≤ 2 k ∑ λ i 2 ,
one lays a flat template of 2 k 2^k 2 k equal amplitudes over the 2 k 2^k 2 k largest Schmidt values and picks the best k k k .
Fermions obey parity superselection first established by Wick, Wightman, and Wigner (WWW) [39 ] — physical operations on A A A must commute with its fermion parity P A P_A P A , so local unitaries must be parity-even .5 For a state of definite total parity the Schmidt vectors can be chosen with definite parity, and the ES splits into an even and an odd sector, λ = ( λ ( + ) , λ ( − ) ) \lambda=(\lambda^{(+)},\lambda^{(-)}) λ = ( λ ( + ) , λ ( − ) ) , the graded ES. Parity-even unitaries cannot move a Schmidt value from one sector to the other, and two states of equal total parity are related by them if and only if their graded spectra coincide (Sec. S2 of the SM).
Parity also restricts the flat spectra that are available. Only stabilizer states of definite parity matter as targets: projecting a stabilizer state with ⟨ P ⟩ = 0 \langle P\rangle=0 ⟨ P ⟩ = 0 onto the parity of ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ gives another stabilizer state with twice the fidelity (Sec. S2 A of the SM). These are the Majorana stabilizer states, and neither they nor M α M_\alpha M α depend on the Jordan-Wigner ordering, since reordering mode gates makes fermionic swaps, which are Clifford gates. For a stabilizer state of definite parity, either ± P A \pm P_A ± P A is itself a stabilizer and all 2 k 2^k 2 k levels lie in one sector (a pure split ), or ⟨ P A ⟩ = 0 \langle P_A\rangle=0 ⟨ P A ⟩ = 0 and the levels are shared equally by the two sectors (a balanced split ). A balanced split is the ES of k k k fermionic Bell pairs ( ∣ 00 ⟩ + ∣ 11 ⟩ ) / 2 (|00\rangle+|11\rangle)/\sqrt{2} ( ∣00 ⟩ + ∣11 ⟩) / 2 ; a pure split is that of k k k Bell pairs together with one mode on each side that stores their total parity. Unequal splits, such as three even levels and one odd, which qubits allow, are excluded. Our figure of merit is the nonlocal stabilizer fidelity
Φ S N L = max U A ⊗ U B [ U A , P A ] = [ U B , P B ] = 0 max ∣ s ⟩ ∈ S T A B ∣ ⟨ s ∣ U A ⊗ U B ∣ ψ ⟩ ∣ 2 , (6) \Phi_S^{\mathrm{NL}}=
\max_{\substack{U_A\otimes U_B\\ [U_A,P_A]=[U_B,P_B]=0}}
\max_{|s\rangle\in\mathrm{STAB}}
\left|\langle s|U_A\otimes U_B|\psi\rangle\right|^2, \tag*{(6)} Φ S NL = U A ⊗ U B [ U A , P A ] = [ U B , P B ] = 0 max ∣ s ⟩ ∈ STAB max ∣ ⟨ s ∣ U A ⊗ U B ∣ ψ ⟩ ∣ 2 , (6)
the largest fidelity with a stabilizer state ∣ s ⟩ ∈ S T A B |s\rangle\in\mathrm{STAB} ∣ s ⟩ ∈ STAB that parity-even local unitaries can reach; the subscript S S S denotes this superselection constraint, which distinguishes Φ S N L \Phi_S^{\mathrm{NL}} Φ S NL from the qubit value Φ q \Phi^q Φ q . Let a i ( ± ) a_i^{(\pm)} a i ( ± ) be the square roots of the Schmidt values of the two sectors, sorted in decreasing order and padded with zeros. A balanced split on 2 k 2^k 2 k levels is matched with the 2 k − 1 2^{k-1} 2 k − 1 largest values of each sector, and a pure split with the 2 k 2^k 2 k largest values of one sector. Therefore (Sec. S2 of the SM)
Φ S N L = max { max k ≥ 0 , η = ± 2 − k ( ∑ i = 1 2 k a i ( η ) ) 2 , max k ≥ 1 2 − k ( ∑ i = 1 2 k − 1 ( a i ( + ) + a i ( − ) ) ) 2 } . (7) \Phi_{S}^{\mathrm{NL}}=\max\left\{\max_{\substack{k\geq0,\ \eta=\pm}}2^{-k}\left(\sum_{i=1}^{2^k}a_i^{(\eta)}\right)^2,\ \max_{k\geq1}2^{-k}\left(\sum_{i=1}^{2^{k-1}}\left(a_i^{(+)}+a_i^{(-)}\right)\right)^2\right\}. \tag*{(7)} Φ S NL = max ⎩ ⎨ ⎧ k ≥ 0 , η = ± max 2 − k i = 1 ∑ 2 k a i ( η ) 2 , k ≥ 1 max 2 − k i = 1 ∑ 2 k − 1 ( a i ( + ) + a i ( − ) ) 2 ⎭ ⎬ ⎫ . (7)
The first line contains the pure splits, the second the balanced ones; each term is the fidelity of one flat template, the square of the sum of the amplitudes it covers divided by its number of levels Figure 1 (c). A pure split needs k + 1 k+1 k + 1 modes on each side, one of them for the parity copy, and a balanced split k k k , but these limits never bind: with n n n modes on the smaller side each sector holds at most 2 n − 1 2^{n-1} 2 n − 1 nonzero values, and a term with more slots than that is smaller than the term with half as many. So Φ S N L \Phi_{S}^{\mathrm{NL}} Φ S NL depends only on the graded ES, not on the numbers of modes or on unpaired modes, and below m m m always denotes the number of BCS pairs. We measure the nonlocal magic in bits, D min f , N L = − log 2 Φ S N L D_{\min}^{f,\mathrm{NL}}=-\log_2\Phi_{S}^{\mathrm{NL}} D m i n f , NL = − log 2 Φ S NL . Compared with the qubit value D min N L = − log 2 Φ q D_{\min}^{\mathrm{NL}}=-\log_2\Phi^{q} D m i n NL = − log 2 Φ q , superselection costs at most one bit, Φ S N L ≥ 1 2 Φ q \Phi_{S}^{\mathrm{NL}}\geq\frac{1}{2}\Phi^{q} Φ S NL ≥ 2 1 Φ q : a balanced template with the largest values of each sector contains the best qubit template and is at most twice as large (Sec. S2 of the SM).
A single pair
For one BCS pair (m = 1 m=1 m = 1 ), q ∣ 00 ⟩ + p ∣ 11 ⟩ \sqrt{q}|00\rangle+\sqrt{p}|11\rangle q ∣00 ⟩ + p ∣11 ⟩ with p = sin 2 θ ≤ 1 2 p=\sin^2\theta\leq\frac{1}{2} p = sin 2 θ ≤ 2 1 , each sector holds one Schmidt value, a 1 ( + ) = q a_1^{(+)}=\sqrt{q} a 1 ( + ) = q and a 1 ( − ) = p a_1^{(-)}=\sqrt{p} a 1 ( − ) = p , so only the pure terms with k = 0 k=0 k = 0 and the balanced term with k = 1 k=1 k = 1 matter. Dropping the pure term p ≤ q p\leq q p ≤ q , Eq. (7) gives:
Φ S N L ( m = 1 ) = max { q , 1 2 ( q + p ) 2 } . (8) \Phi_{S}^{\mathrm{NL}}(m=1)=\max\left\{q,\ \frac{1}{2}\left(\sqrt{q}+\sqrt{p}\right)^2\right\}. \tag*{(8)} Φ S NL ( m = 1 ) = max { q , 2 1 ( q + p ) 2 } . (8)
The first entry, q q q , is the fidelity with the empty state ∣ 0 ⟩ A ∣ 0 ⟩ B |0\rangle_A|0\rangle_B ∣0 ⟩ A ∣0 ⟩ B , whereas 1 2 + p q \frac{1}{2}+\sqrt{pq} 2 1 + pq is that with one fermionic Bell pair. The Bell pair is the better target for p ≥ sin 2 π 8 ≈ 0.15 p\geq\sin^2\frac{\pi}{8}\approx0.15 p ≥ sin 2 8 π ≈ 0.15 . At p = sin 2 π 8 p=\sin^2\frac{\pi}{8} p = sin 2 8 π , the two targets tie at fidelity cos 2 π 8 ≈ 0.85 \cos^2\frac{\pi}{8}\approx0.85 cos 2 8 π ≈ 0.85 : the pair with the largest Gaussian-restricted magic is the hardest to convert with D min f , N L ≈ − log 2 ( 0.85 ) ≈ 0.23 D_{\min}^{f,\mathrm{NL}}\approx-\log_2(0.85)\approx0.23 D m i n f , NL ≈ − log 2 ( 0.85 ) ≈ 0.23 . For one mode on each side every parity-even unitary is Gaussian; the two answers separate only for m ≥ 3 m\geq3 m ≥ 3 pairs.
The fidelity also controls the SRE: every pure state obeys M 2 ≤ 4 D min M_2\leq4D_{\min} M 2 ≤ 4 D m i n [38 ] [24], with D min D_{\min} D m i n the − log 2 -\log_2 − log 2 of its largest fidelity with a stabilizer state, so M 2 f , N L ≤ 4 D min f , N L M_2^{f,\mathrm{NL}}\leq4D_{\min}^{f,\mathrm{NL}} M 2 f , NL ≤ 4 D m i n f , NL (Sec. S2 A of the SM). Since one T T T gate raises M 2 M_2 M 2 by at most log 2 4 3 \log_2\frac{4}{3} log 2 3 4 (Theorem 2 of Ref. [32 ] ), any Clifford+T T T circuit that prepares ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ exactly from a stabilizer input, up to parity-even local unitaries that are not counted, contains at least M 2 f , N L / log 2 4 3 ≈ 2.41 M 2 f , N L M_2^{f,\mathrm{NL}}/\log_2\frac{4}{3}\approx2.41M_2^{f,\mathrm{NL}} M 2 f , NL / log 2 3 4 ≈ 2.41 M 2 f , NL T T T gates (Corollary 4 there).
Identical pairs: a simple picture
Consider m m m BCS pairs with the same occupation p < 1 2 p<\frac{1}{2} p < 2 1 . Each Schmidt vector is a Fock configuration s ∈ { 0 , 1 } m s\in\{0,1\}^m s ∈ { 0 , 1 } m of the pairs, with s j = 1 s_j=1 s j = 1 if pair j j j is occupied, and its Schmidt value depends only on the number w = ∣ s ∣ w=|s| w = ∣ s ∣ of occupied pairs, λ w = p w q m − w \lambda_w=p^wq^{m-w} λ w = p w q m − w . Its entanglement energy is X w = − log 2 λ w = − m log 2 q + w log 2 ( q / p ) X_w=-\log_2\lambda_w=-m\log_2q+w\log_2(q/p) X w = − log 2 λ w = − m log 2 q + w log 2 ( q / p ) , so each occupied pair raises it by the same log 2 ( q / p ) \log_2(q/p) log 2 ( q / p ) bits. The ES is therefore a ladder of classes: class w w w contains ( m w ) \binom{m}{w} ( w m ) equal Schmidt values, all with parity ( − 1 ) w (-1)^w ( − 1 ) w , and carries the total weight b w = ( m w ) λ w b_w=\binom{m}{w}\lambda_w b w = ( w m ) λ w , the binomial distribution Figure 2 (a). For large m m m this distribution is close to a Gaussian of mean m p mp m p and variance m p q mpq m pq (de Moivre–Laplace theorem, the simplest local limit theorem), so the most likely class carries weight ≃ 1 / 2 π m p q \simeq1/\sqrt{2\pi mpq} ≃ 1/ 2 π m pq .
Figure 2. How the Bell-pair fidelity arises, against the entanglement energy X = − log 2 λ X=-\log_2\lambda X = − log 2 λ measured from its mean. (a) m = 40 m=40 m = 40 identical pairs, p = 1 4 p=\frac{1}{4} p = 4 1 . Light bars: weights b m b_m b m of the classes of Sec. III C (blue: even w w w , red: odd w w w ). Dark bars: contribution of each class to the optimal fidelity Φ S N L = 0.40 \Phi_{S}^{\mathrm{NL}}=0.40 Φ S NL = 0.40 of Eq. (7) (k = 32 k=32 k = 32 Bell pairs). The even template reaches past the cutoff of Eq. (10) (red dashed), where its levels contribute slightly negatively, a cost of parity superselection. (b) Haar-random Gaussian state of N = 400 N=400 N = 400 modes cut in half (C E = 47 C_E=47 C E = 47 ). Light: the density ρ ( X ) \rho(X) ρ ( X ) , nearly Gaussian with standard deviation σ = C E / ln 2 = 9.9 \sigma=\sqrt{C_E}/\ln2=9.9 σ = C E / ln 2 = 9.9 bits. Dark: ρ ( X ) b ( X − y ∗ ) \rho(X)b(X-y^*) ρ ( X ) b ( X − y ∗ ) , whose area Φ ∗ = 0.22 \Phi_*=0.22 Φ ∗ = 0.22 matches 4 / 2 π ( C E + 5 ) 4/\sqrt{2\pi(C_E+5)} 4/ 2 π ( C E + 5 ) (Sec. IV A). (c) Fidelity for p j = 1 4 p_j=\frac{1}{4} p j = 4 1 : exact value Eq. (7) , the lower bounds of Eqs. (9) and (15) , and the cap ( 13 / 16 ) m / 4 (13/16)^{m/4} ( 13/16 ) m /4 for local Gaussian unitaries [Eq. (16) ].
Local Gaussian unitaries cannot exploit these classes: their fidelity with every stabilizer state decays exponentially in m m m (Sec. IV C). Leaving the pairs as they are, as m m m fermionic Bell pairs with losses that multiply, gives ( 1 2 + p q ) m ≈ 0.933 m (\frac{1}{2}+\sqrt{pq})^m\approx0.933^m ( 2 1 + pq ) m ≈ 0.93 3 m at p = 1 4 p=\frac{1}{4} p = 4 1 [Eq. (8) ].
Non-Gaussian unitaries can do much better, because the levels of a class are all equal, exactly like those of a flat template. A template that keeps 2 k w 2^{k_w} 2 k w levels of class w w w , with k w = ⌊ log 2 ( m w ) ⌋ k_w = \lfloor\log_2 {m \choose w}\rfloor k w = ⌊ log 2 ( w m ) ⌋ , covers more than half of the class, since ( m w ) < 2 k w + 1 {m \choose w} < 2^{k_w+1} ( w m ) < 2 k w + 1 , and a local Fock-state permutation, parity-even but not Gaussian, moves these levels onto the Schmidt vectors of k w k_w k w Bell pairs with their parity stored in one more mode. The fidelity ( 2 k w λ w ) 2 / 2 k w = 2 k w λ w \left(2^{k_w}\sqrt{\lambda_w}\right)^2/2^{k_w}=2^{k_w}\lambda_w ( 2 k w λ w ) 2 / 2 k w = 2 k w λ w is thus more than half the class weight, and the most likely class gives
Φ S N L ≥ max w 2 k w λ w > 1 2 max w b w ≃ 1 2 2 π m p q . (9) \Phi_{\mathrm{S}}^{\mathrm{NL}} \ge\max_w 2^{k_w}\lambda_w > \frac{1}{2}\max_w b_w \simeq\frac{1}{2\sqrt{2\pi m p q}}. \tag*{(9)} Φ S NL ≥ w max 2 k w λ w > 2 1 w max b w ≃ 2 2 π m pq 1 . (9)
At small m m m a single class is worse than leaving the pairs alone, but the product ( 1 2 + p q ) m \left(\frac{1}{2}+\sqrt{p q}\right)^m ( 2 1 + pq ) m decays exponentially in m m m , whereas the single-class fidelity decays only as 1 / m 1/\sqrt{m} 1/ m , like the probability of the most likely class. At p = 1 4 p=\frac{1}{4} p = 4 1 the product is the exact optimum of Eq. (7) up to m = 5 m=5 m = 5 ; the optimum first exceeds it at m = 6 m=6 m = 6 (fidelity 0.72 against 0.66), and from m = 8 m=8 m = 8 on it exceeds the Gaussian cap, so non-Gaussian unitaries then provably beat every local Gaussian unitary [Fig. 2(c)].
Free fermions: from extensive to logarithmic
We first bound the fidelity of every state by its capacity (A) and then show that for m m m BCS pairs the capacity grows at most linearly in m m m (B). We then compare with local Gaussian unitaries (C), translate fidelities into the SRE and T T T -gate counts (D), and give asymptotic laws for many pairs and for critical chains (E). Identical pairs with p = 1 4 p=\frac{1}{4} p = 4 1 (Sec. III C) serve as a running example.
A bound for every state
For a general state the Schmidt values are not degenerate and have no class structure, although nearly equal values can play the role of a class, so we follow their distribution statistically. To do this, regard the entanglement energy in bits, X X X of Sec. II, as a random variable that takes the value X i X_i X i with probability λ i \lambda_i λ i ; its mean is S 1 S_1 S 1 and its variance is C E / ln 2 2 C_E/\ln^2 2 C E / ln 2 2 [Eq. (5)]. If we drop the two restrictions in Eq. (7) , so that the number of levels of the template need not be a power of two and the parity sectors are ignored, the best flat template obeys the exact identity
Φ ∗ ≡ max K 1 K ( ∑ i = 1 K λ i ) 2 = sup y E b ( X − y ) , (10) \Phi_* \equiv\max_K \frac{1}{K}\left(\sum_{i=1}^{K}\sqrt{\lambda_i}\right)^2 = \sup_y \mathbb{E}\,b(X-y), \tag*{(10)} Φ ∗ ≡ K max K 1 ( i = 1 ∑ K λ i ) 2 = y sup E b ( X − y ) , (10)
with the Schmidt values sorted, λ 1 ≥ λ 2 ≥ ⋯ \lambda_1 \ge\lambda_2 \ge\cdots λ 1 ≥ λ 2 ≥ ⋯ , and the kernel b ( z ) = ( 2 ⋅ 2 z / 2 − 2 z ) + b(z)=(2\cdot2^{z/2}-2^z)_+ b ( z ) = ( 2 ⋅ 2 z /2 − 2 z ) + ; it follows by completing a square (Sec. S2 B of the SM). Here Φ ∗ \Phi_* Φ ∗ is the best fidelity that local unitaries reach with a maximally entangled state of any Schmidt rank K K K , which need not be a stabilizer state. In the derivation each kept level earns 2 τ λ i 2\tau\sqrt{\lambda_i} 2 τ λ i and pays τ 2 \tau^2 τ 2 for its slot, and at the optimum τ = 2 − y / 2 \tau=2^{-y/2} τ = 2 − y /2 is the average amplitude of the kept levels. Per unit of probability a level thus contributes b = 2 r − r 2 = 1 − ( 1 − r ) 2 b=2r-r^2=1-(1-r)^2 b = 2 r − r 2 = 1 − ( 1 − r ) 2 , with r = τ / λ i = 2 ( X i − y ) / 2 r=\tau/\sqrt{\lambda_i}=2^{(X_i-y)/2} r = τ / λ i = 2 ( X i − y ) /2 . The kernel is an exchange rate between probability and fidelity, and it depends on a level only through its energy. Hence Φ ∗ \Phi_* Φ ∗ depends on the state only through the distribution of X X X , for identical pairs a histogram of m + 1 m+1 m + 1 classes rather than a list of 2 m 2^m 2 m levels. The kernel is a window. A level at energy X X X contributes λ b ( X − y ) \lambda b(X-y) λb ( X − y ) to the fidelity, which vanishes above the cutoff y + 2 y+2 y + 2 , is largest at X = y X=y X = y , and halves every two bits further below, where the levels have larger Schmidt values but are too few to fill the template. The fidelity is thus collected from a band a few bits wide: the class picture of Sec. III C, smeared over about six neighboring classes for identical pairs [Fig. 2(a)]. Quantitatively, Φ ∗ \Phi_* Φ ∗ is 4 / ln 2 4/\ln2 4/ ln 2 times the peak of the density of X X X smoothed over about 3.2 bits, which raises its variance to ( C E + 5 ) / ln 2 2 (C_E+5)/\ln^2 2 ( C E + 5 ) / ln 2 2 . For a nearly Gaussian X X X , as for a generic free-fermion state [Fig. 2(b)], this gives Φ ∗ ≃ 4 / 2 π ( C E + 5 ) \Phi_* \simeq4/\sqrt{2\pi(C_E+5)} Φ ∗ ≃ 4/ 2 π ( C E + 5 ) , which falls as 1 / C E 1/\sqrt{C_E} 1/ C E . The window has area ∫ b = 4 / ln 2 ≈ 5.8 \int b=4/\ln2\approx5.8 ∫ b = 4/ ln 2 ≈ 5.8 bits, so this estimate needs X X X to be spread over many more bits than that; a distribution narrower than the window fits into it almost entirely, and then Φ ∗ \Phi_* Φ ∗ is of order one.
A density of variance σ 2 \sigma^2 σ 2 has a peak of at least ( 12 σ 2 ) − 1 / 2 (12\sigma^2)^{-1/2} ( 12 σ 2 ) − 1/2 , the value for a uniform density, so Φ ∗ ≥ 2 / 3 ( C E + 5 ) \Phi_* \ge2/\sqrt{3(C_E+5)} Φ ∗ ≥ 2/ 3 ( C E + 5 ) for every state. Each of the two restrictions only removes templates, so Φ S N L ≤ Φ Q ≤ Φ ∗ \Phi_{\mathrm{S}}^{\mathrm{NL}}\le\Phi^{\mathrm{Q}}\le\Phi_* Φ S NL ≤ Φ Q ≤ Φ ∗ . For four identical pairs at p = 1 4 p=\frac{1}{4} p = 4 1 , for example, the best template of any size holds the 11 levels with at most two occupied pairs, with Φ ∗ = 0.811 \Phi_*=0.811 Φ ∗ = 0.811 , whereas the best admissible one holds all 16 levels, four fermionic Bell pairs, with Φ S N L = Φ Q = 0.758 \Phi_{\mathrm{S}}^{\mathrm{NL}}=\Phi^{\mathrm{Q}}=0.758 Φ S NL = Φ Q = 0.758 . The loss is bounded: rounding the optimal size K K K , with 2 k ≤ K < 2 k + 1 2^k\le K<2^{k+1} 2 k ≤ K < 2 k + 1 , down or up to a power of two keeps at least a fraction max ( 2 k / K , K / 2 k + 1 ) ≥ 1 / 2 \max(2^k/K,K/2^{k+1})\ge1/\sqrt{2} max ( 2 k / K , K / 2 k + 1 ) ≥ 1/ 2 of the fidelity, so Φ Q ≥ Φ ∗ / 2 \Phi^{\mathrm{Q}}\ge\Phi_*/\sqrt{2} Φ Q ≥ Φ ∗ / 2 . Restoring the parity and power-of-two parity restrictions costs at most a factor 2 − 1 \sqrt{2}-1 2 − 1 (Proposition S1 of the SM), so every pure fermionic state of definite parity obeys
Φ S N L ≥ 2 ( 2 − 1 ) 3 ( C E + 5 ) ≈ 0.478 C E + 5 . (11) \Phi_{\mathrm{S}}^{\mathrm{NL}} \ge\frac{2(\sqrt{2}-1)}{\sqrt{3(C_E+5)}} \approx\frac{0.478}{\sqrt{C_E+5}}. \tag*{(11)} Φ S NL ≥ 3 ( C E + 5 ) 2 ( 2 − 1 ) ≈ C E + 5 0.478 . (11)
that is, using D min f , N L = − log 2 Φ S N L D_{\min}^{f,\mathrm{NL}}=-\log_2\Phi_{\mathrm{S}}^{\mathrm{NL}} D m i n f , NL = − log 2 Φ S NL , we have D min f , N L ≤ 1 2 log 2 [ 3 4 ( 3 + 2 2 ) ( C E + 5 ) ] D_{\min}^{f,\mathrm{NL}}\le\frac{1}{2}\log_2\left[\frac{3}{4}(3+2\sqrt{2})(C_E+5)\right] D m i n f , NL ≤ 2 1 log 2 [ 4 3 ( 3 + 2 2 ) ( C E + 5 ) ] . The nonlocal magic of every state is thus bounded, at every size, by the thermal fluctuations of its entanglement Hamiltonian.
The capacity of a BCS state and a fidelity of order 1 / m 1/\sqrt{m} 1/ m
Start with a single pair, q ∣ 00 ⟩ + p ∣ 11 ⟩ \sqrt{q}|00\rangle+\sqrt{p}|11\rangle q ∣00 ⟩ + p ∣11 ⟩ . Its reduced density matrix on A A A has two eigenvalues, q q q and p p p , so the entanglement Hamiltonian H E = − ln ρ A H_E=-\ln\rho_A H E = − ln ρ A has two levels, − ln q -\ln q − ln q and − ln p -\ln p − ln p , occupied with probabilities q q q and p p p . The capacity is the variance of this two-valued random variable. A variable that takes two values with probabilities q q q and p p p has variance p q pq pq times the square of their difference, so the capacity of one pair is
c ( p ) = p q ln 2 q p . (12) c(p)=pq\ln^2\frac{q}{p}. \tag*{(12)} c ( p ) = pq ln 2 p q . (12)
It vanishes in two cases. For a product pair, p = 0 p=0 p = 0 , only one level is occupied and nothing fluctuates. For a Bell pair, p = 1 2 p=\frac{1}{2} p = 2 1 , the two levels coincide, the ES is flat, and again nothing fluctuates. In between c ( p ) c(p) c ( p ) is positive, so it has a maximum. Setting the derivative to zero gives ln ( q / p ) = 2 / ( 1 − 2 p ) \ln(q/p)=2/(1-2p) ln ( q / p ) = 2/ ( 1 − 2 p ) , so at the maximum c = 4 p q / ( 1 − 2 p ) 2 c=4pq/(1-2p)^2 c = 4 pq / ( 1 − 2 p ) 2 . The maximum is attained at p ≈ 0.0832 p\approx0.0832 p ≈ 0.0832 and at 1 − 0.0832 1-0.0832 1 − 0.0832 , and it is
c ∗ = max p c ( p ) ≈ 0.439. (13) c_*=\max_p c(p)\approx0.439. \tag*{(13)} c ∗ = p max c ( p ) ≈ 0.439. (13)
Now take m m m pairs. The reduced density matrix of a product of pairs is a product, ρ A = ⨂ j ρ A ( j ) \rho_A=\bigotimes_j\rho_A^{(j)} ρ A = ⨂ j ρ A ( j ) , so the entanglement Hamiltonian is a sum, H E = ∑ j H E ( j ) H_E=\sum_j H_E^{(j)} H E = ∑ j H E ( j ) , and in the state ρ A \rho_A ρ A the terms are independent random variables. The variance of a sum of independent variables is the sum of their variances, exactly as for the number of heads in m m m coin flips, so
C E = ∑ j = 1 m c ( p j ) ≤ c ∗ m . (14) C_E=\sum_{j=1}^{m}c(p_j)\le c_*m. \tag*{(14)} C E = j = 1 ∑ m c ( p j ) ≤ c ∗ m . (14)
For identical pairs with p = 1 / 4 p=1/4 p = 1/4 , C E = 0.226 m C_E=0.226m C E = 0.226 m . In bits the entanglement energy thus has standard deviation C E / ln 2 ≤ c ∗ m / ln 2 ≈ 0.96 m \sqrt{C_E}/\ln2\le\sqrt{c_*m}/\ln2\approx0.96\sqrt{m} C E / ln 2 ≤ c ∗ m / ln 2 ≈ 0.96 m : its mean, the entropy, grows like m m m , but its spread only like m \sqrt{m} m , as for the number of heads in m m m coin flips. For identical pairs, X X X is a constant plus log 2 ( q / p ) \log_2(q/p) log 2 ( q / p ) times the binomial number of occupied pairs. This linear growth is the only place where the free-fermion structure enters. A general state with m m m modes on each side can spread its Schmidt values evenly over m m m bits of entanglement energy, for example λ R ∝ 1 / R \lambda_R\propto1/R λ R ∝ 1/ R for R = 1 , … , 2 m R=1,\ldots,2^m R = 1 , … , 2 m . Then C E C_E C E grows as m 2 m^2 m 2 , and its fidelity can be as small as order 1 / m 1/m 1/ m , which is the worst case: a dyadic-shell bound gives Φ S N L ≥ 3 ( 2 + 2 ) / ( 8 n ) \Phi_S^{\mathrm{NL}}\ge3(2+\sqrt{2})/(8n) Φ S NL ≥ 3 ( 2 + 2 ) / ( 8 n ) , that is, D min f , N L ≤ log 2 n − 0.35 D_{\min}^{f,\mathrm{NL}}\le\log_2 n-0.35 D m i n f , NL ≤ log 2 n − 0.35 , for every state with n ≥ 2 n\ge2 n ≥ 2 modes on the smaller side (Proposition S2 of the SM). Independent pairs cannot spread their energy this widely: each adds at most c ∗ c_* c ∗ to the variance, which halves the logarithmic rate. Inserting C E ≤ c ∗ m = 0.439 m C_E\le c_*m=0.439m C E ≤ c ∗ m = 0.439 m into Eq. (11) , and writing c ∗ m + 5 = c ∗ ( m + 5 / c ∗ ) c_*m+5=c_*(m+5/c_*) c ∗ m + 5 = c ∗ ( m + 5/ c ∗ ) , gives for every BCS state of m m m pairs
Φ S N L ≥ 2 ( 2 − 1 ) 3 ( c ∗ m + 5 ) ≥ 0.72 m + 11.4 , (15) \Phi_S^{\mathrm{NL}}\ge\frac{2(\sqrt{2}-1)}{\sqrt{3(c_*m+5)}}\ge\frac{0.72}{\sqrt{m+11.4}}, \tag*{(15)} Φ S NL ≥ 3 ( c ∗ m + 5 ) 2 ( 2 − 1 ) ≥ m + 11.4 0.72 , (15)
The fidelity thus decays no faster than 1 / m 1/\sqrt{m} 1/ m . For partially entangled pairs it also decays no slower. The kernel vanishes above y + 2 y+2 y + 2 and halves every two bits below y y y , so Φ S N L ≤ Φ ∗ \Phi_S^{\mathrm{NL}}\le\Phi_* Φ S NL ≤ Φ ∗ is at most a few times the largest probability that X X X falls into an interval a few bits wide, and by the Berry–Esseen theorem this probability is of order 1 / C E 1/\sqrt{C_E} 1/ C E . Each pair adds at least c δ = min { c ( δ ) , c ( 1 2 − δ ) } > 0 c_\delta=\min\{c(\delta),c(\frac{1}{2}-\delta)\}>0 c δ = min { c ( δ ) , c ( 2 1 − δ )} > 0 to C E C_E C E , so Φ S N L ≤ 2 a δ / c δ m \Phi_S^{\mathrm{NL}}\le2^{a_\delta}/\sqrt{c_\delta m} Φ S NL ≤ 2 a δ / c δ m with a constant a δ a_\delta a δ that depends only on δ \delta δ (Proposition S3 of the SM); for p j = 1 4 p_j=\frac{1}{4} p j = 4 1 this reads Φ S N L ≤ 10.4 / m \Phi_S^{\mathrm{NL}}\le10.4/\sqrt{m} Φ S NL ≤ 10.4/ m . Hence Φ S N L = Θ ( m − 1 / 2 ) \Phi_S^{\mathrm{NL}}=\Theta(m^{-1/2}) Φ S NL = Θ ( m − 1/2 ) and D min N L = 1 2 log 2 m + O ( 1 ) D_{\min}^{\mathrm{NL}}=\frac{1}{2}\log_2m+O(1) D m i n NL = 2 1 log 2 m + O ( 1 ) . Pairs that are nearly Bell pairs or nearly empty must be excluded: they add almost nothing to C E C_E C E and keep the fidelity close to one.
Local Gaussian unitaries do exponentially worse
Now compare with local Gaussian unitaries. No local unitary changes the graded ES; what non-Gaussian unitaries add is the freedom to place the Schmidt values on the Fock states of a flat template, class by class (Sec. III C). Local Gaussian unitaries lack this freedom, and the precise statement uses magic. Every state in the local-Gaussian orbit of ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ has SRE at least M 2 ( p ) \mathcal{M}_2(p) M 2 ( p ) , the Gaussian minimum of Eq. (4) , by Theorem 1 of Ref. [22]; and a pure state with SRE M 2 \mathcal{M}_2 M 2 has fidelity at most 2 − M 2 / 4 2^{-\mathcal{M}_2/4} 2 − M 2 /4 with every stabilizer state, since M 2 ≤ 4 D min \mathcal{M}_2\le4D_{\min} M 2 ≤ 4 D m i n (Sec. III). So local Gaussian unitaries reach a fidelity of at most 2 − M 2 ( p ) / 4 2^{-\mathcal{M}_2(p)/4} 2 − M 2 ( p ) /4 . This half of the separation rests entirely on that theorem, which covers every pure Gaussian state, every cut, and all local Gaussian unitaries, including those that change the particle number.
Since M 2 ( p ) \mathcal{M}_2(p) M 2 ( p ) is a sum of m m m nonnegative terms, one per pair, it grows linearly in m m m as long as each term is bounded away from zero. For partially entangled pairs (Sec. II) the smallest term belongs to the allowed pairs closest to Bell pairs, p j = 1 2 ± δ p_j=\frac{1}{2}\pm\delta p j = 2 1 ± δ (with u = 1 − 2 p j u=1-2p_j u = 1 − 2 p j the term is − log 2 ( 1 − u 2 + u 4 ) -\log_2(1-u^2+u^4) − log 2 ( 1 − u 2 + u 4 ) , and u 2 − u 4 u^2-u^4 u 2 − u 4 is smaller at u = 2 δ u=2\delta u = 2 δ than at u = 1 − 2 δ u=1-2\delta u = 1 − 2 δ for δ ≤ 1 4 \delta\le\frac{1}{4} δ ≤ 4 1 ), so that
2 − M 2 ( p ) / 4 ≤ [ 1 − 4 δ 2 ( 1 − 4 δ 2 ) ] m / 4 , (16) 2^{-\mathcal{M}_2(p)/4}\le\left[1-4\delta^2(1-4\delta^2)\right]^{m/4}, \tag*{(16)} 2 − M 2 ( p ) /4 ≤ [ 1 − 4 δ 2 ( 1 − 4 δ 2 ) ] m /4 , (16)
which decays exponentially in m m m . For the running example each pair contributes M 2 = log 2 16 13 = 0.30 \mathcal{M}_2=\log_2\frac{16}{13}=0.30 M 2 = log 2 13 16 = 0.30 bits, so local Gaussian unitaries reach a fidelity of at most ( 13 / 16 ) m / 4 ≈ 0.949 m (13/16)^{m/4}\approx0.949^m ( 13/16 ) m /4 ≈ 0.94 9 m , which is 9.6 × 10 − 10 9.6\times10^{-10} 9.6 × 1 0 − 10 at m = 400 m=400 m = 400 . General parity-even unitaries do far better at the same m m m : inserting C E = 400 ⋅ 3 16 ln 2 3 = 90.5 C_E=400\cdot\frac{3}{16}\ln^2 3=90.5 C E = 400 ⋅ 16 3 ln 2 3 = 90.5 into Eq. (11) guarantees a fidelity at most 0.049 0.049 0.049 , and the exact optimum of Eq. (7) , evaluated on the 401 weight classes, is 0.152 0.152 0.152 .
For a few pairs the comparison is different (Figure 3 (d)). For up to three pairs no parity-even local unitary brings the SRE below M 2 ( p ) \mathcal{M}_2(p) M 2 ( p ) , so local Gaussian unitaries are optimal (Proposition S4 of the SM, whose proof combines a parity grading of the Pauli projector with three results of Ref. [28]). Superselection is essential here: among all qubit unitaries the statement already fails at three pairs. From four pairs on local Gaussian unitaries are not always optimal: for four equal pairs, a diagonal Clifford gate followed by the exchange of ∣ 1000 ⟩ \lvert1000\rangle ∣ 1000 ⟩ and ∣ 0111 ⟩ \lvert0111\rangle ∣ 0111 ⟩ on each side lowers the SRE below M 2 \mathcal{M}_{2} M 2 for every p ∉ { 0 , 1 2 , 1 } p \notin\{0,\frac{1}{2},1\} p ∈ / { 0 , 2 1 , 1 } (Proposition S5 of the SM), and for p j = 1 4 p_j=\frac{1}{4} p j = 4 1 the numerical gap of Figure 3 (d) grows with m m m .
Consequences for magic and T T T gates
The SRE is controlled by the fidelity through M 2 f , N L ≤ 4 D min f , N L \mathcal{M}_{2}^{f,\mathrm{NL}} \le4D_{\min}^{f,\mathrm{NL}} M 2 f , NL ≤ 4 D m i n f , NL (Sec. III). Inserting Eq. (11) with C E ≤ c ∗ m C_{E}\le c_*m C E ≤ c ∗ m for the non-Gaussian side, and Eq. (16) for the Gaussian side, gives
M 2 f , N L ≤ 4 D min f , N L ≤ 2 log 2 ( 1.921 m + 22 ) , \mathcal{M}_{2}^{f,\mathrm{NL}} \le4D_{\min}^{f,\mathrm{NL}} \le2\log_{2}(1.921m+22), M 2 f , NL ≤ 4 D m i n f , NL ≤ 2 log 2 ( 1.921 m + 22 ) ,
M 2 ( p ) ≥ − m log 2 [ 1 − 4 δ 2 ( 1 − 4 δ 2 ) ] ≥ 4 ln 2 δ 2 ( 1 − 4 δ 2 ) m . (17) \mathcal{M}_{2}(p)\ge-m\log_{2}\left[1-4\delta^{2}(1-4\delta^{2})\right] \ge\frac{4}{\ln2}\delta^{2}(1-4\delta^{2})m. \tag*{(17)} M 2 ( p ) ≥ − m log 2 [ 1 − 4 δ 2 ( 1 − 4 δ 2 ) ] ≥ ln 2 4 δ 2 ( 1 − 4 δ 2 ) m . (17)
The first line is Eq. (11) with C E ≤ c ∗ m C_{E}\le c_*m C E ≤ c ∗ m , multiplied by four, and holds for every BCS state; the second is the logarithm of Eq. (16) . Allowing non-Gaussian local unitaries thus lowers the nonlocal SRE from linear to logarithmic in m m m , and with it the lower bound M 2 / log 2 4 3 \mathcal{M}_{2}/\log_{2}\frac{4}{3} M 2 / log 2 3 4 on the number of T T T gates needed beyond local processing. The logarithm cannot be improved: the dyadic-shell bound of Ref. [28] gives M 2 N L ≥ 3 D min N L − 10.36 \mathcal{M}_{2}^{\mathrm{NL}}\ge3D_{\min}^{\mathrm{NL}}-10.36 M 2 NL ≥ 3 D m i n NL − 10.36 , so with the converse of Sec. IV B and the one-bit cost of superselection (Sec. III A) the nonlocal SRE of partially entangled pairs is Θ ( log m ) \Theta(\log m) Θ ( log m ) , although this lower bound becomes nontrivial only at very large m m m . For the running example (p = 1 / 4 p=1/4 p = 1/4 ) at m = 400 m=400 m = 400 , M 2 = 119.8 \mathcal{M}_{2}=119.8 M 2 = 119.8 bits, so at least 119.8 / log 2 ( 4 / 3 ) ≈ 289 119.8/\log_{2}(4/3)\approx289 119.8/ log 2 ( 4/3 ) ≈ 289 T T T gates are needed if only Gaussian processing is allowed; with general parity-even processing M 2 f , N L ≤ 19.3 \mathcal{M}_{2}^{f,\mathrm{NL}}\le19.3 M 2 f , NL ≤ 19.3 bits by Eq. (17) , so the certified lower bound drops to at most 47 T T T gates, and the exact Φ S N L = 0.152 \Phi_{S}^{\mathrm{NL}}=0.152 Φ S NL = 0.152 gives M 2 f , N L ≤ 4 D min f , N L = 10.9 \mathcal{M}_{2}^{f,\mathrm{NL}}\le4D_{\min}^{f,\mathrm{NL}}=10.9 M 2 f , NL ≤ 4 D m i n f , NL = 10.9 bits and at most 27.
Capacity laws and critical systems
The capacity bound is a lower bound. For many pairs the asymptotic value of the fidelity itself can be estimated. Suppose that C E → ∞ C_{E}\to\infty C E → ∞ while the third absolute moments of the pair energies stay comparable to their variances, that the energy gaps log 2 ( q j / p j ) \log_{2}(q_j/p_j) log 2 ( q j / p j ) do not cluster near multiples of a common spacing that stays bounded away from zero, which identical pairs violate, and that the parity is neither frozen, as it is when almost all pairs are empty, nor locked to the energy on such a lattice, as it is on the ladder of identical pairs (a spacing that shrinks, as for the critical chains below, is allowed but slows the convergence). Then we expect that the entanglement energy obeys a parity-resolved local limit theorem: even on the scale of one bit, which the kernel of Eq. (10) probes, X X X is Gaussian, and every energy window is shared equally by the two parity sectors. Equation Eq. (10) then gives Φ ∗ = 4 / 2 π C E ( 1 + o ( 1 ) ) \Phi_* = 4/\sqrt{2\pi C_E}(1+o(1)) Φ ∗ = 4/ 2 π C E ( 1 + o ( 1 )) , and if balanced stabilizer templates attain it, as we expect when the two sectors share every window, then as C E → ∞ C_E\to\infty C E → ∞
D min f , N L = 1 2 log 2 ( 2 π C E ) − 2 + o ( 1 ) , (18) D_{\min}^{f,\mathrm{NL}}=\frac{1}{2}\log_{2}(2\pi C_E)-2+o(1), \tag*{(18)} D m i n f , NL = 2 1 log 2 ( 2 π C E ) − 2 + o ( 1 ) , (18)
and the qubit value D min N L D_{\min}^{\mathrm{NL}} D m i n NL obeys the same law. Since no stabilizer fidelity exceeds Φ ∗ \Phi_* Φ ∗ , the right-hand side is a lower bound on both; the equality, derived for smooth qubit spectra in Ref. [25], we support numerically [Figure 3 (b)] but do not prove. Even the running example (p = 1 / 4 p=1/4 p = 1/4 ), which violates the second and third conditions, is close: at m = 400 m = 400 m = 400 the law gives with C E = 400 ( 3 / 16 ) ln 2 ( 3 ) ≈ 90.52 C_E = 400(3/16)\ln^2(3) \approx90.52 C E = 400 ( 3/16 ) ln 2 ( 3 ) ≈ 90.52 , Φ ∗ = 4 / 2 π ⋅ ( 90.52 ) ≈ 0.1677 \Phi_* = 4/\sqrt{2\pi}\cdot(90.52) \approx0.1677 Φ ∗ = 4/ 2 π ⋅ ( 90.52 ) ≈ 0.1677 , against the exact 0.152.
For the SRE the coefficient of log 2 C E \log_2 C_E log 2 C E is three times larger. The fidelity probes the peak of the density ρ \rho ρ of X X X , whose height is of order C E − 1 / 2 C_E^{-1/2} C E − 1/2 , whereas the nonlocal SRE is fixed, up to a bounded constant, by the dyadic rank shells of the ES [28], which probe ∫ ρ 4 ∝ C E − 3 / 2 \int\rho^4 \propto C_E^{-3/2} ∫ ρ 4 ∝ C E − 3/2 : multiplying C E C_E C E by 4 costs one bit of fidelity-based magic and three bits of SRE, consistent with M 2 f , N L ≤ 4 D min f , N L M_{2}^{f,\mathrm{NL}} \le4D_{\min}^{f,\mathrm{NL}} M 2 f , NL ≤ 4 D m i n f , NL . Quantitatively, Ref. [28] places M 2 N L M_{2}^{\mathrm{NL}} M 2 NL between − log 2 Q s h − log 2 C s h -\log_2 Q_{\mathrm{sh}}-\log_2 C_{\mathrm{sh}} − log 2 Q sh − log 2 C sh and − log 2 Q s h + 4 -\log_2 Q_{\mathrm{sh}}+4 − log 2 Q sh + 4 , with C s h ≈ 654.5 C_{\mathrm{sh}}\approx654.5 C sh ≈ 654.5 and the shell sum Q s h = λ 1 4 + ∑ k ≥ 1 ( 2 k − 1 λ 2 k − 1 + 1 ) 4 Q_{\mathrm{sh}}=\lambda_1^4+\sum_{k\ge1}(2^{k-1}\lambda_{2^{k-1}+1})^4 Q sh = λ 1 4 + ∑ k ≥ 1 ( 2 k − 1 λ 2 k − 1 + 1 ) 4 . For Gaussian X X X , Q s h ≃ ∫ ρ 4 d x / ln 4 2 Q_{\mathrm{sh}}\simeq\int\rho^4\,dx/\ln^4 2 Q sh ≃ ∫ ρ 4 d x / ln 4 2 and − log 2 Q s h = 3 2 log 2 C E + c G + o ( 1 ) -\log_2 Q_{\mathrm{sh}}=\frac{3}{2}\log_2 C_E+c_G+o(1) − log 2 Q sh = 2 3 log 2 C E + c G + o ( 1 ) , with c G = log 2 ( 4 2 π 3 / 2 ln 2 ) ≈ 4.45 c_G=\log_2(4\sqrt{2}\pi^{3/2}\ln2)\approx4.45 c G = log 2 ( 4 2 π 3/2 ln 2 ) ≈ 4.45 . Hence, as C E → ∞ C_E\to\infty C E → ∞ ,
M 2 f , N L ≥ M 2 N L = 3 2 log 2 C E + O ( 1 ) , M_{2}^{f,\mathrm{NL}}\ge M_{2}^{\mathrm{NL}}=\frac{3}{2}\log_2 C_E+O(1), M 2 f , NL ≥ M 2 NL = 2 3 log 2 C E + O ( 1 ) ,
M 2 N L ≥ 3 2 log 2 C E − 4.91 + o ( 1 ) , (19) M_{2}^{\mathrm{NL}}\ge\frac{3}{2}\log_2 C_E-4.91+o(1), \tag*{(19)} M 2 NL ≥ 2 3 log 2 C E − 4.91 + o ( 1 ) , (19)
where − 4.91 = c G − log 2 C s h -4.91=c_G-\log_2 C_{\mathrm{sh}} − 4.91 = c G − log 2 C sh . For M 2 f , N L M_{2}^{f,\mathrm{NL}} M 2 f , NL , Ref. [28] gives only the lower bound, and Eq. (11) only M 2 f , N L ≤ 4 D min f , N L ≤ 2 log 2 C E + O ( 1 ) M_{2}^{f,\mathrm{NL}}\le4D_{\min}^{f,\mathrm{NL}}\le2\log_2 C_E+O(1) M 2 f , NL ≤ 4 D m i n f , NL ≤ 2 log 2 C E + O ( 1 ) . The graded reference states [Figure 3 (c)], which parity-even unitaries reach, do better: for Gaussian X X X we estimate their offset to tend to c G − log 2 K ≈ − 2.39 c_G-\log_2 K\approx-2.39 c G − log 2 K ≈ − 2.39 , with K ≈ 114 K\approx114 K ≈ 114 the numerically extrapolated growth rate of a lattice sum. So M 2 f , N L M_{2}^{f,\mathrm{NL}} M 2 f , NL very likely obeys the same law with its own offset, at accessible sizes all offsets in Figure 3 (c) still lie above − 2.39 -2.39 − 2.39 .
Both laws use the pair structure only through the local limit theorem, so they apply to any family of fermionic states that obeys it. We expect it to hold for blocks of ℓ \ell ℓ sites of critical hopping chains whose Fermi sea is a single interval of momenta, such as the XX chain. There C E = 1 3 ln ℓ ( 1 + o ( 1 ) ) C_E=\frac{1}{3}\ln\ell(1+o(1)) C E = 3 1 ln ℓ ( 1 + o ( 1 )) , so the nonlocal magic should grow only doubly logarithmically, D min f , N L = 1 2 log 2 ( 2 π 3 ln ℓ ) − 2 + o ( 1 ) D_{\min}^{f,\mathrm{NL}}=\frac{1}{2}\log_2\left(\frac{2\pi}{3}\ln\ell\right)-2+o(1) D m i n f , NL = 2 1 log 2 ( 3 2 π ln ℓ ) − 2 + o ( 1 ) , as found for qubit chains [30 ] [28, 29]. This regime is reached very slowly: for ℓ ≤ 8192 \ell\le8192 ℓ ≤ 8192 the XX chain has C E ≤ 3.53 C_E\le3.53 C E ≤ 3.53 , and at such sizes the certificate $4D_{\min}^{f,\mathrm{NL}}$ still exceeds M 2 \mathcal{M}_2 M 2 , so for critical chains the separation of Eq. (17) is an asymptotic statement. It is visible at accessible sizes for states with large C E C_E C E : the 2D Fermi sea and Haar-random Gaussian states [40 ] approach Eq. (18) to within a few hundredths of a bit at $C_E\approx100 [{xref}Fig. 3(b)], and at the largest sizes ($L=96$, $C_E=218$; $N=4096$, $C_E=479$) $4D_{\min}^{f,\mathrm{NL}}$ lies below $\mathcal{M}_2$ by factors of 11.5 and 27 [{xref}Fig. 3(a)`].
Interacting fermions. The exact fidelity of Sec. III and the capacity bound of Sec. IV A hold for every state, including interacting fermionic states. We evaluated both of these for DMRG ground states of t t t -V V V chains (up to 257 sites) and Hubbard chains (up to 64 sites), next to half filling (methods in Sec. S1 of the SM). For the SRE, graded reference states of Eq. (32) , and the staircase bound of Ref. [28], Eq. (21) of the SM, bounds M 2 N L ≤ M 2 f , N L M_{2}^{\mathrm{NL}}\le M_{2}^{f,\mathrm{NL}} M 2 NL ≤ M 2 f , NL from below. For strong repulsion (V ≥ 2 t V\ge2t V ≥ 2 t , U ≥ 4 t U\ge4t U ≥ 4 t ) these lower and upper bounds meet within 0.06 bits, so in that case the nonlocal SRE of an interacting chain is essentially known exactly. At these sizes C E ≤ 2.04 C_E\le2.04 C E ≤ 2.04 , and the chains look like free fermions; but so do non-integrable chains [41 , 42 ] and generic random spectra, so this agreement does not yet test the capacity laws, which need larger capacities.
Discussion and implications
In this paper, we probe the question of where the magic of a free-fermion state resides. If we do a bipartition of the state into two regions, and if each region may only be rotated by Gaussian (free-fermion) unitaries, the magic that cannot be removed (nonlocal magic) grows in proportion to the number m m m of partially entangled pairs across the cut/bipartition. If in each region, we can apply any unitary that respects fermion parity, a Gaussian rotation followed by a reshuffling of Fock states (non-Gaussian) brings the state to Bell pairs with a fidelity that decays only as 1 / m 1/\sqrt{m} 1/ m , and the magic left across the cut grows only like log m \log m log m . The state itself is a free-fermion state throughout; the local unitaries only reveal which part of its magic is local and nonlocal. Almost all the magic of a free-fermion (Gaussian) state is therefore local : it is stored within the regions, not in the correlations between them. The reason is straightforward. Bell pairs have a flat entanglement spectrum, so a state can be far from Bell pairs only if its Schmidt weights are spread over many orders of magnitude, that is, only if its capacity of entanglement C E C_E C E , related to the variance of Schmidt weights is large.
Two caveats apply to our results. First, we do not estimate the cost of the local unitaries, although the Fock-state permutations are themselves non-Clifford circuits. Second, magic measures the cost of a state for stabilizer (Clifford) simulation, not its classical complexity in general, so our results do not affect the efficient classical simulation of free fermions.
Fermionic quantum computation. Fermionic quantum processors [7, 17] manipulate fermionic modes directly, and on them parity superselection holds exactly. For operations within the parts of a register the parity-even nonlocal magic is therefore the relevant quantity. In case the hardware also conserves the particle number, the stricter number-conserving value applies, and for identical pairs its fidelity-based version stays extensive. Free-fermion gates supply the magic that makes the combination of matchgate and Clifford circuits universal [6, 10]. Our results quantify how much of that magic must be shared between two parts of a register. Across any partition, the lower bound on the number of T T T gates that processing within the parts cannot replace grows linearly with the number m m m of partially entangled pairs if that processing must be Gaussian, but only as log m \log m log m if it may be any parity-even unitary including non-Gaussian unitaries.
Hartree–Fock and mean-field states. Slater determinants (Hartree–Fock states) and BCS states are free-fermion states, so all our results apply to the mean-field reference states of quantum chemistry and condensed matter. These are also standard initial states of quantum algorithms [12], and their magic has recently been examined [43 ] . Our bounds need only the number m m m of partially occupied orbital pairs across the cut, not the occupations themselves. Inserting C E ≤ c ∗ m C_E \leq c_*m C E ≤ c ∗ m into Eq. (11) gives D min f , N L ≤ 1 2 log 2 ( 1.921 m + 22 ) D_{\min}^{f,\mathrm{NL}} \leq\frac{1}{2}\log_2(1.921m+22) D m i n f , NL ≤ 2 1 log 2 ( 1.921 m + 22 ) for every such state [Eq. (17)]. For Hartree–Fock and BCS states, the exact Φ S N L \Phi_{\mathrm{S}}^{\mathrm{NL}} Φ S NL , C E C_E C E , and M 2 \mathcal{M}_2 M 2 also follow from the two-point correlations within region A A A alone, ⟨ c i † c j ⟩ \langle c_i^\dagger c_j\rangle ⟨ c i † c j ⟩ and ⟨ c i c j ⟩ \langle c_i c_j\rangle ⟨ c i c j ⟩ with i , j ∈ A i,j\in A i , j ∈ A [44 ] , which can be estimated efficiently from measurements [45 ] .
Classical simulation of interacting fermions. Many classical methods for interacting fermions combine a tensor network or a sampling scheme with transformations that are cheap to apply: orbital rotations, which are Gaussian unitaries, in orbital-optimized DMRG [15] and Gaussian matrix product states [14]; superpositions of a few Gaussian states for quantum impurity models [16]; simulators of circuits dominated by free-fermion operations [46 , 47 ] ; and Clifford circuits in Clifford-augmented matrix product states (CAMPS) [48 , 49 ] and stabilizer-rank methods [50 ] . Our results explains what such transformations can achieve across a single cut. Local Gaussian rotations cannot bring the SRE below the Gaussian-restricted nonlocal SRE M 2 \mathcal{M}_2 M 2 , whereas local Fock-state permutations reduce it to O ( log ( C E + 5 ) ) O(\log(C_E+5)) O ( log ( C E + 5 )) . This does not make tensor networks cheaper: no local unitary changes the Schmidt spectrum, so the bond dimension needed at the cut stays the same. It can help only methods whose cost grows with magic, such as stabilizer-rank methods. Moreover, the permutations are not Clifford circuits, so pushing Pauli observables through them, as Clifford-augmented methods do, is in general expensive, and efficient circuits for them are not known in general. Whether permutation-augmented methods can be made efficient is open. For interacting fermions, Eq. (11) still bounds the nonlocal magic of every state by its C E C_E C E , from critical chains [51 ] to Sachdev–Ye–Kitaev models [52 , 53 ] .
Other connections. The capacity C E C_E C E follows from the entanglement energies, which entanglement-Hamiltonian tomography measures in quantum simulators [54 , 55 ] . Equation (11) is thus, in principle, an experimentally accessible bound on the nonlocal magic of any pure fermionic state, including the fermionic matter of lattice gauge theories, whose magic and non-Gaussianity are now being studied [56 , 57 , 58 ] . In holography, nonlocal magic has been tied to gravitational backreaction [19], and C E C_E C E has been used as its proxy [59 ] . The qubit form of Eq. (11) makes this link quantitative: the fidelity-based nonlocal magic is at most 1 2 log 2 C E \frac{1}{2}\log_2 C_E 2 1 log 2 C E plus a constant, so a capacity of order 1 / G N 1/G_N 1/ G N , with G N G_N G N Newton’s constant, allows only O ( log ( 1 / G N ) ) O(\log(1/G_N)) O ( log ( 1/ G N )) bits of nonlocal magic.
Outlook. The capacity of entanglement (a single number) obtained from the variance of entanglement energies bounds the nonlocal magic of every state. Whatever magic a state carries, the part that must be shared between two regions grows at most logarithmically with the fluctuation C E C_E C E of its entanglement Hamiltonian. For free fermions this fluctuation grows at most linearly in the number of entangled pairs. Three questions remain open. (1) Interacting matter. The bound of Eq. (11) holds for every state, but whether the capacity laws, including the slope 2 3 \frac{2}{3} 3 2 of the SRE law, persist for interacting systems at large C E C_E C E is not known. (2) The cost of the local operations. The Fock-state permutations that reduce the nonlocal magic are themselves neither Gaussian nor Clifford. A complete accounting must therefore weigh the magic that these permutations shift from the cut into the regions against their own cost: their non-Gaussianity, for which computable measures and experimentally accessible bounds were recently developed [60 ] , and the T T T gates they need within each region. Efficient parity-respecting circuits for them are still to be found, in particular for pairs with unequal occupations. (3) Particle-number conservation. Without a phase reference [61 ] (a shared reservoir, such as a superconductor, with which both region A A A and B B B can coherently exchange particles), the local unitaries must also conserve the particle number in each region, not only its parity. For identical pairs the fidelity-based nonlocal magic then stays extensive, at least for stabilizer targets of definite particle number (Proposition S6 of the SM). However, the behavior of SRE, and both measures (fidelity-based and SRE) for unequal pairs, is still open to our knowledge. Two constants also remain to be pinned down: the best prefactor in the capacity bound, Eq. (11) , and the offset of the SRE law, Eq. (19) , which we bound from below by − 4.91 -4.91 − 4.91 and, through reference states, estimate to be at most about − 2.4 -2.4 − 2.4 . We leave these for future work.
Note added. While this work was being completed, a paper was posted on arXiv [62 ] studying the Gaussian-restricted nonlocal SRE M 2 \mathcal{M}_2 M 2 of Eq. (4) in higher-dimensional free-fermion systems. Both our work and theirs highlight the capacity of entanglement, but under different sets of local operations. Unlike Ref. [62], we allow non-Gaussian, parity-respecting local unitaries and show that the nonlocal magic of every pure fermionic state grows at most logarithmically with C E C_E C E [Eq. (11)]. Furthermore, Ref. [62] does not consider parity superselection, stabilizer fidelity, or the optimality of local Gaussian unitaries for up to three pairs, all of which are key components of the present work. In addition, another paper [25] posted earlier in the same month, connected nonlocal magic to the capacity of entanglement for qubits and used this link to study nonlocal magic in chaotic and integrable dynamics, with consequences for entanglement embezzlement. Its saturation at a value that grows logarithmically with the subsystem size in integrable dynamics is consistent with the logarithmic laws found here. Our capacity bound, Eq. (11) , is tighter than theirs to our understanding, but Ref. [25] considers neither parity superselection nor the optimality of local Gaussian unitaries at small sizes.
Acknowledgments
We thank Norman Hogan and Heba A. Labib for comments on the draft. R.G.J., D.L., A.F.K., and B.N.B. were supported by the U.S. Department of Energy, Advanced Scientific Computing Research, under contract number DE-SC0025384. The authors acknowledge assistance of Anthropic’s Claude Fable 5.0 and Opus 5.5 in preparing this manuscript especially in the kernel approach of Sec. IV and OpenAI’s Astra for proving the Gaussian optimality for m = 3 m = 3 m = 3 pairs. The authors have checked and assume complete responsibility.
Data availability
The code and data that support the findings of this submission can be obtained from the corresponding author after publication. We include Mathematica notebook with this submission to verify some results in this paper.
References
Supplemental Material for “Nonlocal magic of fermions is bounded by entanglement fluctuations and is logarithmic for free fermions”
Raghav G. Jha; Dengyao Luo; Alexander F. Kemper; Bojko N. Bakalov
This Supplemental Material (SM) contains the proofs and numerical methods. Section S1 fixes the notation and summarizes the Gaussian-restricted result. Section S2 derives the exact parity-resolved stabilizer fidelity, the capacity bound, and the finite-size estimates. Section S3 proves Gaussian optimality for up to three BCS pairs and gives four-pair counterexamples. Section S4 treats local particle-number conservation.
Setting, notation, and prior work
Setting and conventions. We consider pure fermionic states of definite total parity on N = N A + N B N = N_A + N_B N = N A + N B modes, with the Jordan–Wigner (JW) ordering chosen with A A A first. The nonlocal SRE M α N L M^{\mathrm{NL}}_\alpha M α NL is defined in Eq. (2) , with M α M_\alpha M α the stabilizer Rényi entropy [26]. For fermions, local parity superselection restricts U A , U B U_A,U_B U A , U B to unitaries commuting with the respective local parities [39 , 63 ] ; this gives M α f , N L M^{f,\mathrm{NL}}_\alpha M α f , NL and the parity-resolved fidelity Φ S N L \Phi^{\mathrm{NL}}_S Φ S NL . These local unitaries need not be Gaussian.
The graded Schmidt spectrum is invariant under the parity-even local group. The fidelity problem of Sec. S2 maximizes the overlap with stabilizer targets of allowed flat spectra; the SRE problem in Sec. S3 minimizes Pauli moments over the entire allowed orbit. A fidelity bound controls the SRE but need not determine its exact minimum. The corresponding qubit quantities without the parity constraint are denoted D min N L D^{\mathrm{NL}}_{\min} D m i n NL and M α N L M^{\mathrm{NL}}_\alpha M α NL . Logarithms log 2 \log_2 log 2 are in bits, whereas ln \ln ln denotes the natural logarithm. Table S1 collects the notation.
Table S1. TABLE S1. Notation.
Gaussian-restricted nonlocal SRE. For BCS occupations p j p_j p j , q j = 1 − p j q_j = 1-p_j q j = 1 − p j , and x j = ( 1 − 2 p j ) 2 x_j = (1-2p_j)^2 x j = ( 1 − 2 p j ) 2 , the Gaussian-restricted expression studied in Refs. [22, 23] is
M α ( p ) = ∑ j 1 1 − α log 2 [ 1 + x j α + ( 1 − x j ) α 2 ] . (20) \mathcal{M}_\alpha(p) = \sum_j \frac{1}{1-\alpha}\log_2\left[\frac{1+x_j^\alpha+(1-x_j)^\alpha}{2}\right]. \tag*{(20)} M α ( p ) = j ∑ 1 − α 1 log 2 [ 2 1 + x j α + ( 1 − x j ) α ] . (20)
Its identification with the minimum over all local Gaussian unitaries is proved at α = 2 \alpha= 2 α = 2 for arbitrary subsystem size in Ref. [22]; the corresponding statement for α > 2 \alpha> 2 α > 2 is conjectural there. At α = 2 \alpha= 2 α = 2 , using 1 − x j = 4 p j q j 1-x_j = 4p_jq_j 1 − x j = 4 p j q j , Eq. (20) reduces to M 2 ( p ) M_2(p) M 2 ( p ) in Eq. (4) of the main text. For occupations bounded away from 0 0 0 , 1 2 \frac{1}{2} 2 1 , and 1 1 1 , this Gaussian-restricted quantity is extensive in the number of BCS pairs.
Relation to prior work. For qubits, the exact nonlocal stabilizer fidelity is determined by the Schmidt spectrum [24, 27], while dyadic spectral bounds control the nonlocal SRE up to additive constants [28]. Capacity-based bounds and smooth-spectrum asymptotics were studied in Ref. [25]. The relationship between spectrum flatness and magic was examined in Refs. [64 ] [19, 29]. The pure-state inequality M 2 ≤ 4 D min M_2 \le4D_{\min} M 2 ≤ 4 D m i n [38 ] [24] is extended to α > 1 \alpha> 1 α > 1 in Eq. (25) . The Gaussian-restricted result used here is from Refs. [22, 23].
Numerical methods. For many BCS pairs, listing all 2 m 2^m 2 m Schmidt levels to evaluate Eq. (6) is impractical. Write X = X min + Y X = X_{\min} + Y X = X m i n + Y , where Y Y Y is a sum of independent Bernoulli energy increments with gaps g j g_j g j as defined in Proposition S3. We propagate the distribution of Y Y Y with parity resolution on a 10 − 3 10^{-3} 1 0 − 3 -bit grid, splitting weights linearly between neighboring bins. Cumulative probabilities and root sums evaluate the pure and balanced flat-template fidelities, without enumerating Schmidt levels; truncation at Y c u t Y_{\mathrm{cut}} Y cut is exact for thresholds below it. Against full enumeration for random occupations at m = 16 , 20 m = 16,20 m = 16 , 20 , the error is at most 1.5 × 10 − 7 1.5 \times10^{-7} 1.5 × 1 0 − 7 bits; against the exact weight classes of 2048 identical pairs it is at most 1.3 × 10 − 3 1.3 \times10^{-3} 1.3 × 1 0 − 3 bits. The occupations are extracted from block correlation matrices: an antiperiodic ring of 65536 sites for chains and a 1024 × 1024 1024 \times1024 1024 × 1024 torus for the two-dimensional Fermi sea (ε k = − 2 ( cos k x + cos k y ) < μ F = − 1 \varepsilon_{\mathbf{k}} = -2(\cos k_x + \cos k_y) < \mu_F = -1 ε k = − 2 ( cos k x + cos k y ) < μ F = − 1 , filling 0.31 0.31 0.31 ). Haar-random Gaussian states are averaged over seeds. The contributions in Figure 2 (b) use the kernel of Eq. (26) ; reference-state SREs use the Pauli fourth-moment formula [Eq. (65) ] evaluated with fast Walsh–Hadamard transforms.
The ground states of the interacting chains discussed in Sec. IV E were obtained using DMRG: open boundaries, a cut in the middle, conserved particle number (and $S^z = 0 for Hubbard chains), bond dimension 128 (t t t –V V V ) and 384 (Hubbard), and truncation errors at most 6 × 10 − 4 6 \times10^{-4} 6 × 1 0 − 4 . The chains are at half filling; for odd L L L we take N = ( L + 1 ) / 2 N=(L+1)/2 N = ( L + 1 ) /2 , which selects one charge-density-wave pattern for V > 2 t V>2t V > 2 t , as does an edge pinning field for even L L L . At V = 0 V=0 V = 0 and U = 0 U=0 U = 0 the exact C E C_E C E is reproduced to within 0.3%. For these states M 2 f , N L M_{2}^{f,\mathrm{NL}} M 2 f , NL is bounded from above by the graded reference state of Eq. (32) and from below by the staircase bound of Ref. [28]: if ν ↑ \nu^\uparrow ν ↑ repeats λ 2 k − 1 + 1 \lambda_{2^{k-1}+1} λ 2 k − 1 + 1 on each dyadic shell 2 k − 1 < R ≤ 2 k 2^{k-1}<R\le2^k 2 k − 1 < R ≤ 2 k (and ν 1 ↑ = λ 1 \nu_1^\uparrow=\lambda_1 ν 1 ↑ = λ 1 ), then
M 2 f , N L ≥ M 2 N L ≥ − log 2 W ( ν ↑ ) (21) M_{2}^{f,\mathrm{NL}} \ge M_{2}^{\mathrm{NL}} \ge-\log_2 W(\sqrt{\nu^\uparrow}) \tag*{(21)} M 2 f , NL ≥ M 2 NL ≥ − log 2 W ( ν ↑ ) (21)
with W W W the sum of Eq. (65) on the ordered amplitudes ν ↑ \sqrt{\nu^\uparrow} ν ↑ . When W > 1 W>1 W > 1 this bound is negative and we use M 2 f , N L ≥ 0 M_{2}^{f,\mathrm{NL}}\ge0 M 2 f , NL ≥ 0 instead.
Exact nonlocal stabilizer fidelity
Exact fidelity from graded spectra
Let ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ be a pure state of definite total fermion parity on N = N A + N B N=N_A+N_B N = N A + N B modes, and let n = min ( N A , N B ) n=\min(N_A,N_B) n = min ( N A , N B ) be the number of modes on the smaller side. Each parity sector contains at most 2 n − 1 2^{n-1} 2 n − 1 nonzero Schmidt values. The reduced state ρ A \rho_A ρ A commutes with the local parity P A P_A P A , so its entanglement spectrum splits into λ = ( λ ( + ) , λ ( − ) ) \lambda=(\lambda^{(+)},\lambda^{(-)}) λ = ( λ ( + ) , λ ( − ) ) . Write a i ( η ) ↓ = λ i ( η ) ↓ a_i^{(\eta)\downarrow}=\sqrt{\lambda_i^{(\eta)\downarrow}} a i ( η ) ↓ = λ i ( η ) ↓ for the Schmidt amplitudes sorted within each sector η = ± \eta=\pm η = ± .
Local orbits. Two pure states of the same total parity are related by parity-even local unitaries if and only if their graded Schmidt spectra agree. More generally, the largest overlap between the local orbits of states ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ and ϕ \phi ϕ with graded spectra λ \lambda λ and ν \nu ν is
max U A , U B ∣ ⟨ ϕ ∣ U A ⊗ U B ∣ ψ ⟩ ∣ = β ( λ , ν ) : = ∑ η = ± ∑ i λ i ( η ) ↓ ν i ( η ) ↓ . (22) \max_{U_A,U_B}|\langle\phi|U_A\otimes U_B|\psi\rangle|=\beta(\lambda,\nu):=\sum_{\eta=\pm}\sum_i\sqrt{\lambda_i^{(\eta)\downarrow}\nu_i^{(\eta)\downarrow}}. \tag*{(22)} U A , U B max ∣ ⟨ ϕ ∣ U A ⊗ U B ∣ ψ ⟩ ∣ = β ( λ , ν ) := η = ± ∑ i ∑ λ i ( η ) ↓ ν i ( η ) ↓ . (22)
Indeed, the Schmidt vectors can be chosen to have definite local parity. Parity-even unitaries rotate their bases independently in the two sectors, and von Neumann’s trace inequality bounds each sector’s overlap by the sum of products of the ordered Schmidt amplitudes. Aligning both bases and their relative phases attains the sum in Eq. (22) (the graded counterpart of Ref. [24]).
Allowed stabilizer spectra. For stabilizer targets of definite total parity, the nonzero Schmidt values are flat, and their parity split must be one of two kinds: pure , with all 2 k 2^k 2 k levels in one sector (0 ≤ k ≤ n − 1 0\le k\le n-1 0 ≤ k ≤ n − 1 ), or balanced , with 2 k − 1 2^{k-1} 2 k − 1 levels in each sector (1 ≤ k ≤ n 1\le k\le n 1 ≤ k ≤ n ). Denote this set of spectra by S ♭ \mathcal{S}_\flat S ♭ .
Here is the classification and its converse. For a stabilizer state s s s , let S A S_A S A be its stabilizers supported on A A A . Then ρ A = 2 − N A ∑ g ∈ S A g \rho_A=2^{-N_A}\sum_{g\in S_A}g ρ A = 2 − N A ∑ g ∈ S A g is proportional to a projector, and all g ∈ S A g\in S_A g ∈ S A commute with P A P_A P A . If ± P A ∈ S A \pm P_A\in S_A ± P A ∈ S A , the support lies in one parity sector; otherwise Tr ( P A ρ A ) = 0 \operatorname{Tr}(P_A\rho_A)=0 Tr ( P A ρ A ) = 0 , so the flat levels occur equally in both sectors. Conversely, k k k fermionic Bell pairs realize a balanced split, whereas
2 − k / 2 ∑ s ∈ { 0 , 1 } k ∣ s , ∣ s ∣ m o d 2 ⟩ ∣ s , ∣ s ∣ m o d 2 ⟩ 2^{-k/2}\sum_{s\in\{0,1\}^k}|s,|s|\mathbin{\mathrm{mod}}2\rangle|s,|s|\mathbin{\mathrm{mod}}2\rangle 2 − k /2 s ∈ { 0 , 1 } k ∑ ∣ s , ∣ s ∣ mod 2 ⟩ ∣ s , ∣ s ∣ mod 2 ⟩
realizes a pure split: the final mode on each side stores the parity of k k k Bell pairs. Both are stabilizer states, the latter obtained by local Clifford gates. Appropriate bit flips realize either sector and either total parity. Finally, an indefinite-parity stabilizer target cannot improve the fidelity: if P ∣ ψ ⟩ = π ∣ ψ ⟩ P|\psi\rangle=\pi|\psi\rangle P ∣ ψ ⟩ = π ∣ ψ ⟩ and ⟨ s ∣ P ∣ s ⟩ = 0 \langle s|P|s\rangle=0 ⟨ s ∣ P ∣ s ⟩ = 0 , then ( I + π P ) ∣ s ⟩ / 2 (I+\pi P)|s\rangle/\sqrt{2} ( I + π P ) ∣ s ⟩ / 2 is a parity-definite stabilizer target whose squared overlap with ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ is twice as large.
Exact formula. The fermionic nonlocal stabilizer fidelity is Φ S N L = max U A , U B max s ∈ S T A B ∣ ⟨ s ∣ U A ⊗ U B ∣ ψ ⟩ ∣ 2 \Phi_S^{\mathrm{NL}}=\max_{U_A,U_B}\max_{s\in\mathrm{STAB}}|\langle s|U_A\otimes U_B|\psi\rangle|^2 Φ S NL = max U A , U B max s ∈ STAB ∣ ⟨ s ∣ U A ⊗ U B ∣ ψ ⟩ ∣ 2 , with parity-even U A , U B U_A,U_B U A , U B . Maximizing Eq. (22) over ν ∈ S ♭ \nu\in\mathcal{S}_\flat ν ∈ S ♭ gives
Φ S N L = max { max 0 ≤ k ≤ n − 1 η = ± 2 − k ( ∑ i = 1 2 k a i ( η ) ) 2 , max 1 ≤ k ≤ n 2 − k ( ∑ i = 1 2 k − 1 [ a i ( + ) + a i ( − ) ] ) 2 } . (23) \Phi_S^{\mathrm{NL}}=\max\left\{\max_{\substack{0\le k\le n-1\\ \eta=\pm}}2^{-k}\left(\sum_{i=1}^{2^k}a_i^{(\eta)}\right)^2,\quad\max_{1\le k\le n}2^{-k}\left(\sum_{i=1}^{2^{k-1}}\left[a_i^{(+)}+a_i^{(-)}\right]\right)^2\right\}. \tag*{(23)} Φ S NL = max ⎩ ⎨ ⎧ 0 ≤ k ≤ n − 1 η = ± max 2 − k i = 1 ∑ 2 k a i ( η ) 2 , 1 ≤ k ≤ n max 2 − k i = 1 ∑ 2 k − 1 [ a i ( + ) + a i ( − ) ] 2 ⎭ ⎬ ⎫ . (23)
The two lines correspond, respectively, to pure and balanced splits. Thus Φ S N L = max ν ∈ S ♭ β ( λ , ν ) 2 \Phi_S^{\mathrm{NL}}=\max_{\nu\in\mathcal{S}_\flat}\beta(\lambda,\nu)^2 Φ S NL = max ν ∈ S ♭ β ( λ , ν ) 2 depends only on the graded spectrum. Operationally, the targets are Bell pairs or Bell pairs with a copied parity bit, reached by parity-even local unitaries without communication. Appending local vacuum modes does not change Eq. (23) : newly admissible templates have more slots than the old parity-sector dimensions, but no additional nonzero Schmidt amplitudes. For the same reason the upper limits on k k k never bind: a sector holds at most 2 n − 1 2^{n-1} 2 n − 1 nonzero values, and a term with more slots than that is smaller than the term with as many, so the maxima may run over all k k k , as in Eq. (7) of the main text, and Φ S N L \Phi_S^{\mathrm{NL}} Φ S NL depends neither on N A , N B N_A,N_B N A , N B nor on unpaired modes.
For m m m BCS pairs, the graded Schmidt spectrum of the Gaussian normal form [1] has probabilities ∏ j p j s j ( 1 − p j ) 1 − s j \prod_j p_j^{s_j}(1-p_j)^{1-s_j} ∏ j p j s j ( 1 − p j ) 1 − s j and sector ( − 1 ) ∣ s ∣ + c (-1)^{|s|+c} ( − 1 ) ∣ s ∣ + c , where c ∈ { 0 , 1 } c\in\{0,1\} c ∈ { 0 , 1 } accounts for unpaired occupied modes.
Two immediate consequences. Write Φ q = 2 − D m N L \Phi^q=2^{-D_m^{\mathrm{NL}}} Φ q = 2 − D m NL for the qubit fidelity and D m f , N L = − log 2 Φ S N L D_m^{f,\mathrm{NL}}=-\log_2\Phi_S^{\mathrm{NL}} D m f , NL = − log 2 Φ S NL . Parity super-selection costs at most one bit:
D m N L ≤ D m f , N L ≤ D m N L + 1. (24) D_m^{\mathrm{NL}}\le D_m^{f,\mathrm{NL}}\le D_m^{\mathrm{NL}}+1. \tag*{(24)} D m NL ≤ D m f , NL ≤ D m NL + 1. (24)
To see the upper bound, take the 2 k 2^k 2 k levels of an optimal qubit flat template. If k < n k<n k < n , they fit inside a balanced fermionic template of total rank 2 k + 1 2^{k+1} 2 k + 1 , with at least half the qubit fidelity. If k = n k=n k = n , the full template is already balanced.
The pure-state SRE–fidelity inequality [38 ] [24] extends to α > 1 \alpha>1 α > 1 : if ∣ s ⟩ |s\rangle ∣ s ⟩ has stabilizer group G G G , the power mean of order 2 α 2\alpha 2 α gives ∣ ⟨ s ∣ ψ ⟩ ∣ 2 = 2 − N ∑ g ∈ G ⟨ ψ ∣ g ∣ ψ ⟩ ≤ ( 2 − N ∑ P ⟨ ψ ∣ P ∣ ψ ⟩ 2 α ) 1 / ( 2 α ) = 2 − ( α − 1 ) M α / ( 2 α ) |\langle s|\psi\rangle|^2 = 2^{-N}\sum_{g\in G}\langle\psi|g|\psi\rangle\le\left(2^{-N}\sum_P\langle\psi|P|\psi\rangle^{2\alpha}\right)^{1/(2\alpha)} = 2^{-(\alpha-1)M_\alpha/(2\alpha)} ∣ ⟨ s ∣ ψ ⟩ ∣ 2 = 2 − N ∑ g ∈ G ⟨ ψ ∣ g ∣ ψ ⟩ ≤ ( 2 − N ∑ P ⟨ ψ ∣ P ∣ ψ ⟩ 2 α ) 1/ ( 2 α ) = 2 − ( α − 1 ) M α / ( 2 α ) , so M α ≤ α α − 1 D min M_\alpha\le\frac{\alpha}{\alpha-1}D_{\min} M α ≤ α − 1 α D m i n . Hence, for α > 1 \alpha> 1 α > 1 ,
M α f , N L ( ψ ) ≤ 2 α α − 1 D min f , N L , (25) M_\alpha^{f,\mathrm{NL}}(\psi) \le\frac{2\alpha}{\alpha-1}D_{\min}^{f,\mathrm{NL}}, \tag*{(25)} M α f , NL ( ψ ) ≤ α − 1 2 α D m i n f , NL , (25)
M 2 f , N L ( ψ ) ≤ 4 D min f , N L ( α = 2 ) . M_2^{f,\mathrm{NL}}(\psi) \le4D_{\min}^{f,\mathrm{NL}} \qquad(\alpha=2). M 2 f , NL ( ψ ) ≤ 4 D m i n f , NL ( α = 2 ) .
Apply the inequality to a parity-even local unitary attaining Φ S f , N L \Phi_S^{f,\mathrm{NL}} Φ S f , NL . This bound also supplies the T T T -count interpretation discussed in the main text, using the single-gate bound of Ref. [32 ] .
A universal bound from the capacity of entanglement
Let λ 1 ≥ λ 2 ≥ ⋯ \lambda_1 \ge\lambda_2 \ge\cdots λ 1 ≥ λ 2 ≥ ⋯ be the sorted ungraded Schmidt probabilities. Regard X = − log 2 λ X=-\log_2\lambda X = − log 2 λ as a random variable taking X i = − log 2 λ i X_i=-\log_2\lambda_i X i = − log 2 λ i with probability λ i \lambda_i λ i . Its variance is Var X = C E / ( ln 2 ) 2 \operatorname{Var}X=C_E/(\ln2)^2 Var X = C E / ( ln 2 ) 2 , where C E C_E C E is the capacity of entanglement in natural units [36 , 37 ] .
The flat-truncation identity. Dropping the parity and power-of-two constraints, and defining b ( z ) : = ( 2 ⋅ 2 z / 2 − 2 z ) + b(z):=(2\cdot2^{z/2}-2^z)_+ b ( z ) := ( 2 ⋅ 2 z /2 − 2 z ) + , gives the identity
Φ ∗ : = max K ≥ 1 1 K ( ∑ i = 1 K λ i ) 2 = sup y ∈ R E λ b ( X − y ) . (26) \Phi_* := \max_{K\ge1}\frac{1}{K}\left(\sum_{i=1}^{K}\sqrt{\lambda_i}\right)^2=\sup_{y\in\mathbb{R}}\mathbb{E}_{\lambda}b(X-y). \tag*{(26)} Φ ∗ := K ≥ 1 max K 1 ( i = 1 ∑ K λ i ) 2 = y ∈ R sup E λ b ( X − y ) . (26)
To prove the identity, put a i = λ i a_i=\sqrt{\lambda_i} a i = λ i . Completing the square gives, for every nonempty set of indices T T T ,
( ∑ i ∈ T a i ) 2 ∣ T ∣ = max t ≥ 0 ∑ i ∈ T ( 2 t a i − t 2 ) . \frac{\left(\sum_{i\in T}a_i\right)^2}{|T|}=\max_{t\ge0}\sum_{i\in T}(2ta_i-t^2). ∣ T ∣ ( ∑ i ∈ T a i ) 2 = t ≥ 0 max i ∈ T ∑ ( 2 t a i − t 2 ) .
At fixed t t t , the maximizing set contains exactly the positive summands; at fixed size, it contains the largest amplitudes. Consequently,
Φ ∗ = max t ≥ 0 ∑ i ( 2 t a i − t 2 ) + . \Phi_*=\max_{t\ge0}\sum_i(2ta_i-t^2)_+. Φ ∗ = t ≥ 0 max i ∑ ( 2 t a i − t 2 ) + .
Substituting t = 2 − y / 2 t=2^{-y/2} t = 2 − y /2 turns each term into λ i b ( X i − y ) \lambda_i b(X_i-y) λ i b ( X i − y ) and proves Eq. (26) .
Proposition S1 (Capacity bound). Every pure state of definite fermion parity obeys
Φ ∗ ≥ 2 3 ( C E + 5 ) , Φ ∗ q ≥ Φ ∗ 2 , \Phi_* \ge\frac{2}{\sqrt{3(C_E+5)}}, \qquad\Phi_*^{q}\ge\frac{\Phi_*}{\sqrt{2}}, Φ ∗ ≥ 3 ( C E + 5 ) 2 , Φ ∗ q ≥ 2 Φ ∗ ,
Φ S N L ≥ ( 2 − 1 ) Φ ∗ ≥ 2 ( 2 − 1 ) 3 ( C E + 5 ) . (27) \Phi_S^{\mathrm{NL}}\ge(\sqrt{2}-1)\Phi_*\ge\frac{2(\sqrt{2}-1)}{\sqrt{3(C_E+5)}}. \tag*{(27)} Φ S NL ≥ ( 2 − 1 ) Φ ∗ ≥ 3 ( C E + 5 ) 2 ( 2 − 1 ) . (27)
In particular,
D min N L ≤ 1 2 log 2 [ 3 2 ( C E + 5 ) ] , D_{\min}^{\mathrm{NL}}\le\frac{1}{2}\log_2\left[\frac{3}{2}(C_E+5)\right], D m i n NL ≤ 2 1 log 2 [ 2 3 ( C E + 5 ) ] ,
D min f , N L ≤ 1 2 log 2 [ 3 4 ( 3 + 2 2 ) ( C E + 5 ) ] . D_{\min}^{f,\mathrm{NL}}\le\frac{1}{2}\log_2\left[\frac{3}{4}(3+2\sqrt{2})(C_E+5)\right]. D m i n f , NL ≤ 2 1 log 2 [ 4 3 ( 3 + 2 2 ) ( C E + 5 ) ] .
Balanced targets alone, i.e. fermionic Bell-pair conversion, achieve fidelity at least [ 6 ( C E + 5 ) ] − 1 / 2 [6(C_E+5)]^{-1/2} [ 6 ( C E + 5 ) ] − 1/2 . The bounds on Φ ∗ \Phi_* Φ ∗ and Φ q \Phi^q Φ q also hold for qubits.
Proof. Smoothing and the peak bound. Set γ = ln 2 \gamma=\ln2 γ = ln 2 . The function
g ( t ) : = γ 4 b ( 2 − t ) = γ ( e − γ t / 2 − e − γ t ) , t ≥ 0 , g(t):=\frac{\gamma}{4}b(2-t)=\gamma(e^{-\gamma t/2}-e^{-\gamma t}), \qquad t\ge0, g ( t ) := 4 γ b ( 2 − t ) = γ ( e − γ t /2 − e − γ t ) , t ≥ 0 ,
(and g ( t ) = 0 g(t)=0 g ( t ) = 0 for t < 0 t<0 t < 0 ) is the density of E 1 + E 2 E_1+E_2 E 1 + E 2 , where E 1 , E 2 E_1,E_2 E 1 , E 2 are independent exponential random variables of rates γ / 2 \gamma/2 γ /2 and γ \gamma γ . With V = 2 − E 1 − E 2 V=2-E_1-E_2 V = 2 − E 1 − E 2 independent of X X X , Eq. (26) reads
Φ ∗ = 4 γ ∥ ρ X − V ∥ ∞ , Var ( X − V ) = C E + 5 γ 2 . \Phi_*=\frac{4}{\gamma}\|\rho_{X-V}\|_\infty,\qquad\operatorname{Var}(X-V)=\frac{C_E+5}{\gamma^2}. Φ ∗ = γ 4 ∥ ρ X − V ∥ ∞ , Var ( X − V ) = γ 2 C E + 5 .
For any probability density ρ \rho ρ bounded by M M M , its variance is at least 1 / ( 12 M 2 ) 1/(12M^2) 1/ ( 12 M 2 ) : among densities bounded by M M M , the uniform density of height M M M on an interval minimizes the second moment about any fixed center. Taking the center to be the mean yields ∥ ρ ∥ ∞ ≥ ( 12 Var ρ ) − 1 / 2 \|\rho\|_\infty\ge(12\operatorname{Var}\rho)^{-1/2} ∥ ρ ∥ ∞ ≥ ( 12 Var ρ ) − 1/2 . Applying this to ρ X − V \rho_{X-V} ρ X − V gives the first inequality in Eq. (S8).
Rounding to a power of two. For any nonnegative decreasing sequence z i z_i z i , let f ( h ) = ( ∑ i = 1 h z i ) 2 / h f(h)=(\sum_{i=1}^{h}z_i)^2/h f ( h ) = ( ∑ i = 1 h z i ) 2 / h . If L ≤ h ≤ U L\le h\le U L ≤ h ≤ U with L , U L,U L , U consecutive powers of two, monotonicity implies f ( L ) ≥ ( L / h ) f ( h ) f(L)\ge(L/h)f(h) f ( L ) ≥ ( L / h ) f ( h ) and f ( U ) ≥ ( h / U ) f ( h ) f(U)\ge(h/U)f(h) f ( U ) ≥ ( h / U ) f ( h ) . Hence
max { f ( L ) , f ( U ) } ≥ f ( h ) 2 ( L ≤ h ≤ U ) . \max\{f(L),f(U)\}\ge\frac{f(h)}{\sqrt{2}}\qquad(L\le h\le U). max { f ( L ) , f ( U )} ≥ 2 f ( h ) ( L ≤ h ≤ U ) .
Applying this to the ungraded amplitudes proves Φ q ≥ Φ ∗ / 2 \Phi^q\ge\Phi_*/\sqrt{2} Φ q ≥ Φ ∗ / 2 .
Rounding with parity. Put R = 2 n − 1 R=2^{n-1} R = 2 n − 1 and S η ( h ) = ∑ i = 1 h a i η S_\eta(h)=\sum_{i=1}^{h}a_i^\eta S η ( h ) = ∑ i = 1 h a i η . Let F g r F_{\mathrm{gr}} F gr be the maximum of
S + ( h ) 2 h , S − ( h ) 2 h , [ S + ( h ) + S − ( h ) ] 2 2 h , \frac{S_+(h)^2}{h},\qquad\frac{S_-(h)^2}{h},\qquad\frac{[S_+(h)+S_-(h)]^2}{2h}, h S + ( h ) 2 , h S − ( h ) 2 , 2 h [ S + ( h ) + S − ( h ) ] 2 ,
over integers 1 ≤ h ≤ R 1\le h\le R 1 ≤ h ≤ R .
Choose a maximizing truncation for Φ ∗ \Phi_* Φ ∗ , containing h h h levels of one parity and ℓ ≤ h \ell\le h ℓ ≤ h of the other, with amplitude sums A A A and B B B . Then
Φ ∗ = ( A + B ) 2 h + ℓ , h ≤ R . \Phi_*=\frac{(A+B)^2}{h+\ell},\qquad h\le R. Φ ∗ = h + ℓ ( A + B ) 2 , h ≤ R .
If ℓ = 0 \ell=0 ℓ = 0 , then F g r ≥ Φ ∗ F_{\mathrm{gr}}\ge\Phi_* F gr ≥ Φ ∗ . For ℓ > 0 \ell>0 ℓ > 0 , the pure and balanced candidates give
F g r ≥ max { A 2 h , B 2 ℓ , ( A + B ) 2 2 h } . F_{\mathrm{gr}}\ge\max\left\{\frac{A^2}{h},\frac{B^2}{\ell},\frac{(A+B)^2}{2h}\right\}. F gr ≥ max { h A 2 , ℓ B 2 , 2 h ( A + B ) 2 } .
Since
max { A 2 h , B 2 ℓ } ≥ ( A + B ) 2 ( h + ℓ ) 2 , \max\left\{\frac{A^2}{h},\frac{B^2}{\ell}\right\}\ge\frac{(A+B)^2}{(\sqrt{h}+\sqrt{\ell})^2}, max { h A 2 , ℓ B 2 } ≥ ( h + ℓ ) 2 ( A + B ) 2 ,
setting x = ℓ / h x=\ell/h x = ℓ / h yields
F g r Φ ∗ ≥ max { 1 + x ( 1 + x ) 2 , 1 + x 2 } ≥ 2 − 2 . \frac{F_{\mathrm{gr}}}{\Phi_*}\ge\max\left\{\frac{1+x}{(1+\sqrt{x})^2},\frac{1+x}{2}\right\}\ge2-\sqrt{2}. Φ ∗ F gr ≥ max { ( 1 + x ) 2 1 + x , 2 1 + x } ≥ 2 − 2 .
The two expressions cross at x = ( 2 − 1 ) 2 x=(\sqrt{2}-1)^2 x = ( 2 − 1 ) 2 . Applying the preceding rounding estimate to a i ( + ) a_i^{(+)} a i ( + ) , a i ( − ) a_i^{(-)} a i ( − ) , and a i ( + ) + a i ( − ) a_i^{(+)}+a_i^{(-)} a i ( + ) + a i ( − ) gives
Φ S N L ≥ F g r 2 ≥ ( 2 − 1 ) Φ ∗ . \Phi_S^{\mathrm{NL}} \ge\frac{F_{\mathrm{gr}}}{\sqrt{2}} \ge(\sqrt{2}-1)\Phi_\ast. Φ S NL ≥ 2 F gr ≥ ( 2 − 1 ) Φ ∗ .
This proves Eq. (S8). Finally, the balanced template containing the 2 k 2^k 2 k largest ungraded levels has at least half the corresponding qubit fidelity (and agrees with it when k = n k=n k = n ). Therefore the best balanced target achieves Φ S b a l ≥ Φ q / 2 ≥ [ 6 ( C E + 5 ) ] − 1 / 2 \Phi_S^{\mathrm{bal}}\ge\Phi^q/2\ge[6(C_E+5)]^{-1/2} Φ S bal ≥ Φ q /2 ≥ [ 6 ( C E + 5 ) ] − 1/2 , as claimed.
The capacity bound holds at every system size; it gives no converse inequality from C E C_E C E alone. A single Schmidt level of weight close to one can keep Φ S N L \Phi_S^{\mathrm{NL}} Φ S NL close to one even when the remaining levels make C E C_E C E arbitrarily large. For BCS states, C E ≤ c ∗ m C_E\le c_\ast m C E ≤ c ∗ m [Eq. (14) ].
The bound of Proposition S1 improves the capacity-based estimate of Ref. [25], including the one-bit parity-superselection correction in Eq. (24) . The first inequality of Eq. (S8) is asymptotically sharp: for spectra λ R ∝ R − 1 \lambda_R\propto R^{-1} λ R ∝ R − 1 supported on 1 ≤ R ≤ 2 L 1\le R\le2^L 1 ≤ R ≤ 2 L , the ratio of its two sides tends to one as L → ∞ L\to\infty L → ∞ .
Bounds from the number of modes and from the spread of X X X
We complement the capacity bound with two finite-size estimates: a lower bound in terms of the number of local modes, valid for every pure state, and a converse for BCS spectra. Together they establish the optimal m − 1 / 2 m^{-1/2} m − 1/2 fidelity scaling when all BCS occupations stay uniformly away from 0 0 0 , 1 2 \frac{1}{2} 2 1 , and 1 1 1 .
Proposition S2 (Dimension-dependent fidelity bound). For a pure state of definite total fermion parity with n ≥ 2 n\ge2 n ≥ 2 modes on the smaller side,
Φ S N L ≥ 3 ( 2 + 2 ) 8 n , D f N L ≤ log 2 n − log 2 3 ( 2 + 2 ) 8 . \Phi_S^{\mathrm{NL}}\ge\frac{3(2+\sqrt{2})}{8n},\qquad D_f^{\mathrm{NL}}\le\log_2 n-\log_2\frac{3(2+\sqrt{2})}{8}. Φ S NL ≥ 8 n 3 ( 2 + 2 ) , D f NL ≤ log 2 n − log 2 8 3 ( 2 + 2 ) .
Proof. Let λ 1 ≥ ⋯ ≥ λ 2 M ≥ 0 \lambda_1\ge\cdots\ge\lambda_{2^M}\ge0 λ 1 ≥ ⋯ ≥ λ 2 M ≥ 0 have total weight w w w , where M ≥ 1 M\ge1 M ≥ 1 , and define F k = 2 − k ( ∑ j ≤ 2 k λ j ) 2 F_k=2^{-k}\left(\sum_{j\le2^k}\sqrt{\lambda_j}\right)^2 F k = 2 − k ( ∑ j ≤ 2 k λ j ) 2 . Monotonicity gives F k ≥ 2 − k ∑ j ≤ 2 k ( 2 j − 1 ) λ j F_k\ge2^{-k}\sum_{j\le2^k}(2j-1)\lambda_j F k ≥ 2 − k ∑ j ≤ 2 k ( 2 j − 1 ) λ j . The sum of these linear minorants is minimized over decreasing spectra of total weight w w w at an initial flat spectrum: λ 1 = ⋯ = λ h = w / h \lambda_1=\cdots=\lambda_h=w/h λ 1 = ⋯ = λ h = w / h , with all later entries zero. Indeed, every decreasing spectrum is a convex combination of such vectors. For 2 K ≤ h < 2 K + 1 2^K\le h<2^{K+1} 2 K ≤ h < 2 K + 1 , putting t = h / 2 K t=h/2^K t = h / 2 K gives the sum w ( α / t + β t ) w(\alpha/t+\beta t) w ( α / t + β t ) , where α = 2 − 2 − K \alpha=2-2^{-K} α = 2 − 2 − K and β = 1 − 2 K − M \beta=1-2^{K-M} β = 1 − 2 K − M . At h = 1 h=1 h = 1 and h = 2 M h=2^M h = 2 M this sum equals w ( 2 − 2 − M ) ≥ 3 2 w w(2-2^{-M})\ge\frac{3}{2}w w ( 2 − 2 − M ) ≥ 2 3 w ; for the other values, K ≥ 1 K\ge1 K ≥ 1 , α ≥ 3 2 \alpha\ge\frac{3}{2} α ≥ 2 3 , β ≥ 1 2 \beta\ge\frac{1}{2} β ≥ 2 1 , and α / t + β t ≥ 2 α β ≥ 3 \alpha/t+\beta t\ge2\sqrt{\alpha\beta}\ge\sqrt{3} α / t + β t ≥ 2 α β ≥ 3 . Consequently,
∑ k = 0 M F k ≥ 3 2 w . (28) \sum_{k=0}^{M}F_k\ge\frac{3}{2}w. \tag*{(28)} k = 0 ∑ M F k ≥ 2 3 w . (28)
Apply this to the two parity sectors with M = n − 1 M=n-1 M = n − 1 . Write x j 2 , y j 2 x_j^2,y_j^2 x j 2 , y j 2 for their pure-template fidelities at rank 2 j 2^j 2 j . The balanced template of total rank 2 j + 1 2^{j+1} 2 j + 1 has fidelity ( x j + y j ) 2 / 2 (x_j+y_j)^2/2 ( x j + y j ) 2 /2 , and max { x 2 , y 2 , ( x + y ) 2 / 2 } ≥ ( 2 + 2 ) ( x 2 + y 2 ) / 4 \max\{x^2,y^2,(x+y)^2/2\}\ge(2+\sqrt{2})(x^2+y^2)/4 max { x 2 , y 2 , ( x + y ) 2 /2 } ≥ ( 2 + 2 ) ( x 2 + y 2 ) /4 for x , y ≥ 0 x,y\ge0 x , y ≥ 0 . Since ∑ j = 0 n − 1 ( x j 2 + y j 2 ) ≥ 3 2 \sum_{j=0}^{n-1}(x_j^2+y_j^2)\ge\frac{3}{2} ∑ j = 0 n − 1 ( x j 2 + y j 2 ) ≥ 2 3 , one of the n n n admissible ranks attains the stated bound.
The two expressions cross at x = ( 2 − 1 ) 2 x=(\sqrt{2}-1)^2 x = ( 2 − 1 ) 2 . Applying the preceding rounding estimate to a i ( + ) a_i^{(+)} a i ( + ) , a i ( − ) a_i^{(-)} a i ( − ) , and a i ( + ) + a i ( − ) a_i^{(+)}+a_i^{(-)} a i ( + ) + a i ( − ) gives
Φ S N L ≥ F g r 2 ≥ ( 2 − 1 ) Φ ∗ . (29) \Phi_S^{\mathrm{NL}}\ge\frac{F_{\mathrm{gr}}}{\sqrt{2}}\ge(\sqrt{2}-1)\Phi_\ast. \tag*{(29)} Φ S NL ≥ 2 F gr ≥ ( 2 − 1 ) Φ ∗ . (29)
This proves Eq. (S8). Finally, the balanced template containing the 2 k 2^k 2 k largest ungraded levels has at least half the corresponding qubit fidelity (and agrees with it when k = n k=n k = n ). Therefore the best balanced target achieves Φ S b a l ≥ Φ q / 2 ≥ [ 6 ( C E + 5 ) ] − 1 / 2 \Phi_S^{\mathrm{bal}}\ge\Phi^q/2\ge[6(C_E+5)]^{-1/2} Φ S bal ≥ Φ q /2 ≥ [ 6 ( C E + 5 ) ] − 1/2 , as claimed.
The capacity bound holds at every system size; it gives no converse inequality from C E C_E C E alone. A single Schmidt level of weight close to one can keep Φ S N L \Phi_S^{\mathrm{NL}} Φ S NL close to one even when the remaining levels make C E C_E C E arbitrarily large. For BCS states, C E ≤ c ∗ m C_E\le c_\ast m C E ≤ c ∗ m [Eq. (14) ].
The bound of Proposition S1 improves the capacity-based estimate of Ref. [25], including the one-bit parity-superselection correction in Eq. (24) . The first inequality of Eq. (S8) is asymptotically sharp: for spectra λ R ∝ R − 1 \lambda_R\propto R^{-1} λ R ∝ R − 1 supported on 1 ≤ R ≤ 2 L 1\le R\le2^L 1 ≤ R ≤ 2 L , the ratio of its two sides tends to one as L → ∞ L\to\infty L → ∞ .
For the converse, let X = − log 2 λ X=-\log_2\lambda X = − log 2 λ under the ungraded Schmidt distribution and set Q L = sup x Pr ( x − L < X ≤ x ) Q_L=\sup_x\Pr(x-L<X\le x) Q L = sup x Pr ( x − L < X ≤ x ) for L > 0 L>0 L > 0 .
Proposition S3 (Converse for BCS spectra). For every pure state and L > 0 L>0 L > 0 ,
Φ S N L ≤ Φ q ≤ Φ ∗ ≤ 2 Q L 1 − 2 − L / 2 . (30) \Phi_S^{\mathrm{NL}}\le\Phi^q\le\Phi_\ast\le\frac{2Q_L}{1-2^{-L/2}}. \tag*{(30)} Φ S NL ≤ Φ q ≤ Φ ∗ ≤ 1 − 2 − L /2 2 Q L . (30)
*For m m m BCS pairs with C E > 0 C_E>0 C E > 0 , omit deterministic pairs with p j ∈ { 0 , 1 } p_j\in\{0,1\} p j ∈ { 0 , 1 } and set p ˉ j = min ( p j , 1 − p j ) \bar p_j=\min(p_j,1-p_j) p ˉ j = min ( p j , 1 − p j ) , g j = log 2 [ ( 1 − p ˉ j ) / p ˉ j ] g_j=\log_2[(1-\bar p_j)/\bar p_j] g j = log 2 [( 1 − p ˉ j ) / p ˉ j ] , and s 2 = Var X = C E / ln 2 s^2=\operatorname{Var}X=C_E/\ln2 s 2 = Var X = C E / ln 2 . Define the nonnegative quantity r = s − 2 ∑ j p ˉ j ( 1 − p ˉ j ) [ p ˉ j 2 + ( 1 − p ˉ j ) 2 ] g j 3 ≤ max j g j r=s^{-2}\sum_j\bar p_j(1-\bar p_j)[\bar p_j^2+(1-\bar p_j)^2]g_j^3\le\max_j g_j r = s − 2 ∑ j p ˉ j ( 1 − p ˉ j ) [ p ˉ j 2 + ( 1 − p ˉ j ) 2 ] g j 3 ≤ max j g j . Then, with a Berry–Esseen constant C 0 ≤ 0.56 C_0\le0.56 C 0 ≤ 0.56 ,
Φ S N L ≤ 2 a B E ( r ) C E , \Phi_S^{\mathrm{NL}}\le\frac{2^{a_{\mathrm{BE}}(r)}}{\sqrt{C_E}}, Φ S NL ≤ C E 2 a BE ( r ) ,
a B E ( r ) = inf L > 0 log 2 [ 2 ln 2 ( L / 2 π + 2 C 0 r ) 1 − 2 − L / 2 ] . (31) a_{\mathrm{BE}}(r)=\inf_{L>0}\log_2\left[\frac{2\ln2\,(L/\sqrt{2\pi}+2C_0r)}{1-2^{-L/2}}\right]. \tag*{(31)} a BE ( r ) = L > 0 inf log 2 [ 1 − 2 − L /2 2 ln 2 ( L / 2 π + 2 C 0 r ) ] . (31)
If p ˉ j ∈ [ δ , 1 2 − δ ] \bar p_j\in[\delta,\frac{1}{2}-\delta] p ˉ j ∈ [ δ , 2 1 − δ ] for all j j j , 0 < δ < 1 4 0<\delta<\frac{1}{4} 0 < δ < 4 1 , then C E ≥ c δ m C_E\ge c_\delta m C E ≥ c δ m , where c δ = min { c ( δ ) , c ( 1 2 − δ ) } > 0 c_\delta=\min\{c(\delta),c(\frac{1}{2}-\delta)\}>0 c δ = min { c ( δ ) , c ( 2 1 − δ )} > 0 and c ( p ) = p ( 1 − p ) ln 2 [ ( 1 − p ) / p ] c(p)=p(1-p)\ln^2[(1-p)/p] c ( p ) = p ( 1 − p ) ln 2 [( 1 − p ) / p ] . Φ S N L = O ( m − 1 / 2 ) \Phi_S^{\mathrm{NL}}=O(m^{-1/2}) Φ S NL = O ( m − 1/2 ) : for p j = 1 4 p_j=\frac{1}{4} p j = 4 1 , one obtains r = 5 8 log 2 3 r=\frac{5}{8}\log_2 3 r = 8 5 log 2 3 , a B E ( r ) ≃ 2.30 a_{\mathrm{BE}}(r)\simeq2.30 a BE ( r ) ≃ 2.30 , and Φ S N L ≤ 10.4 / m \Phi_S^{\mathrm{NL}}\le10.4/\sqrt{m} Φ S NL ≤ 10.4/ m .
Proof. Choose K K K attaining Φ ∗ = A 2 / K \Phi_\ast=A^2/K Φ ∗ = A 2 / K , with A = ∑ i ≤ K λ i A=\sum_{i\le K}\sqrt{\lambda_i} A = ∑ i ≤ K λ i and x = − log 2 λ K x=-\log_2\lambda_K x = − log 2 λ K . The variational characterization in Eq. (26) implies that the optimal threshold t = A / K t=A/K t = A / K satisfies t ≤ 2 λ K = 2 − 1 − x / 2 t\le2\sqrt{\lambda_K}=2^{-1-x/2} t ≤ 2 λ K = 2 − 1 − x /2 . Moreover, A = ∑ i ≤ K λ i / λ i ≤ 2 x 2 λ K / 2 x A=\sum_{i\le K}\lambda_i/\sqrt{\lambda_i}\le2^x\sqrt{2}\sqrt{\lambda_K/2^x} A = ∑ i ≤ K λ i / λ i ≤ 2 x 2 λ K / 2 x . Partition ( − ∞ , x ] (-\infty,x] ( − ∞ , x ] into intervals of length L L L . Each carries probability at most Q L Q_L Q L , hence A ≤ 2 x / 2 Q L / ( 1 − 2 − L / 2 ) A\le2^{x/2}Q_L/(1-2^{-L/2}) A ≤ 2 x /2 Q L / ( 1 − 2 − L /2 ) . Multiplication by t t t proves Eq. (30) .
For BCS pairs, X X X is a constant plus independent Bernoulli increments of sizes g j g_j g j . Their variance is s 2 s^2 s 2 and the sum of their third absolute central moments is r s 2 rs^2 r s 2 . Berry–Esseen bounds the distance between their distribution function and the normal distribution with the same mean and variance by C 0 r / s C_0r/s C 0 r / s . It follows that Q L ≤ [ L / 2 π + 2 C 0 r ] / s Q_L\le[L/\sqrt{2\pi}+2C_0r]/s Q L ≤ [ L / 2 π + 2 C 0 r ] / s . Use this in Eq. (30) and optimize over L L L . The occupation condition bounds r r r and ensures C E ≥ c δ m C_E\ge c_\delta m C E ≥ c δ m , giving the stated scaling.
If C E = 0 C_E=0 C E = 0 , every BCS pair is empty, filled, or maximally entangled, so the state is a stabilizer state and Φ S N L = 1 \Phi_S^{\mathrm{NL}}=1 Φ S NL = 1 . Together with Proposition S1 and C E ≤ c ∗ m C_E\le c_\ast m C E ≤ c ∗ m , the converse establishes Φ S N L = Θ ( m − 1 / 2 ) \Phi_S^{\mathrm{NL}}=\Theta(m^{-1/2}) Φ S NL = Θ ( m − 1/2 ) under the stated uniform occupation condition.
Small sizes: when local Gaussian unitaries are optimal
Graded reference states
Let λ = ( λ ( + ) , λ ( − ) ) \lambda=(\lambda^{(+)},\lambda^{(-)}) λ = ( λ ( + ) , λ ( − ) ) be the graded Schmidt spectrum of a pure even-parity state ∣ ψ ⟩ \lvert\psi\rangle ∣ ψ ⟩ . After padding the two sides with local vacuum modes if necessary, let n n n be the number of modes on each side. A graded labeling a a a assigns distinct Fock strings a i ∈ { 0 , 1 } n a_i\in\{0,1\}^{n} a i ∈ { 0 , 1 } n to its Schmidt levels, preserving the parity of each level. The associated graded reference state is
∣ ϕ a ⟩ = ∑ i λ i ∣ a i ⟩ A ∣ a i ⟩ B . (32) \lvert\phi_a\rangle=\sum_i\sqrt{\lambda_i}\lvert a_i\rangle_A\lvert a_i\rangle_B . \tag*{(32)} ∣ ϕ a ⟩ = i ∑ λ i ∣ a i ⟩ A ∣ a i ⟩ B . (32)
Its graded spectrum agrees with that of ∣ ψ ⟩ \lvert\psi\rangle ∣ ψ ⟩ . By the graded Schmidt orbit characterization in Sec. S2 A, ϕ a \phi_a ϕ a is therefore reachable by parity-even local unitaries, and M α f , N L ( ∣ ψ ⟩ ) ≤ M α ( ϕ a ) M^{\mathrm{f,NL}}_{\alpha}(\lvert\psi\rangle)\le M_{\alpha}(\phi_a) M α f , NL (∣ ψ ⟩) ≤ M α ( ϕ a ) . For odd total parity, flip a fixed bit of every B B B label; the same conclusion holds. Thus optimizing over graded labelings supplies explicit trial states.
For m m m BCS pairs, the product labeling a = s a=s a = s uses the pair-occupation string s s s . In A A A -first Jordan–Wigner order, the physical BCS state and ϕ s \phi_s ϕ s differ by the parity-even diagonal Clifford ∏ i < j C Z A i A j \prod_{i<j}C Z_{A_iA_j} ∏ i < j C Z A i A j , which removes the phase ( − 1 ) ∣ s ∣ ( ∣ s ∣ − 1 ) / 2 (-1)^{\lvert s\rvert(\lvert s\rvert-1)/2} ( − 1 ) ∣ s ∣ (∣ s ∣ − 1 ) /2 . Consequently, at the Rényi index α = 2 \alpha=2 α = 2 used here, M 2 ( ϕ s ) = M 2 ( p ) M_2(\phi_s)=\mathcal{M}_2(p) M 2 ( ϕ s ) = M 2 ( p ) . In particular, any graded labeling with M 2 ( ϕ a ) < M 2 ( p ) M_2(\phi_a)<\mathcal{M}_2(p) M 2 ( ϕ a ) < M 2 ( p ) witnesses failure of Gaussian optimality.
Gaussian optimality for up to three BCS pairs
We prove that the Gaussian minimum of Eq. (4) is also the minimum over parity-even local unitaries for at most three BCS pairs.
Proposition S4 (Gaussian optimality for up to three pairs). For m ≤ 3 m\le3 m ≤ 3 , let Ψ m \Psi_m Ψ m consist of m m m BCS pairs with occupations p j ∈ [ 0 , 1 ] p_j\in[0,1] p j ∈ [ 0 , 1 ] , q j = 1 − p j q_j=1-p_j q j = 1 − p j , and any additional local modes in the vacuum. Then
M 2 f , N L ( Ψ m ) = M 2 ( p ) = − ∑ j log 2 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] (33) M^{\mathrm{f,NL}}_2(\Psi_m)=\mathcal{M}_2(p)
=-\sum_j\log_2\left[1-4p_jq_j(1-2p_j)^2\right] \tag*{(33)} M 2 f , NL ( Ψ m ) = M 2 ( p ) = − j ∑ log 2 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] (33) The minimum is attained by the BCS state Ψ m \Psi_m Ψ m itself.
Reduction to three pairs. For m < 3 m<3 m < 3 , append 3 − m 3-m 3 − m vacuum pairs with occupations p j = 0 p_j=0 p j = 0 . Every output of an allowed parity-even local unitary on the original system embeds, by tensoring these vacuum modes, into the local orbit covered by the three-pair statement. Since tensoring with a stabilizer vacuum leaves M 2 M_2 M 2 unchanged, the three-pair bound implies M 2 f , N L ( Ψ m ) ≥ M 2 ( p ) M^{\mathrm{f,NL}}_2(\Psi_m)\ge\mathcal{M}_2(p) M 2 f , NL ( Ψ m ) ≥ M 2 ( p ) . The canonical BCS representative attains equality.
Proof ingredients. We use the one-unitary reduction [28, 29], the Pauli-projector identity, and the compression estimate (Theorem 22 of Ref. [28]). Lemmas S1–S4 establish the parity decomposition and the rank-four estimates used below. Figure S1 summarizes the proof.
Figure S1. Proof roadmap for Proposition S4.
Setting. We treat first the interior case 0 < p j < 1 0<p_j<1 0 < p j < 1 for all j j j and then extend to the boundary by continuity. Let n ≥ 3 n\ge3 n ≥ 3 be the number of local modes on each side after padding with vacuum modes if needed, and let d = 2 n d=2^n d = 2 n , D = d / 2 D=d/2 D = d /2 , and q = 4 / d = 2 / D q=4/d=2/D q = 4/ d = 2/ D ; q q q is the constant q q q of Ref. [28], not an occupation. The local parity is P = Z ⊗ n P=Z^{\otimes n} P = Z ⊗ n , and the parity-even local unitary group is G = { U ∈ U ( d ) : [ U , P ] = 0 } ≃ U ( D ) × U ( D ) G=\{U\in U(d):[U,P]=0\}\simeq U(D)\times U(D) G = { U ∈ U ( d ) : [ U , P ] = 0 } ≃ U ( D ) × U ( D ) , block diagonal in the two parity sectors. Let E , O ⊂ F 2 n \mathcal{E},\mathcal{O}\subset\mathbb{F}_2^n E , O ⊂ F 2 n be the Fock strings of even and odd weight, read as integers with the last mode as the least significant bit; this order fixes the signs of the sorted wedges below. The eigenspaces of P P P with eigenvalues ± 1 \pm1 ± 1 are H + = E \mathcal{H}_+=\mathcal{E} H + = E and H − = O \mathcal{H}_-=\mathcal{O} H − = O .
Choose the BCS product labeling by the pair-occupation strings s ∈ F 2 3 s\in\mathbb{F}_2^3 s ∈ F 2 3 , and define
λ s = ∏ j = 1 3 q j 1 − s j p j s j , μ s = λ s . (34) \lambda_s=\prod_{j=1}^{3}q_j^{1-s_j}p_j^{s_j},\qquad\mu_s=\sqrt{\lambda_s}. \tag*{(34)} λ s = j = 1 ∏ 3 q j 1 − s j p j s j , μ s = λ s . (34)
Let M M M be the diagonal d × d d\times d d × d matrix with entries μ s \mu_s μ s on the eight Fock configurations of the three paired modes, with all spectator modes empty, and zero elsewhere. Let
ϕ M = ∑ x , y ( M ) x y ∣ x ⟩ ∣ y ⟩ (35) \phi_M=\sum_{x,y}(M)_{xy}\lvert x\rangle\lvert y\rangle \tag*{(35)} ϕ M = x , y ∑ ( M ) x y ∣ x ⟩ ∣ y ⟩ (35)
be the corresponding product-labeled reference state. In the A A A -first Jordan–Wigner order, the physical BCS state differs from ϕ M \phi_M ϕ M by the parity-even diagonal Clifford ∏ 1 ≤ i < j ≤ 3 C Z A i A j \prod_{1\le i<j\le3}CZ_{A_iA_j} ∏ 1 ≤ i < j ≤ 3 C Z A i A j on A A A . Thus they have the same stabilizer Rényi entropy and the same optimized value of M 2 M_2 M 2 . The state ϕ M \phi_M ϕ M is a tensor product of three single-pair states, so its stabilizer purity, defined below, is ∏ j ( 1 − 4 u j + 16 u j 2 ) = 2 − M 2 ( p ) \prod_j(1-4u_j+16u_j^2)=2^{-\mathcal{M}_2(p)} ∏ j ( 1 − 4 u j + 16 u j 2 ) = 2 − M 2 ( p ) with u j = p j q j u_j=p_jq_j u j = p j q j , by Eq. (4) .
The one-unitary reduction of Refs. [28, 29] applies to the parity-even subgroup G G G because G G G is closed under complex conjugation. Its proof (Lemma 1 of Ref. [28]) is a Cauchy–Schwarz inequality for the Pauli projector, which bounds the stabilizer purity of ( U ⊗ V ) ϕ M (U\otimes V)\phi_M ( U ⊗ V ) ϕ M by the geometric mean of the purity of ( U ⊗ U ‾ ) ϕ M (U\otimes\overline{U})\phi_M ( U ⊗ U ) ϕ M and ( V ⊗ V ‾ ) ϕ M (V\otimes\overline{V})\phi_M ( V ⊗ V ) ϕ M . For U , V ∈ G U,V\in G U , V ∈ G both pairs on the right lie again in G × G G\times G G × G , and ( U ⊗ U ‾ ) ϕ M = ϕ M U T (U\otimes\overline{U})\phi_M= \phi_MU^T ( U ⊗ U ) ϕ M = ϕ M U T . Hence
min U A , U B ∈ G M 2 ( ( U A ⊗ U B ) ϕ M ) = min U ∈ G M 2 ( ϕ M U T ) . (36) \min_{U_A,U_B\in G}M_2((U_A\otimes U_B)\phi_M)=\min_{U\in G}M_2(\phi_MU^T). \tag*{(36)} U A , U B ∈ G min M 2 (( U A ⊗ U B ) ϕ M ) = U ∈ G min M 2 ( ϕ M U T ) . (36)
We therefore fix U ∈ G U\in G U ∈ G , set N = U M U T N=UMU^T N = U M U T , and write ϕ N \phi_N ϕ N for the corresponding bipartite pure state.
Since U U U preserves parity, N N N is block diagonal with respect to H + ⊕ H − \mathcal{H}_+\oplus\mathcal{H}_- H + ⊕ H − . N = A ⊕ B N=A\oplus B N = A ⊕ B , where A A A and B B B are positive semidefinite matrices acting on H + \mathcal{H}_+ H + and H − \mathcal{H}_- H − , respectively. Each has rank four. The nonzero eigenvalues α 1 , … , α 4 \alpha_1,\ldots,\alpha_4 α 1 , … , α 4 of A A A are the amplitudes μ s \mu_s μ s with s s s even, and
Proposition S4. For m ≤ 3 m \le3 m ≤ 3 BCS pairs,
M 2 f , N L ( Ψ m ) = M 2 ( p ) = − ∑ j = 1 m log 2 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] M_{2}^{f,\mathrm{NL}}(\Psi_m)=\mathcal{M}_2(p)=-\sum_{j=1}^{m}\log_2\left[1-4p_jq_j(1-2p_j)^2\right] M 2 f , NL ( Ψ m ) = M 2 ( p ) = − j = 1 ∑ m log 2 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] the BCS state minimizes M 2 M_2 M 2 over its parity-even local-unitary orbit
those of B B B , β 1 , … , β 4 \beta_1,\ldots,\beta_4 β 1 , … , β 4 , the amplitudes with s s s odd. Define
u = ∏ a = 1 4 α a = ∏ a = 1 4 β a = ∏ j = 1 3 p j q j , u=\prod_{a=1}^{4}\alpha_a=\prod_{a=1}^{4}\beta_a=\prod_{j=1}^{3}p_jq_j, u = a = 1 ∏ 4 α a = a = 1 ∏ 4 β a = j = 1 ∏ 3 p j q j ,
T = ∑ a < b ( α a α b ) 4 = ∑ a < b ( β a β b ) 4 . (37) T=\sum_{a<b}(\alpha_a\alpha_b)^4=\sum_{a<b}(\beta_a\beta_b)^4. \tag*{(37)} T = a < b ∑ ( α a α b ) 4 = a < b ∑ ( β a β b ) 4 . (37)
The two equalities follow from the three-pair BCS spectrum. In each parity sector the four Schmidt values multiply to ( ∏ j p j q j ) 2 (\prod_jp_jq_j)^2 ( ∏ j p j q j ) 2 , and their six pairwise products are { p j 2 , q j 2 } × ∏ k ≠ j p k q k \{p_j^2,q_j^2\}\times\prod_{k\ne j}p_kq_k { p j 2 , q j 2 } × ∏ k = j p k q k for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 ; the two multisets of products therefore coincide.
We use the stabilizer purity W W W , as listed in Table S1 :
W ( ϕ ) : = 2 − M 2 ( ϕ ) . (38) W(\phi):=2^{-M_2(\phi)}. \tag*{(38)} W ( ϕ ) := 2 − M 2 ( ϕ ) . (38)
With P n = { I , X , Y , Z } ⊗ n \mathcal{P}_n=\{I,X,Y,Z\}^{\otimes n} P n = { I , X , Y , Z } ⊗ n , define
Π W = d − 2 ∑ σ ∈ P n σ ⊗ 4 , Π F = Π W ∣ Λ 4 C d . (39) \Pi_W=d^{-2}\sum_{\sigma\in\mathcal{P}_n}\sigma^{\otimes4},\qquad\Pi_F=\Pi_W|_{\Lambda^4\mathbb{C}^d}. \tag*{(39)} Π W = d − 2 σ ∈ P n ∑ σ ⊗ 4 , Π F = Π W ∣ Λ 4 C d . (39)
The antisymmetric subspace is invariant under Π W \Pi_W Π W , so Π F \Pi_F Π F is an orthogonal projector. For an operator C C C on C d \mathbb{C}^d C d let Λ 4 ( C ) \Lambda^4(C) Λ 4 ( C ) be the operator it induces on Λ 4 C d \Lambda^4\mathbb{C}^d Λ 4 C d , and put
Q ( 14 ) ( N ) = ∥ Π F Λ 4 ( N ) Π F ∥ H S 2 , Q^{(14)}(N)=\left\|\Pi_F\Lambda^4(N)\Pi_F\right\|_{HS}^2, Q ( 14 ) ( N ) = Π F Λ 4 ( N ) Π F H S 2 ,
c F ( C ) = Tr [ Π F Λ 4 ( C ) ] . (40) c_F(C)=\operatorname{Tr}\left[\Pi_F\Lambda^4(C)\right]. \tag*{(40)} c F ( C ) = Tr [ Π F Λ 4 ( C ) ] . (40)
and I ( μ ) = 7 ( ∑ s μ s 2 ) 2 − 6 ∑ s μ s 8 I(\mu)=7(\sum_s\mu_s^2)^2-6\sum_s\mu_s^8 I ( μ ) = 7 ( ∑ s μ s 2 ) 2 − 6 ∑ s μ s 8 . Proposition 2 of Ref. [28] [its Eq. (42)] is the identity
W ( ϕ N ) = I ( μ ) + 6 d 2 Q ( 14 ) ( N ) + 18 d c F ( N 2 ) − D ( N ) , (41) W(\phi_N)=I(\mu)+6d^2Q^{(14)}(N)+18d\,c_F(N^2)-D(N), \tag*{(41)} W ( ϕ N ) = I ( μ ) + 6 d 2 Q ( 14 ) ( N ) + 18 d c F ( N 2 ) − D ( N ) , (41)
valid for every unitary U U U , with a defect D ( N ) ≥ 0 D(N)\ge0 D ( N ) ≥ 0 that vanishes at N = M N=M N = M . Dropping the defect gives
W ( ϕ N ) ≤ I ( μ ) + 6 d 2 Q ( 14 ) ( N ) + 18 d c F ( N 2 ) . (42) W(\phi_N)\le I(\mu)+6d^2Q^{(14)}(N)+18d\,c_F(N^2). \tag*{(42)} W ( ϕ N ) ≤ I ( μ ) + 6 d 2 Q ( 14 ) ( N ) + 18 d c F ( N 2 ) . (42)
We write e F = e x 1 ∧ e x 2 ∧ e x 3 ∧ e x 4 e_F=e_{x_1}\wedge e_{x_2}\wedge e_{x_3}\wedge e_{x_4} e F = e x 1 ∧ e x 2 ∧ e x 3 ∧ e x 4 , with x 1 < x 2 < x 3 < x 4 x_1<x_2<x_3<x_4 x 1 < x 2 < x 3 < x 4 , for the sorted unit wedges of computational basis vectors, so that ⟨ e F , Λ 4 ( C ) e F ′ ⟩ = det C [ F , F ′ ] \langle e_F,\Lambda^4(C)e_{F'}\rangle=\det C[F,F'] ⟨ e F , Λ 4 ( C ) e F ′ ⟩ = det C [ F , F ′ ] . A four-set F F F with x 1 ⊕ x 2 ⊕ x 3 ⊕ x 4 = 0 x_1\oplus x_2\oplus x_3\oplus x_4=0 x 1 ⊕ x 2 ⊕ x 3 ⊕ x 4 = 0 is an affine plane, a coset of the two-dimensional subspace L = { 0 , x 1 ⊕ x 2 , x 1 ⊕ x 3 , x 1 ⊕ x 4 } L=\{0,x_1\oplus x_2,x_1\oplus x_3,x_1\oplus x_4\} L = { 0 , x 1 ⊕ x 2 , x 1 ⊕ x 3 , x 1 ⊕ x 4 } of F 2 n \mathbb{F}_2^n F 2 n . An h h h -edge is a pair { x , x ⊕ h } \{x,x\oplus h\} { x , x ⊕ h } , and e { x , x ⊕ h } e_{\{x,x\oplus h\}} e { x , x ⊕ h } is its wedge with the smaller string first. We now bound Q ( 14 ) ( N ) Q^{(14)}(N) Q ( 14 ) ( N ) and c F ( N 2 ) c_F(N^2) c F ( N 2 ) in terms of T T T and u u u .
Parity grading of Π F \Pi_F Π F . Since N N N commutes with P P P , Λ 4 H \Lambda^4\mathcal{H} Λ 4 H splits into the bidegrees ( k , 4 − k ) (k,4-k) ( k , 4 − k ) , spanned by wedges of k k k vectors from H + \mathcal{H}_+ H + and 4 − k 4-k 4 − k from H − \mathcal{H}_- H − , and Λ 4 ( N ) \Lambda^4(N) Λ 4 ( N ) is block diagonal in them. The next two lemmas show that Π F \Pi_F Π F lives on the pure bidegrees ( 4 , 0 ) (4,0) ( 4 , 0 ) and ( 0 , 4 ) (0,4) ( 0 , 4 ) and on the mixed bidegree ( 2 , 2 ) (2,2) ( 2 , 2 ) , and describe both parts.
Lemma S1 (Parity selection). If v 1 , … , v 4 v_1,\ldots,v_4 v 1 , … , v 4 are eigenvectors of P P P and an odd number of them lie in H − \mathcal{H}_- H − , then Π F ( v 1 ∧ v 2 ∧ v 3 ∧ v 4 ) = 0 \Pi_F(v_1\wedge v_2\wedge v_3\wedge v_4)=0 Π F ( v 1 ∧ v 2 ∧ v 3 ∧ v 4 ) = 0 . Hence Π F \Pi_F Π F vanishes on the bidegrees ( 3 , 1 ) (3,1) ( 3 , 1 ) and ( 1 , 3 ) (1,3) ( 1 , 3 ) .
Proof. Since P P P is itself a Pauli string, σ ↦ P σ \sigma\mapsto P\sigma σ ↦ P σ permutes P n \mathcal{P}_n P n up to phases whose fourth powers are one, so P ⊗ 4 Π W = Π W P^{\otimes4}\Pi_W=\Pi_W P ⊗ 4 Π W = Π W , and P ⊗ 4 P^{\otimes4} P ⊗ 4 acts as + 1 +1 + 1 on the range of Π F \Pi_F Π F . It acts as − 1 -1 − 1 on the wedge, which is therefore orthogonal to that range. □ \square □
Lemma S2 (Graded projector decomposition). The antisymmetric Pauli projector Π F \Pi_F Π F admits the following decomposition.
(i) Full projector. For each two-dimensional subspace L ⊆ F 2 n L\subseteq\mathbb{F}_2^n L ⊆ F 2 n , define
ψ L = 4 D ∑ F ∈ F 2 n / L e F . \psi_L=\sqrt{\frac{4}{D}}\sum_{F\in\mathbb{F}_2^n/L}e_F. ψ L = D 4 F ∈ F 2 n / L ∑ e F . where e F e_F e F is the sorted wedge associated with the four-element coset F F F . These vectors are orthonormal, and
Π F = ∑ dim L = 2 ∣ ψ L ⟩ ⟨ ψ L ∣ . \Pi_F=\sum_{\dim L=2}|\psi_L\rangle\langle\psi_L|. Π F = d i m L = 2 ∑ ∣ ψ L ⟩ ⟨ ψ L ∣. (ii) Pure parity bidegrees. If L ⊆ E L\subseteq E L ⊆ E , every coset of L L L lies entirely in E E E or O O O . Define
ψ L η = 4 D ∑ F ∈ F 2 n / L F ⊆ η e F , η ∈ { E , O } . \psi_L^\eta=\sqrt{\frac{4}{D}}\sum_{\substack{F\in\mathbb{F}_2^n/L\\F\subseteq\eta}}e_F,\qquad\eta\in\{E,O\}. ψ L η = D 4 F ∈ F 2 n / L F ⊆ η ∑ e F , η ∈ { E , O } . Then
ψ L = 2 − 1 / 2 ( ψ L E + ψ L O ) . \psi_L=2^{-1/2}(\psi_L^E+\psi_L^O). ψ L = 2 − 1/2 ( ψ L E + ψ L O ) . The compression of Π F \Pi_F Π F to Λ 4 H + \Lambda^4\mathcal{H}_+ Λ 4 H + is 1 2 Π D \frac{1}{2}\Pi_D 2 1 Π D , where
Π D = ∑ L ⊆ E dim L = 2 ∣ ψ L E ⟩ ⟨ ψ L E ∣ . \Pi_D=\sum_{\substack{L\subseteq E\\\dim L=2}}|\psi_L^E\rangle\langle\psi_L^E|. Π D = L ⊆ E d i m L = 2 ∑ ∣ ψ L E ⟩ ⟨ ψ L E ∣. This is the corresponding projector for n − 1 n-1 n − 1 modes, under the identification E ≅ F 2 n − 1 E\cong\mathbb{F}_2^{n-1} E ≅ F 2 n − 1 obtained by dropping the last bit. The same statement holds for Λ 4 H − \Lambda^4\mathcal{H}_- Λ 4 H − , using O O O in place of E E E .
(iii) Mixed parity bidegree. If L ⊈ E L\not\subseteq E L ⊆ E , each coset contains two even and two odd strings. Its contribution lies in the ( 2 , 2 ) (2,2) ( 2 , 2 ) bidegree, where
Π F m i x = ∑ i = 1 ( D / 2 2 ) ∣ f i ∧ g i ⟩ ⟨ f i ∧ g i ∣ . (43) \Pi_F^{\mathrm{mix}}=\sum_{i=1}^{\binom{D/2}{2}}|f_i\wedge g_i\rangle\langle f_i\wedge g_i|. \tag*{(43)} Π F mix = i = 1 ∑ ( 2 D /2 ) ∣ f i ∧ g i ⟩ ⟨ f i ∧ g i ∣. (43) Here { f i } \{f_i\} { f i } and { g i } \{g_i\} { g i } are matched real orthonormal bases of Λ 2 H + \Lambda^2\mathcal{H}_+ Λ 2 H + and Λ 2 H − \Lambda^2\mathcal{H}_- Λ 2 H − . Each basis vector is supported on a perfect matching of D / 2 D/2 D /2 disjoint edges, with coefficients ± q \pm\sqrt{q} ± q .
Explicitly, fix a 0 ∈ O a_0\in O a 0 ∈ O . For h ∈ E ∖ { 0 } h\in E\setminus\{0\} h ∈ E ∖ { 0 } and each character χ \chi χ of E / ⟨ h ⟩ E/\langle h\rangle E / ⟨ h ⟩ , the matched basis vectors are
f h , χ = q ∑ [ x ] ∈ E / ⟨ h ⟩ χ ( [ x ] ) e { x , x ⊕ h } , f_{h,\chi}=\sqrt{q}\sum_{[x]\in E/\langle h\rangle}\chi([x])e_{\{x,x\oplus h\}}, f h , χ = q [ x ] ∈ E / ⟨ h ⟩ ∑ χ ([ x ]) e { x , x ⊕ h } , g h , χ = q ∑ [ y ] ∈ O / ⟨ h ⟩ χ ( [ y ⊕ a 0 ] ) e { y , y ⊕ h } . (44) g_{h,\chi}=\sqrt{q}\sum_{[y]\in O/\langle h\rangle}\chi([y\oplus a_0])e_{\{y,y\oplus h\}}. \tag*{(44)} g h , χ = q [ y ] ∈ O / ⟨ h ⟩ ∑ χ ([ y ⊕ a 0 ]) e { y , y ⊕ h } . (44) Proof. (i) Projector. In Π W \Pi_W Π W , the Z ⊗ 4 Z^{\otimes4} Z ⊗ 4 require x 1 ⊕ ⋯ ⊕ x 4 = 0 x_1\oplus\cdots\oplus x_4=0 x 1 ⊕ ⋯ ⊕ x 4 = 0 ; thus a surviving four-set is a coset F F F of a two-dimensional subspace L L L . The operators X ⊗ 4 X^{\otimes4} X ⊗ 4 translate these cosets. They preserve the orientation of the sorted four-wedge: for each nonzero difference in an affine plane, the two corresponding unordered pairs have the same inversion parity under translation. The number of inversions is therefore even. Translation invariance gives the uniform vector ψ L \psi_L ψ L of Lemma S2; distinct L L L have disjoint supports, proving (i).
(ii) Pure parity bidegrees. If L ⊆ E L\subseteq E L ⊆ E , every coset has definite parity, with D / 4 D/4 D /4 cosets in each sector. Hence ψ L = 2 − 1 / 2 ( ψ L E + ψ L O ) \psi_L=2^{-1/2}(\psi_L^E+\psi_L^O) ψ L = 2 − 1/2 ( ψ L E + ψ L O ) . Deleting the last bit identifies either parity sector, with its ordering, with F 2 n − 1 \mathbb{F}_2^{n-1} F 2 n − 1 . Thus the two diagonal pure-sector compressions of Π F \Pi_F Π F are each 1 2 Π D \frac{1}{2}\Pi_D 2 1 Π D .
(iii) Mixed parity bidegrees. When L ⊈ E L\not\subseteq E L ⊆ E , there is a unique nonzero direction h ∈ L ∩ E h\in L\cap E h ∈ L ∩ E , and each coset consists of an even and an odd h h h -edge. To determine their relative wedge orientation, let k k k be the highest set bit of h h h , and let x , y x,y x , y be the smaller endpoints of the two edges. Both have zero bit k k k . With ℓ ≠ k \ell\ne k ℓ = k the highest set bit of x ⊕ y x\oplus y x ⊕ y , the sorting sign is
ε h , L = { + 1 , ℓ > k , ( − 1 ) 1 + h ℓ , ℓ < k . \varepsilon_{h,L}=
\begin{cases}
+1, & \ell>k,\\
(-1)^{1+h_\ell}, & \ell<k.
\end{cases} ε h , L = { + 1 , ( − 1 ) 1 + h ℓ , ℓ > k , ℓ < k . Indeed, for ℓ > k \ell>k ℓ > k the two oriented edges occur as separate blocks, while for ℓ < k \ell<k ℓ < k counting cross-edge inversions gives the stated parity. Moreover, x ⊕ y x\oplus y x ⊕ y is the representative of the odd coset { a , a ⊕ h } \{a,a\oplus h\} { a , a ⊕ h } with zero bit k k k . The sign therefore depends only on L L L , not on its coset. In the projector ∣ ψ L ⟩ ⟨ ψ L ∣ |\psi_L\rangle\langle\psi_L| ∣ ψ L ⟩ ⟨ ψ L ∣ the common sign cancels.
For fixed h h h , the mixed planes are indexed by the D / 2 D/2 D /2 elements of E / ⟨ h ⟩ E/\langle h\rangle E / ⟨ h ⟩ , translating a chosen odd direction a 0 a_0 a 0 . Their even and odd h h h -edge bases are paired by these translations. The Walsh transform on E / ⟨ h ⟩ E/\langle h\rangle E / ⟨ h ⟩ diagonalizes the pairing: ∑ t χ ( t ) χ ′ ( t ) = ( D / 2 ) δ χ , χ ′ \sum_t \chi(t)\chi'(t)=(D/2)\delta_{\chi,\chi'} ∑ t χ ( t ) χ ′ ( t ) = ( D /2 ) δ χ , χ ′ . This gives the matched vectors f h , χ f_{h,\chi} f h , χ , g h , χ g_{h,\chi} g h , χ and the mixed projector decomposition stated in the lemma. Different h h h have disjoint edge supports; for fixed h h h the Walsh characters are orthogonal. Since there are ( D − 1 ) D / 2 = ( D 2 ) (D-1)D/2=\binom{D}{2} ( D − 1 ) D /2 = ( 2 D ) such vectors, both matched families are orthonormal bases, as claimed.
Each flat basis vector f i f_i f i corresponds to a real skew-symmetric matrix S i S_i S i , defined by
f i = ∑ x , y ∈ E ( S i ) x y e x ∧ e y . f_i=\sum_{x,y\in E}(S_i)_{xy}e_x\wedge e_y. f i = x , y ∈ E ∑ ( S i ) x y e x ∧ e y .
Since its nonzero coefficients are ± q \pm\sqrt{q} ± q on the edges of a perfect matching, S i / q S_i/\sqrt{q} S i / q is an orthogonal signed permutation matrix. Hence S i T S i = q I D S_i^{\mathsf T}S_i=qI_D S i T S i = q I D , and for unit vectors v , w ∈ H + v,w\in H_+ v , w ∈ H + ,
∣ ⟨ f i , v ∧ w ⟩ ∣ 2 = ∣ v T S i w ∣ 2 ≤ q . (45) |\langle f_i,v\wedge w\rangle|^2=|v^{\mathsf T}S_iw|^2\le q. \tag*{(45)} ∣ ⟨ f i , v ∧ w ⟩ ∣ 2 = ∣ v T S i w ∣ 2 ≤ q . (45)
The same estimate holds for the odd-sector basis { g i } \{g_i\} { g i } .
Lemma S3 (Diagonal rank-four estimate). Let H H H be positive semidefinite of rank at most four on one parity sector. Denote its nonzero eigenvalues, padded with zeros, by h 1 , … , h 4 h_1,\ldots,h_4 h 1 , … , h 4 . Then, for the corresponding flat basis,
∑ i ⟨ f i , Λ 2 ( H ) f i ⟩ 2 ≤ q [ ∑ a < b ( h a h b ) 2 + 6 ∏ a = 1 4 h a ] . (46) \sum_i\langle f_i,\Lambda^2(H)f_i\rangle^2\le q\left[\sum_{a<b}(h_ah_b)^2+6\prod_{a=1}^{4}h_a\right]. \tag*{(46)} i ∑ ⟨ f i , Λ 2 ( H ) f i ⟩ 2 ≤ q [ a < b ∑ ( h a h b ) 2 + 6 a = 1 ∏ 4 h a ] . (46) The same inequality holds with g i g_i g i in place of f i f_i f i in the odd sector.
Proof. Let v 1 , … , v 4 v_1,\ldots,v_4 v 1 , … , v 4 be orthonormal eigenvectors of H H H and V = ( v 1 ⋯ v 4 ) V=(v_1\ \cdots\ v_4) V = ( v 1 ⋯ v 4 ) and let e = { a , b } e=\{a,b\} e = { a , b } range over the six edges of K 4 K_4 K 4 . Set
t a b = h a h b , y i ( a b ) = q − 1 ∣ ⟨ f i , v a ∧ v b ⟩ ∣ 2 . t_{ab}=h_ah_b,\qquad y_i(ab)=q^{-1}|\langle f_i,v_a\wedge v_b\rangle|^2. t ab = h a h b , y i ( ab ) = q − 1 ∣ ⟨ f i , v a ∧ v b ⟩ ∣ 2 .
Then y i ( a b ) ≥ 0 y_i(ab)\ge0 y i ( ab ) ≥ 0 and
⟨ f i , Λ 2 ( H ) f i ⟩ = q ∑ a < b t a b y i ( a b ) . \langle f_i,\Lambda^2(H)f_i\rangle=q\sum_{a<b}t_{ab}y_i(ab). ⟨ f i , Λ 2 ( H ) f i ⟩ = q a < b ∑ t ab y i ( ab ) .
Matching constraints. Represent f i f_i f i by its real skew matrix S i S_i S i , so that S i / q S_i/\sqrt{q} S i / q is an orthogonal signed-permutation matrix. The 4 × 4 4\times4 4 × 4 matrix V T S i V / q V^{\mathsf T}S_iV/\sqrt{q} V T S i V / q is a complex skew-symmetric contraction. Its row norms give the vertex constraints, while each 3 × 3 3\times3 3 × 3 principal skew contraction has singular values ( σ , σ , 0 ) (\sigma,\sigma,0) ( σ , σ , 0 ) , σ ≤ 1 \sigma\le1 σ ≤ 1 , giving the triangle constraints:
∑ b ≠ a y i ( a b ) ≤ 1 , ∑ { a , b } ⊂ U y i ( a b ) ≤ 1 ( ∣ U ∣ = 3 ) . \sum_{b\ne a}y_i(ab)\le1,\qquad\sum_{\{a,b\}\subset U}y_i(ab)\le1\quad(|U|=3). b = a ∑ y i ( ab ) ≤ 1 , { a , b } ⊂ U ∑ y i ( ab ) ≤ 1 ( ∣ U ∣ = 3 ) .
Together with nonnegativity these characterize the matching polytope of K 4 K_4 K 4 (Edmonds’ theorem [65 ] ). Thus
y i = ∑ M θ i M 1 M , y_i=\sum_M\theta_{iM}\mathbf{1}_M, y i = M ∑ θ i M 1 M ,
where M M M runs over the empty matching, single edges, and the three perfect matchings, with θ i M ≥ 0 \theta_{iM}\ge0 θ i M ≥ 0 and ∑ M θ i M = 1 \sum_M\theta_{iM}=1 ∑ M θ i M = 1 .
Summation over the flat basis. Completeness of { f i } \{f_i\} { f i } implies ∑ i y i ( e ) = q − 1 \sum_i y_i(e)=q^{-1} ∑ i y i ( e ) = q − 1 for each edge e e e . Put Θ M = ∑ i θ i M \Theta_M=\sum_i\theta_{iM} Θ M = ∑ i θ i M . Every edge e e e belongs to a unique perfect matching M = { e , e ′ } M=\{e,e'\} M = { e , e ′ } , so the aggregate weight of the singleton { e } \{e\} { e } is q − 1 − Θ M q^{-1}-\Theta_M q − 1 − Θ M . By Jensen’s inequality,
∑ i ( ∑ e t e y i ( e ) ) 2 ≤ ∑ M = { e , e ′ } p e r f e c t [ Θ M ( t e + t e ′ ) 2 + ( q − 1 − Θ M ) ( t e 2 + t e ′ 2 ) ] ≤ q − 1 ∑ M = { e , e ′ } p e r f e c t ( t e + t e ′ ) 2 = q − 1 [ ∑ a < b ( h a h b ) 2 + 6 ∏ a = 1 4 h a ] . \begin{aligned}
\sum_i\left(\sum_e t_ey_i(e)\right)^2
\le
\sum_{M=\{e,e'\}\ {\rm perfect}}
\left[\Theta_M(t_e+t_{e'})^2+(q^{-1}-\Theta_M)(t_e^2+t_{e'}^2)\right] \\
&\le q^{-1}\sum_{M=\{e,e'\}\ {\rm perfect}}(t_e+t_{e'})^2 \\
&=q^{-1}\left[\sum_{a<b}(h_ah_b)^2+6\prod_{a=1}^{4}h_a\right].
\end{aligned} i ∑ ( e ∑ t e y i ( e ) ) 2 ≤ M = { e , e ′ } perfect ∑ [ Θ M ( t e + t e ′ ) 2 + ( q − 1 − Θ M ) ( t e 2 + t e ′ 2 ) ] ≤ q − 1 M = { e , e ′ } perfect ∑ ( t e + t e ′ ) 2 = q − 1 [ a < b ∑ ( h a h b ) 2 + 6 a = 1 ∏ 4 h a ] .
The second inequality uses t e ≥ 0 t_e\ge0 t e ≥ 0 and 0 ≤ Θ M ≤ q − 1 0\le\Theta_M\le q^{-1} 0 ≤ Θ M ≤ q − 1 ; the last identity follows from t e t e ′ = ∏ a = 1 4 h a t_et_{e'}=\prod_{a=1}^{4}h_a t e t e ′ = ∏ a = 1 4 h a for every perfect matching. Multiplication by q 2 q^2 q 2 proves the lemma. The argument for the g i g_i g i basis is identical. □ \square □
Lemma S4 (Fourth-moment estimate). Let H H H be a positive semidefinite matrix acting on one parity sector, of rank four. Let ρ 1 , … , ρ 4 \rho_1,\ldots,\rho_4 ρ 1 , … , ρ 4 be its nonzero eigenvalues, let
τ : = ∏ a = 1 4 ρ a , T ( H ) : = ∑ a < b ( ρ a ρ b ) 4 , (47) \tau:=\prod_{a=1}^{4}\rho_a,\qquad T(H):=\sum_{a<b}(\rho_a\rho_b)^4, \tag*{(47)} τ := a = 1 ∏ 4 ρ a , T ( H ) := a < b ∑ ( ρ a ρ b ) 4 , (47) and let
X i j = ⟨ f i , Λ 2 ( H ) f j ⟩ . (48) X_{ij}=\langle f_i,\Lambda^2(H)f_j\rangle. \tag*{(48)} X ij = ⟨ f i , Λ 2 ( H ) f j ⟩ . (48) Then
∑ i , j ∣ X i j ∣ 4 ≤ q 2 [ T ( H ) + 18 τ 2 ] . (49) \sum_{i,j}|X_{ij}|^4\le q^2[T(H)+18\tau^2]. \tag*{(49)} i , j ∑ ∣ X ij ∣ 4 ≤ q 2 [ T ( H ) + 18 τ 2 ] . (49) The same statement holds in the odd sector with g i g_i g i .
Proof. Step 1: two flat columns. Let c , c ′ ∈ C ( D 2 ) c,c'\in\mathbb{C}^{\binom{D}{2}} c , c ′ ∈ C ( 2 D ) be orthonormal with ∣ c i ∣ 2 , ∣ c i ′ ∣ 2 ≤ q |c_i|^2,|c'_i|^2\le q ∣ c i ∣ 2 , ∣ c i ′ ∣ 2 ≤ q , and let α , β ≥ 0 \alpha,\beta\ge0 α , β ≥ 0 . We first establish
∑ i ∣ α c i + β c i ′ ∣ 4 ≤ q ( α 4 + β 4 + 6 α 2 β 2 ) . (50) \sum_i|\alpha c_i+\beta c'_i|^4\le q(\alpha^4+\beta^4+6\alpha^2\beta^2). \tag*{(50)} i ∑ ∣ α c i + β c i ′ ∣ 4 ≤ q ( α 4 + β 4 + 6 α 2 β 2 ) . (50) Put γ = α 2 \gamma=\alpha^2 γ = α 2 , γ ′ = β 2 \gamma'=\beta^2 γ ′ = β 2 , r i = ∣ c i ∣ 2 r_i=|c_i|^2 r i = ∣ c i ∣ 2 , r i ′ = ∣ c i ′ ∣ 2 r'_i=|c'_i|^2 r i ′ = ∣ c i ′ ∣ 2 , and ϱ i = Re ( c i c i ′ ‾ ) \varrho_i=\operatorname{Re}(c_i\overline{c'_i}) ϱ i = Re ( c i c i ′ ) . Orthogonality gives ∑ i ϱ i = 0 \sum_i\varrho_i=0 ∑ i ϱ i = 0 , while ∣ ϱ i ∣ ≤ r i r i ′ |\varrho_i|\le\sqrt{r_ir'_i} ∣ ϱ i ∣ ≤ r i r i ′ . Define the nonnegative deficits
Δ 1 = q − ∑ i r i 2 , Δ 2 = q − ∑ i ( r i ′ ) 2 , \Delta_1=q-\sum_i r_i^2,\qquad\Delta_2=q-\sum_i(r'_i)^2, Δ 1 = q − i ∑ r i 2 , Δ 2 = q − i ∑ ( r i ′ ) 2 , Δ 12 = q − ∑ i r i r i ′ ≥ 1 2 ( Δ 1 + Δ 2 ) ≥ Δ 1 Δ 2 . \Delta_{12}=q-\sum_i r_ir'_i\ge\frac12(\Delta_1+\Delta_2)\ge\sqrt{\Delta_1\Delta_2}. Δ 12 = q − i ∑ r i r i ′ ≥ 2 1 ( Δ 1 + Δ 2 ) ≥ Δ 1 Δ 2 .
With w i = γ ( q − r i ) + γ ′ ( q − r i ′ ) ≥ 0 w_i=\gamma(q-r_i)+\gamma'(q-r'_i)\ge0 w i = γ ( q − r i ) + γ ′ ( q − r i ′ ) ≥ 0 , weighted Cauchy–Schwarz gives
∣ ∑ i w i ϱ i ∣ 2 ≤ ( γ Δ 1 + γ ′ Δ 12 ) ( γ Δ 12 + γ ′ Δ 2 ) . \left|\sum_i w_i\varrho_i\right|^2\le(\gamma\Delta_1+\gamma'\Delta_{12})(\gamma\Delta_{12}+\gamma'\Delta_2). i ∑ w i ϱ i 2 ≤ ( γ Δ 1 + γ ′ Δ 12 ) ( γ Δ 12 + γ ′ Δ 2 ) .
Expanding the fourth powers and using ∑ i ϱ i = 0 \sum_i\varrho_i=0 ∑ i ϱ i = 0 therefore bounds the difference between the right- and left-hand sides of the displayed inequality from below by
x + y + 6 z − 4 ( x + z ) ( y + z ) , x+y+6z-4\sqrt{(x+z)(y+z)}, x + y + 6 z − 4 ( x + z ) ( y + z ) ,
where x = γ 2 Δ 1 x=\gamma^2\Delta_1 x = γ 2 Δ 1 , y = ( γ ′ ) 2 Δ 2 y=(\gamma')^2\Delta_2 y = ( γ ′ ) 2 Δ 2 , and z = γ γ ′ Δ 12 z=\gamma\gamma'\Delta_{12} z = γ γ ′ Δ 12 . Since z 2 ≥ x y z^2\ge xy z 2 ≥ x y , this is at least
( x + y + 2 z − 2 z ) 2 ≥ 0 , \left(\sqrt{x+y+2z}-2\sqrt{z}\right)^2\ge0, ( x + y + 2 z − 2 z ) 2 ≥ 0 ,
proving the claim.
Step 2: four-dimensional two-forms. Let V V V be four-dimensional with unit volume form ζ V \zeta_V ζ V . For t ∈ Λ 2 V t\in\Lambda^2V t ∈ Λ 2 V define t ∧ t = 2 Pf V ( t ) ζ V t\wedge t=2\operatorname{Pf}_V(t)\zeta_V t ∧ t = 2 Pf V ( t ) ζ V . The skew normal form is t = α v 1 ∧ v 2 + β v 3 ∧ v 4 t=\alpha v_1\wedge v_2+\beta v_3\wedge v_4 t = α v 1 ∧ v 2 + β v 3 ∧ v 4 with orthonormal v a v_a v a and α , β ≥ 0 \alpha,\beta\ge0 α , β ≥ 0 . The two columns c i = ⟨ f i , v 1 ∧ v 2 ⟩ c_i=\langle f_i,v_1\wedge v_2\rangle c i = ⟨ f i , v 1 ∧ v 2 ⟩ and c i ′ = ⟨ f i , v 3 ∧ v 4 ⟩ c'_i=\langle f_i,v_3\wedge v_4\rangle c i ′ = ⟨ f i , v 3 ∧ v 4 ⟩ are orthonormal and satisfy ∣ c i ∣ 2 , ∣ c i ′ ∣ 2 ≤ q |c_i|^2,|c'_i|^2\le q ∣ c i ∣ 2 , ∣ c i ′ ∣ 2 ≤ q by Eq. (45) . As ∥ t ∥ 2 = α 2 + β 2 \|t\|^2=\alpha^2+\beta^2 ∥ t ∥ 2 = α 2 + β 2 and ∣ Pf V ( t ) ∣ = α β |\operatorname{Pf}_V(t)|=\alpha\beta ∣ Pf V ( t ) ∣ = α β , Step 1 yields
∑ i ∣ ⟨ f i , t ⟩ ∣ 4 ≤ q ( ∥ t ∥ 4 + 4 ∣ Pf V ( t ) ∣ 2 ) . (51) \sum_i|\langle f_i,t\rangle|^4\le q(\|t\|^4+4|\operatorname{Pf}_V(t)|^2). \tag*{(51)} i ∑ ∣ ⟨ f i , t ⟩ ∣ 4 ≤ q ( ∥ t ∥ 4 + 4∣ Pf V ( t ) ∣ 2 ) . (51)
Step 3: wedge squares. For f = f h , χ f=f_{h,\chi} f = f h , χ , expand f ∧ f f\wedge f f ∧ f into pairs of distinct h h h -edges of E E E . Each such pair spans a plane L ⊂ E L\subset E L ⊂ E containing h h h ; the wedge-orientation sign is constant on its cosets by Lemma S2. For each fixed h h h , Walsh orthogonality shows that summing over the characters contributes twice the projector ∣ ψ L E ⟩ ⟨ ψ L E ∣ |\psi_L^E\rangle\langle\psi_L^E| ∣ ψ L E ⟩ ⟨ ψ L E ∣ for each plane L ∋ h L\ni h L ∋ h . Every plane contains three nonzero such directions, so
∑ i ∣ f i ∧ f i ⟩ ⟨ f i ∧ f i ∣ = 6 Π D . \sum_i|f_i\wedge f_i\rangle\langle f_i\wedge f_i|=6\Pi_D. i ∑ ∣ f i ∧ f i ⟩ ⟨ f i ∧ f i ∣ = 6 Π D .
Let z i z_i z i be the orthogonal projection of f i f_i f i onto Λ 2 V \Lambda^2V Λ 2 V . Since f i ∧ f i f_i\wedge f_i f i ∧ f i and z i ∧ z i z_i\wedge z_i z i ∧ z i have the same component along ζ V \zeta_V ζ V ,
∑ i ∣ Pf V ( z i ) ∣ 2 = 1 4 ∑ i ∣ ⟨ ζ V , f i ∧ f i ⟩ ∣ 2 = 3 2 ⟨ ζ V , Π D ζ V ⟩ ≤ 3 q . (52) \sum_i|\operatorname{Pf}_V(z_i)|^2=\frac14\sum_i|\langle\zeta_V,f_i\wedge f_i\rangle|^2
=\frac32\langle\zeta_V,\Pi_D\zeta_V\rangle\le3q. \tag*{(52)} i ∑ ∣ Pf V ( z i ) ∣ 2 = 4 1 i ∑ ∣ ⟨ ζ V , f i ∧ f i ⟩ ∣ 2 = 2 3 ⟨ ζ V , Π D ζ V ⟩ ≤ 3 q . (52)
The last inequality uses the four-dimensional compression bound ⟨ ζ V , Π D ζ V ⟩ ≤ 4 / D = 2 q \langle\zeta_V,\Pi_D\zeta_V\rangle\le4/D=2q ⟨ ζ V , Π D ζ V ⟩ ≤ 4/ D = 2 q . It follows from Theorem 22 of Ref. [28] for D ≥ 8 D\ge8 D ≥ 8 ; for D = 4 D=4 D = 4 it is immediate.
Step 4: conclusion. Take V = Ran H V=\operatorname{Ran}H V = Ran H , put t j = Λ 2 ( H ) f j ∈ Λ 2 V t_j=\Lambda^2(H)f_j\in\Lambda^2V t j = Λ 2 ( H ) f j ∈ Λ 2 V , and let z j z_j z j be the projection of f j f_j f j onto Λ 2 V \Lambda^2V Λ 2 V . Then X i j = ⟨ f i , t j ⟩ X_{ij}=\langle f_i,t_j\rangle X ij = ⟨ f i , t j ⟩ , ∥ t j ∥ 2 = ( X 2 ) j j \|t_j\|^2=(X^2)_{jj} ∥ t j ∥ 2 = ( X 2 ) j j , and Pf V ( t j ) = τ Pf V ( z j ) \operatorname{Pf}_V(t_j)=\tau\operatorname{Pf}_V(z_j) Pf V ( t j ) = τ Pf V ( z j ) . Step 2, Lemma S3 applied to H 2 H^2 H 2 , and the bound in Step 3 yield
∑ i , j ∣ X i j ∣ 4 ≤ q ∑ j [ ( X 2 ) j j ] 2 + 4 q τ 2 ∑ j ∣ Pf V ( z j ) ∣ 2 ≤ q 2 [ T ( H ) + 6 τ 2 ] + 12 q 2 τ 2 = q 2 [ T ( H ) + 18 τ 2 ] . \sum_{i,j}|X_{ij}|^4\le q\sum_j[(X^2)_{jj}]^2+4q\tau^2\sum_j|\operatorname{Pf}_V(z_j)|^2
\le q^2[T(H)+6\tau^2]+12q^2\tau^2
=q^2[T(H)+18\tau^2]. i , j ∑ ∣ X ij ∣ 4 ≤ q j ∑ [( X 2 ) j j ] 2 + 4 q τ 2 j ∑ ∣ Pf V ( z j ) ∣ 2 ≤ q 2 [ T ( H ) + 6 τ 2 ] + 12 q 2 τ 2 = q 2 [ T ( H ) + 18 τ 2 ] .
The odd sector is identical. □
Proof of Proposition S4. By Lemmas S1 and S2, the projector Π F \Pi_F Π F has a pure contribution on ( 4 , 0 ) ⊕ ( 0 , 4 ) (4,0)\oplus(0,4) ( 4 , 0 ) ⊕ ( 0 , 4 ) and a mixed contribution on ( 2 , 2 ) (2,2) ( 2 , 2 ) . Since Λ 4 ( N ) \Lambda^4(N) Λ 4 ( N ) preserves the parity bidegrees, c F ( N 2 ) c_F(N^2) c F ( N 2 ) and Q ( 14 ) ( N ) Q^{(14)}(N) Q ( 14 ) ( N ) split into pure and mixed terms, which we bound separately.
Mixed part. Define
X i j = ⟨ f i , Λ 2 ( A ) f j ⟩ , Y i j = ⟨ g i , Λ 2 ( B ) g j ⟩ . (53) X_{ij}=\langle f_i,\Lambda^2(A)f_j\rangle,\qquad Y_{ij}=\langle g_i,\Lambda^2(B)g_j\rangle. \tag*{(53)} X ij = ⟨ f i , Λ 2 ( A ) f j ⟩ , Y ij = ⟨ g i , Λ 2 ( B ) g j ⟩ . (53)
On the bidegree ( 2 , 2 ) (2,2) ( 2 , 2 ) , Λ 4 ( N ) ( ω ∧ η ) = Λ 2 ( A ) ω ∧ Λ 2 ( B ) η \Lambda^4(N)(\omega\wedge\eta)=\Lambda^2(A)\omega\wedge\Lambda^2(B)\eta Λ 4 ( N ) ( ω ∧ η ) = Λ 2 ( A ) ω ∧ Λ 2 ( B ) η for ω ∈ Λ 2 H + \omega\in\Lambda^2H_+ ω ∈ Λ 2 H + and η ∈ Λ 2 H − \eta\in\Lambda^2H_- η ∈ Λ 2 H − , and ω ⊗ η ↦ ω ∧ η \omega\otimes\eta\mapsto\omega\wedge\eta ω ⊗ η ↦ ω ∧ η is an isometry. Equation (S23) therefore gives
c F m i x ( N 2 ) = ∑ i ( X 2 ) i i ( Y 2 ) i i , c_F^{\mathrm{mix}}(N^2)=\sum_i(X^2)_{ii}(Y^2)_{ii}, c F mix ( N 2 ) = i ∑ ( X 2 ) ii ( Y 2 ) ii ,
Q ( 14 ) m i x ( N ) = ∑ i , j ∣ X i j Y i j ∣ 2 . (54) Q_{(14)}^{\mathrm{mix}}(N)=\sum_{i,j}|X_{ij}Y_{ij}|^2. \tag*{(54)} Q ( 14 ) mix ( N ) = i , j ∑ ∣ X ij Y ij ∣ 2 . (54)
By Lemma S3 applied to A 2 A^2 A 2 and Lemma S4 applied to A A A , together with (S17),
∑ i [ ( X 2 ) i i ] 2 ≤ q ( T + 6 u 2 ) , \sum_i[(X^2)_{ii}]^2\le q(T+6u^2), i ∑ [( X 2 ) ii ] 2 ≤ q ( T + 6 u 2 ) ,
∑ i , j ∣ X i j ∣ 4 ≤ q 2 ( T + 18 u 2 ) . (55) \sum_{i,j}|X_{ij}|^4\le q^2(T+18u^2). \tag*{(55)} i , j ∑ ∣ X ij ∣ 4 ≤ q 2 ( T + 18 u 2 ) . (55)
and the same estimates hold for Y Y Y , whose invariants u u u and T T T are the same. The Cauchy–Schwarz inequality then gives
c F m i x ( N 2 ) ≤ q ( T + 6 u 2 ) , Q ( 14 ) m i x ( N ) ≤ q 2 ( T + 18 u 2 ) . (56) c_F^{\mathrm{mix}}(N^2)\le q(T+6u^2),\qquad Q_{(14)}^{\mathrm{mix}}(N)\le q^2(T+18u^2). \tag*{(56)} c F mix ( N 2 ) ≤ q ( T + 6 u 2 ) , Q ( 14 ) mix ( N ) ≤ q 2 ( T + 18 u 2 ) . (56)
Pure part. Since A A A and B B B each have rank four,
Λ 4 ( A ) = u ∣ ζ A ⟩ ⟨ ζ A ∣ , Λ 4 ( B ) = u ∣ ζ B ⟩ ⟨ ζ B ∣ , (57) \Lambda^4(A)=u\,|\zeta_A\rangle\langle\zeta_A|,\qquad
\Lambda^4(B)=u\,|\zeta_B\rangle\langle\zeta_B|, \tag*{(57)} Λ 4 ( A ) = u ∣ ζ A ⟩ ⟨ ζ A ∣ , Λ 4 ( B ) = u ∣ ζ B ⟩ ⟨ ζ B ∣ , (57)
for unit volume forms ζ A ∈ Λ 4 Ran ( A ) \zeta_A\in\Lambda^4\operatorname{Ran}(A) ζ A ∈ Λ 4 Ran ( A ) and ζ B ∈ Λ 4 Ran ( B ) \zeta_B\in\Lambda^4\operatorname{Ran}(B) ζ B ∈ Λ 4 Ran ( B ) , and Λ 4 ( A 2 ) = u 2 ∣ ζ A ⟩ ⟨ ζ A ∣ \Lambda^4(A^2)=u^2|\zeta_A\rangle\langle\zeta_A| Λ 4 ( A 2 ) = u 2 ∣ ζ A ⟩ ⟨ ζ A ∣ , Λ 4 ( B 2 ) = u 2 ∣ ζ B ⟩ ⟨ ζ B ∣ \Lambda^4(B^2)=u^2|\zeta_B\rangle\langle\zeta_B| Λ 4 ( B 2 ) = u 2 ∣ ζ B ⟩ ⟨ ζ B ∣ . By Lemma S2(ii) and the compression bound of Eq. (52) ,
k A = ⟨ ζ A , Π F ζ A ⟩ = 1 2 ⟨ ζ A , Π D ζ A ⟩ ≤ q , k_A=\langle\zeta_A,\Pi_F\zeta_A\rangle=\frac12\langle\zeta_A,\Pi_D\zeta_A\rangle\le q, k A = ⟨ ζ A , Π F ζ A ⟩ = 2 1 ⟨ ζ A , Π D ζ A ⟩ ≤ q ,
and likewise k B ≤ q k_B\le q k B ≤ q . Hence
c F p u r e ( N 2 ) = u 2 ( k A + k B ) ≤ 2 q u 2 , c_F^{\mathrm{pure}}(N^2)=u^2(k_A+k_B)\le2qu^2, c F pure ( N 2 ) = u 2 ( k A + k B ) ≤ 2 q u 2 ,
Q ( 14 ) p u r e ( N ) = u 2 ( k A 2 + k B 2 + 2 ∣ ⟨ ζ A , Π F ζ B ⟩ ∣ 2 ) ≤ u 2 ( k A + k B ) 2 ≤ 4 q 2 u 2 . (58) Q_{(14)}^{\mathrm{pure}}(N)=u^2(k_A^2+k_B^2+2|\langle\zeta_A,\Pi_F\zeta_B\rangle|^2)
\le u^2(k_A+k_B)^2\le4q^2u^2. \tag*{(58)} Q ( 14 ) pure ( N ) = u 2 ( k A 2 + k B 2 + 2∣ ⟨ ζ A , Π F ζ B ⟩ ∣ 2 ) ≤ u 2 ( k A + k B ) 2 ≤ 4 q 2 u 2 . (58)
where ∣ ⟨ ζ A , Π F ζ B ⟩ ∣ 2 ≤ k A k B |\langle\zeta_A,\Pi_F\zeta_B\rangle|^2\le k_Ak_B ∣ ⟨ ζ A , Π F ζ B ⟩ ∣ 2 ≤ k A k B by the Cauchy–Schwarz inequality.
Conclusion. Adding (S36) and (S38), we obtain
c F ( N 2 ) ≤ q ( T + 8 u 2 ) , Q ( 14 ) ( N ) ≤ q 2 ( T + 22 u 2 ) . (59) c_F(N^2)\le q(T+8u^2),\qquad Q_{(14)}(N)\le q^2(T+22u^2). \tag*{(59)} c F ( N 2 ) ≤ q ( T + 8 u 2 ) , Q ( 14 ) ( N ) ≤ q 2 ( T + 22 u 2 ) . (59)
Substituting into (S22), with d q = 4 dq=4 d q = 4 , gives
W ( ϕ N ) ≤ I ( μ ) + 168 T + 2688 u 2 . (60) W(\phi_N)\le I(\mu)+168T+2688u^2. \tag*{(60)} W ( ϕ N ) ≤ I ( μ ) + 168 T + 2688 u 2 . (60)
It remains to identify the right-hand side with the canonical BCS value. Let u j = p j q j u_j=p_jq_j u j = p j q j . Then u 2 = u 1 2 u 2 2 u 3 2 u^2=u_1^2u_2^2u_3^2 u 2 = u 1 2 u 2 2 u 3 2 , ∑ s λ s 2 = ∏ j ( 1 − 2 u j ) \sum_s\lambda_s^2=\prod_j(1-2u_j) ∑ s λ s 2 = ∏ j ( 1 − 2 u j ) , ∑ s λ s 4 = ∏ j ( 1 − 4 u j + 2 u j 2 ) \sum_s\lambda_s^4=\prod_j(1-4u_j+2u_j^2) ∑ s λ s 4 = ∏ j ( 1 − 4 u j + 2 u j 2 ) , and, from the pair products below Eq. (37) , T = ∑ l ( 1 − 4 u l + 2 u l 2 ) ∏ k ≠ l u k 2 T=\sum_l(1-4u_l+2u_l^2)\prod_{k\ne l}u_k^2 T = ∑ l ( 1 − 4 u l + 2 u l 2 ) ∏ k = l u k 2 . Expanding in the u j u_j u j gives
I ( μ ) + 168 T + 2688 u 2 = ∏ j = 1 3 [ 1 − 4 u j + 16 u j 2 ] = ∏ j = 1 3 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] . (61) \begin{aligned}
I(\mu)+168T+2688u^2 \\
&=\prod_{j=1}^{3}\left[1-4u_j+16u_j^2\right] \\
&=\prod_{j=1}^{3}\left[1-4p_jq_j(1-2p_j)^2\right]. \tag*{(61)}
\end{aligned} I ( μ ) + 168 T + 2688 u 2 = j = 1 ∏ 3 [ 1 − 4 u j + 16 u j 2 ] = j = 1 ∏ 3 [ 1 − 4 p j q j ( 1 − 2 p j ) 2 ] . (61)
But the product on the right is precisely 2 − M 2 ( p ) 2^{-\mathcal{M}_2(p)} 2 − M 2 ( p ) , i.e. the stabilizer purity of the canonical BCS representative. Hence
W ( ϕ N ) ≤ 2 − M 2 ( p ) . (62) W(\phi_N)\le2^{-\mathcal{M}_2(p)}. \tag*{(62)} W ( ϕ N ) ≤ 2 − M 2 ( p ) . (62)
This is equivalent to
M 2 ( ϕ N ) ≥ M 2 ( p ) . (63) \mathcal{M}_2(\phi_N)\ge\mathcal{M}_2(p). \tag*{(63)} M 2 ( ϕ N ) ≥ M 2 ( p ) . (63)
Since the BCS state attains M 2 ( p ) \mathcal{M}_2(p) M 2 ( p ) , the minimum over the full parity-even local unitary orbit is exactly M 2 ( p ) \mathcal{M}_2(p) M 2 ( p ) . This proves the proposition for m = 3 m=3 m = 3 in the interior case. For fixed U ∈ G U\in G U ∈ G both sides of W ( ϕ N ) ≤ W ( ϕ M ) W(\phi_N)\le W(\phi_M) W ( ϕ N ) ≤ W ( ϕ M ) are continuous in the occupations, so the bound extends to p j ∈ { 0 , 1 } p_j\in\{0,1\} p j ∈ { 0 , 1 } , and the reductions above then give m = 2 m=2 m = 2 and m = 1 m=1 m = 1 as well.
All bounds in the proof are saturated at N = M N=M N = M . The three-mode cube supporting M M M contains fourteen affine planes. Two consist of the even and odd Fock strings, each with amplitude product u u u . The other twelve arise from the three perfect matchings of the four even strings. For a matching with edge products t , t ′ t,t' t , t ′ satisfying t t ′ = u tt'=u t t ′ = u , one direction gives two planes with products t 2 , ( t ′ ) 2 t^2,(t')^2 t 2 , ( t ′ ) 2 , and another gives two planes with product u u u .
Four and more pairs
The first failure of Gaussian optimality occurs at m = 4 m=4 m = 4 pairs. An explicit witness is a parity-even exchange of two Fock states.
Proposition S5 (Four equal pairs). Let m = 4 m=4 m = 4 , p j = p p_j=p p j = p , q = 1 − p q=1-p q = 1 − p , and let Π \Pi Π exchange ∣ 1000 ⟩ |1000\rangle ∣1000 ⟩ and ∣ 0111 ⟩ |0111\rangle ∣0111 ⟩ on each side, acting as the identity on all other Fock states. If ϕ s \phi_s ϕ s is the product-labeled reference state from § 10 A, then
( Π ⊗ Π ) ϕ s = W G + Δ W , (\Pi\otimes\Pi)\phi_s=W_G+\Delta W, ( Π ⊗ Π ) ϕ s = W G + Δ W ,
Δ W = 168 ( p q ) 4 ( 1 − 2 p ) 4 [ 2 − ( 1 − 2 p ) 4 ] , (64) \Delta W=168(pq)^4(1-2p)^4[2-(1-2p)^4], \tag*{(64)} Δ W = 168 ( pq ) 4 ( 1 − 2 p ) 4 [ 2 − ( 1 − 2 p ) 4 ] , (64)
W G = 2 − M 2 ( p ) = [ 1 − 4 p q ( 1 − 2 p ) 2 ] 4 . W_G=2^{-\mathcal{M}_2(p)}=[1-4pq(1-2p)^2]^4. W G = 2 − M 2 ( p ) = [ 1 − 4 pq ( 1 − 2 p ) 2 ] 4 .
Therefore M 2 f , N L ( Ψ 4 ) < M 2 ( p ) \mathcal{M}_2^{\mathrm{f,NL}}(\Psi_4)<\mathcal{M}_2(p) M 2 f , NL ( Ψ 4 ) < M 2 ( p ) whenever p ∉ { 0 , 1 2 , 1 } p\notin\{0,\frac12,1\} p ∈ / { 0 , 2 1 , 1 } .
Proof. Both exchanged strings have odd parity, so Π \Pi Π is an allowed parity-even local unitary. The product-labeled state is related to the physical BCS state by a local parity-even Clifford (§ 10 A). It therefore suffices to compute the stabilizer purity of the permuted reference state.
Here is the finite-sum identity needed for this computation. For real amplitudes v x v_x v x of a reference state ϕ v = ∑ x v x ∣ x ⟩ A ∣ x ⟩ B \phi_v=\sum_xv_x|x\rangle_A|x\rangle_B ϕ v = ∑ x v x ∣ x ⟩ A ∣ x ⟩ B , extended by zero outside its support, write W ( ϕ v ) = 2 − M 2 ( ϕ v ) W(\phi_v)=2^{-\mathcal{M}_2(\phi_v)} W ( ϕ v ) = 2 − M 2 ( ϕ v ) . Then
W ( ϕ v ) = ∑ s ∈ F 2 n ∑ x 1 ⊕ x 2 ⊕ x 3 ⊕ x 4 = 0 ∏ i = 1 4 v x i v x i ⊕ s . (65) W(\phi_v)=\sum_{s\in\mathbb{F}_2^n}\sum_{x_1\oplus x_2\oplus x_3\oplus x_4=0}\prod_{i=1}^{4}v_{x_i}v_{x_i\oplus s}. \tag*{(65)} W ( ϕ v ) = s ∈ F 2 n ∑ x 1 ⊕ x 2 ⊕ x 3 ⊕ x 4 = 0 ∑ i = 1 ∏ 4 v x i v x i ⊕ s . (65)
Indeed, the Pauli expectation on ϕ v \phi_v ϕ v vanishes unless both parties have the same bit-flip string s s s . For fixed s s s , the remaining Pauli phases give Walsh transforms of g s ( x ) = v x v x ⊕ s g_s(x)=v_xv_{x\oplus s} g s ( x ) = v x v x ⊕ s ; expanding their fourth powers and using Walsh orthogonality gives Eq. (65) . Equivalently,
W ( ϕ v ) = 2 − n ∑ s , b ∈ F 2 n [ ∑ x ( − 1 ) b ⋅ x v x v x ⊕ s ] 4 . W(\phi_v)=2^{-n}\sum_{s,b\in\mathbb{F}_2^n}\left[\sum_x(-1)^{b\cdot x}v_xv_{x\oplus s}\right]^4. W ( ϕ v ) = 2 − n s , b ∈ F 2 n ∑ [ x ∑ ( − 1 ) b ⋅ x v x v x ⊕ s ] 4 .
For four identical pairs the unswapped amplitudes are v x = q ( 4 − ∣ x ∣ ) / 2 p ∣ x ∣ / 2 v_x=q^{(4-|x|)/2}p^{|x|/2} v x = q ( 4 − ∣ x ∣ ) /2 p ∣ x ∣/2 . Let v ′ v' v ′ be obtained by interchanging v 1000 v_{1000} v 1000 and v 0111 v_{0111} v 0111 , and put u = p q u=pq u = pq . Substituting the 16 amplitudes into the displayed finite sum and collecting powers of u u u gives the exact polynomial difference
Δ W = W ( ϕ v ′ ) − W ( ϕ v ) = 168 u 4 − 10752 u 6 + 43008 u 7 − 43008 u 8 = 168 u 4 ( 1 − 4 u ) 2 [ 2 − ( 1 − 4 u ) 2 ] . \begin{aligned}
\Delta W=W(\phi_{v'})-W(\phi_v) \\
&=168u^4-10752u^6+43008u^7-43008u^8 \\
&=168u^4(1-4u)^2[2-(1-4u)^2].
\end{aligned} Δ W = W ( ϕ v ′ ) − W ( ϕ v ) = 168 u 4 − 10752 u 6 + 43008 u 7 − 43008 u 8 = 168 u 4 ( 1 − 4 u ) 2 [ 2 − ( 1 − 4 u ) 2 ] .
Since 1 − 4 u = ( 1 − 2 p ) 2 1-4u=(1-2p)^2 1 − 4 u = ( 1 − 2 p ) 2 and W ( ϕ v ) = [ 1 − 4 u ( 1 − 2 p ) 2 ] 4 W(\phi_v)=[1-4u(1-2p)^2]^4 W ( ϕ v ) = [ 1 − 4 u ( 1 − 2 p ) 2 ] 4 , this is precisely Eq. (64) . The difference is strictly positive for 0 < p < 1 0<p<1 0 < p < 1 , p ≠ 1 2 p\ne\frac12 p = 2 1 , proving the claim.
The exchange is non-Gaussian. At p = 1 4 p=\frac14 p = 4 1 it gives M 2 = 1.1173 < M 2 = 1.1982 \mathcal{M}_2=1.1173<\mathcal{M}_2=1.1982 M 2 = 1.1173 < M 2 = 1.1982 . The resulting graded reference state places its larger Schmidt amplitudes on the Fock labels associated with the optimal flat-spectrum target for the w = 1 w=1 w = 1 class discussed in Sec. III C of the main text.
Corollary S1 (From four pairs on). (i) For every m ≥ 4 m\ge4 m ≥ 4 , Gaussian local unitaries fail to minimize M 2 \mathcal{M}_2 M 2 on an open set of BCS occupations.
(ii) If the occupations are independent with common distribution ν \nu ν with ν ( { 0 , 1 2 , 1 } ) < 1 \nu(\{0,\frac12,1\})<1 ν ({ 0 , 2 1 , 1 }) < 1 , then
Pr ( Gaussian optimality ) ≤ ( 1 − π ν ) ⌊ m / 4 ⌋ , \Pr(\text{Gaussian optimality})\le(1-\pi_\nu)^{\lfloor m/4\rfloor}, Pr ( Gaussian optimality ) ≤ ( 1 − π ν ) ⌊ m /4 ⌋ ,
where π ν > 0 \pi_\nu>0 π ν > 0 is the probability that some complementary four-pair exchange x ↔ x ⊕ 1111 x\leftrightarrow x\oplus1111 x ↔ x ⊕ 1111 improves on the Gaussian value.
Proof. Apply the same two-sided exchange to any four selected pairs and the identity to all remaining modes. Additivity of M 2 M_2 M 2 on tensor products preserves the strict improvement in Proposition S5; continuity makes it persist in a neighborhood of four equal nontrivial occupations, proving (i). For (ii), choose a nontrivial p 0 p_0 p 0 in the support of ν \nu ν . Four independent draws all sufficiently near p 0 p_0 p 0 have positive probability and form a successful block. Among ⌊ m / 4 ⌋ \lfloor m/4\rfloor ⌊ m /4 ⌋ disjoint blocks, Gaussian optimality requires every block to fail the improvement test. Independence gives the stated probability bound. □
Figure 3(d) of the main text illustrates the Gaussian-optimality threshold for identical BCS pairs with p j = 1 4 p_j=\frac{1}{4} p j = 4 1 . Local Gaussian unitaries are optimal for m ≤ 3 m\leq3 m ≤ 3 (Proposition S4), whereas the graded reference states shown in the figure achieve smaller SRE from m = 4 m=4 m = 4 onward.
Particle-number-conserving local unitaries
Local particle-number conservation is more restrictive than parity conservation. Consequently, D min f , N L ≤ D min U ( 1 ) , N L D_{\min}^{f,\mathrm{NL}}\leq D_{\min}^{U(1),\mathrm{NL}} D m i n f , NL ≤ D m i n U ( 1 ) , NL and M min f , N L ≤ M min U ( 1 ) , N L M_{\min}^{f,\mathrm{NL}}\leq M_{\min}^{U(1),\mathrm{NL}} M m i n f , NL ≤ M m i n U ( 1 ) , NL . Here M min U ( 1 ) , N L M_{\min}^{U(1),\mathrm{NL}} M m i n U ( 1 ) , NL denotes the minimum SRE under local unitaries conserving N A N_A N A and N B N_B N B . The quantity Φ S U ( 1 ) \Phi_S^{U(1)} Φ S U ( 1 ) is the maximum squared overlap with stabilizer targets of definite total particle number under the same local unitaries, and D min U ( 1 ) , N L = − log 2 Φ S U ( 1 ) D_{\min}^{U(1),\mathrm{NL}}=-\log_2\Phi_S^{U(1)} D m i n U ( 1 ) , NL = − log 2 Φ S U ( 1 ) .
Proposition S6 (Number-conserving conversion of identical pairs). Let
ψ = ⨂ j = 1 m ( q ∣ 10 ⟩ + p ∣ 01 ⟩ ) \psi=\bigotimes_{j=1}^{m}\left(\sqrt{q}|10\rangle+\sqrt{p}|01\rangle\right) ψ = j = 1 ⨂ m ( q ∣10 ⟩ + p ∣01 ⟩ )
consist of m m m identical number-conserving entangled pairs. Then the optimal fidelity under local particle-number-conserving unitaries, with stabilizer targets of definite total particle number, is
Φ S U ( 1 ) = [ max { p , q , 1 2 + p q } ] m = [ Φ S N L ( m = 1 ) ] m (66) \Phi_S^{U(1)}=\left[\max\left\{p,q,\frac{1}{2}+\sqrt{pq}\right\}\right]^m=\left[\Phi_S^{\mathrm{NL}}(m=1)\right]^m \tag*{(66)} Φ S U ( 1 ) = [ max { p , q , 2 1 + pq } ] m = [ Φ S NL ( m = 1 ) ] m (66)
The maximum is attained by products of the one-pair stabilizer states ∣ 10 ⟩ |10\rangle ∣10 ⟩ , ∣ 01 ⟩ |01\rangle ∣01 ⟩ , and ( ∣ 10 ⟩ + ∣ 01 ⟩ ) / 2 (|10\rangle+|01\rangle)/\sqrt{2} ( ∣10 ⟩ + ∣01 ⟩) / 2 . Hence D min U ( 1 ) , N L = m D min f , N L ( m = 1 ) D_{\min}^{U(1),\mathrm{NL}}=mD_{\min}^{f,\mathrm{NL}}(m=1) D m i n U ( 1 ) , NL = m D m i n f , NL ( m = 1 ) .
The fidelity is unchanged by filled or empty spectator modes on A A A and empty spectator modes on B B B for p ≤ 1 2 p\leq\frac{1}{2} p ≤ 2 1 ; for p > 1 2 p>\frac{1}{2} p > 2 1 , exchange the two sides.
Proof. Exchange the two sides if necessary so p ≤ q p\leq q p ≤ q and set t = p / q t=\sqrt{p/q} t = p / q . The ( m w ) \binom{m}{w} ( w m ) amplitudes in charge sector N B = w N_B=w N B = w are all q m / 2 t w q^{m/2}t^w q m /2 t w . Let ∣ σ ⟩ |\sigma\rangle ∣ σ ⟩ be a stabilizer target of Schmidt rank 2 k 2^k 2 k . Its reduced state satisfies ρ B = 2 − k Π V \rho_B=2^{-k}\Pi_V ρ B = 2 − k Π V , whose computational-basis diagonal is uniform on an affine binary coset x 0 + V x_0+V x 0 + V , with dim V ≥ k \dim V\geq k dim V ≥ k : this follows from the Z Z Z -type part of the stabilizer expansion of ρ B \rho_B ρ B . Because the target has definite particle number, the dimension of its N B = w N_B=w N B = w Schmidt sector is
m w = 2 k Pr v ∈ V ( ∣ x 0 ⊕ v ∣ = w ) . m_w=2^k\Pr_{v\in V}(|x_0\oplus v|=w). m w = 2 k v ∈ V Pr ( ∣ x 0 ⊕ v ∣ = w ) .
Sectorwise Schmidt alignment therefore gives
Φ ( ψ , σ ) ≤ 2 − k q m [ ∑ w min { m w , ( m w ) t w } ] 2 (67) \Phi(\psi,\sigma)\leq2^{-k}q^m\left[\sum_w\min\left\{m_w,\binom{m}{w}t^w\right\}\right]^2 \tag*{(67)} Φ ( ψ , σ ) ≤ 2 − k q m [ w ∑ min { m w , ( w m ) t w } ] 2 (67)
For k ≤ m k\leq m k ≤ m , replacing the minimum by m w m_w m w turns this into 2 k q m ( E V t ∣ x 0 ⊕ v ∣ ) 2 2^kq^m(\mathbb{E}_Vt^{|x_0\oplus v|})^2 2 k q m ( E V t ∣ x 0 ⊕ v ∣ ) 2 . Select k k k independent coordinate functionals on V V V ; their values are independent fair bits. Discarding the other factors (each at most one, as t ≤ 1 t\leq1 t ≤ 1 ) shows Φ ( ψ , σ ) ≤ q m [ ( 1 + t ) 2 / 2 ] k \Phi(\psi,\sigma)\leq q^m[(1+t)^2/2]^k Φ ( ψ , σ ) ≤ q m [( 1 + t ) 2 /2 ] k , which is at most [ max { q , 1 2 + p q } ] m \left[\max\{q,\frac{1}{2}+\sqrt{pq}\}\right]^m [ max { q , 2 1 + pq } ] m . For k > m k>m k > m , instead replace the minimum in (S47) by ( m w ) \binom{m}{w} ( w m ) and sum the binomial expansion, giving Φ ≤ 2 − k q m ( 1 + t ) 2 m ≤ ( 1 2 + p q ) m \Phi\leq2^{-k}q^m(1+t)^{2m}\leq(\frac{1}{2}+\sqrt{pq})^m Φ ≤ 2 − k q m ( 1 + t ) 2 m ≤ ( 2 1 + pq ) m . Product stabilizers attain the largest of the one-pair alternatives. The same cost argument accommodates the stated spectator modes. □