betaFeedback

WorksPhysics and Astronomy

The magical memory cost of entanglement spectrum compression

Abstract

How much quantum memory is needed to faithfully retain the entanglement spectrum of a bipartite pure state after separating out Bell pairs through local unitaries? We derive the exact tradeoff between Bell-pair yield, residual memory, and reconstruction fidelity under local unitaries, defining an operational measure of entanglement-spectrum complexity. Smoothed entanglement spread bounds this memory cost from below, while diverging nonlocal magic forces unbounded memory at any fixed positive fidelity. Chaotic circuit dynamics exhibit a transient memory peak whose height increases with system size, followed by declining cost despite continued entropy growth. A large typical Haar-random state requires only an accuracy-dependent memory, whereas stationary fermionic Gaussian states retain a growing cost with size. In the circuits studied, the multiscale spectral structure associated with entanglement catalysis also has a practical consequence for matrix product state compression, yielding a multiplicative overhead not captured by the conventional entropy-based bond dimension estimate.

Introduction

Random quantum dynamics can generate volume-law entanglement from product states [1, 2], raising questions about both the resources needed to represent the resulting states and the entanglement transformations they support. These are closely related through the Schmidt probabilities, the eigenvalues of a reduced density matrix that form the entanglement spectrum [3, 4]. Their cumulative weight determines the bond dimension required for a matrix product state approximation across a cut at a prescribed fidelity [5, 6, 7], while their distribution constrains how entanglement can be manipulated by local operations [8, 9, 10, 11, 12, 13]. For many identical copies, entropy sets the leading rates of quantum source coding and entanglement concentration into Bell pairs [14, 15, 16], while fluctuations of the Schmidt weights introduce additional resource requirements in coherent concentration and dilution [17, 18, 19, 20]. This motivates asking how much of a single state’s entanglement can be coherently compressed into Bell pairs, and how much residual quantum memory must be retained for faithful reconstruction.

For a single copy, local unitaries preserve every Schmidt probability. Bell pairs contribute a uniform factor to the spectrum, so an exact decomposition into Bell pairs and a residual state must retain all spectral nonuniformity in that state. In particular, bb Bell pairs have Schmidt rank 2b2^b and entanglement entropy S1=bS_1=b, yet require no residual. At a fixed reconstruction accuracy, the minimum memory required for such coherent compression thus provides an operational measure of entanglement-spectrum complexity. This residual memory is closely related to the notion of nonlocal magic [21, 22]. While nonstabilizerness [23] is studied as a resource in quantum information and computing [24, 25, 26, 27], and as a way to characterize many-body states and dynamics [28, 29, 30, 31, 32, 33], minimizing it over local changes of basis isolates the contribution associated with shared entanglement. Bipartite pure stabilizer states provide a zero-memory reference, as local Clifford unitaries transform them into Bell pairs and local product states [34]. For general pure states, the fidelity-based measure of nonlocal magic is determined entirely by the entanglement spectrum [35, 36]. Recent studies of random quantum dynamics have further connected near scale-free entanglement spectra to diverging nonlocal magic and universal entanglement embezzlement [37, 38, 39], which allows any fixed entangled target to be borrowed while returning the shared catalyst with vanishing disturbance [12, 40]. These findings further motivate a quantitative formulation of the relation between catalytic capabilities and spectral compressibility.

Here we derive the exact tradeoff between Bell-pair yield, residual memory, and reconstruction fidelity for arbitrary bipartite pure states under unrestricted local unitaries. We show that diverging nonlocal magic forces unbounded residual memory at any fixed positive fidelity, and smoothed entanglement spread provides an accuracy-dependent lower bound. Simulations of chaotic circuits reveal a transient memory peak that recedes while entanglement entropy continues to rise. Around this peak and at fixed fidelity, the near multiscale spectral structure imposes a multiplicative overhead beyond the entropy-based bond dimension estimate 2S12^{S_1}. In contrast, a large Haar-typical state asymptotically admits an accuracy-dependent, size-independent remainder, whereas stationary fermionic Gaussian states retain a cost that grows with size. In particular, every universally embezzling family requires unbounded residual memory at any fixed positive fidelity, establishing an obstruction to representing such families through Bell pairs and a bounded residual.

Coherent entanglement spectrum compression. Consider a (single copy of) pure bipartite state ∣ψ⟩AB\lvert\psi\rangle_{AB} of N=2nN = 2n qubits, with nn qubits held by each party AA and BB. Our goal is to faithfully compress the state using local unitaries by separating its maximally entangled Bell-pair structure from a residual register that carries the remaining nonflat entanglement spectrum. We can write the Schmidt decomposition as ∣ψ⟩AB=∑i=02n−1λi∣ai⟩A∣bi⟩B|\psi\rangle_{AB} = \sum_{i=0}^{2^n-1} \sqrt{\lambda_i}|a_i\rangle_A|b_i\rangle_B, with λ0≥λ1≥⋯≥0\lambda_0 \ge\lambda_1 \ge\cdots\ge0 and zero padding when the Schmidt rank is smaller than 2n2^n. For an allocation (b,k)(b,k) with b+k≤nb+k \le n, we divide each party’s nn qubits into a Bell register of bb qubits, a residual of kk qubits, and r=n−b−kr=n-b-k qubits in a fixed pure product state. The desired factorized approximation is

(U^A⊗U^B)∣ψ⟩AB≈∣Ωb,k⟩,(\hat{U}_A \otimes\hat{U}_B)|\psi\rangle_{AB} \approx|\Omega_{b,k}\rangle,
∣Ωb,k⟩:=∣Φ2⟩AbBb⊗∣γk⟩AkBk⊗∣0⟩Ar⊗∣0⟩Br,(1)|\Omega_{b,k}\rangle:= |\Phi_2\rangle_{A_bB_b} \otimes|\gamma_k\rangle_{A_kB_k} \otimes|0\rangle_{A_r} \otimes|0\rangle_{B_r}, \tag*{(1)}

where ∣Φ2⟩:=(∣00⟩+∣11⟩)/2|\Phi_2\rangle:= (|00\rangle+ |11\rangle)/\sqrt{2} is a symmetric Bell state, and ∣γk⟩|\gamma_k\rangle is an arbitrary normalized bipartite pure state. More generally, ∣Φ2⟩|\Phi_2\rangle may be replaced by any maximally entangled two-qubit state. The state-dependent unitary encoders U^A,B\hat{U}_{A,B} act on their entire subsystems with unrestricted complexity. With a sufficient initial supply of Bell pairs, the same representation may also be seen as a preparation protocol [18, 41, 42].

The separation above distinguishes standardized entanglement from the nonuniform spectral structure that must remain in a quantum register. For given bb and kk, the optimal reconstruction fidelity is Fb,k(ψ):=max⁡U^A,U^B,γk∣⟨Ωb,k(γ)∣U^A⊗U^B∣ψ⟩∣2F_{b,k}(\psi) := \max_{\hat{U}_A,\hat{U}_B,\gamma_k} |\langle\Omega_{b,k}(\gamma)|\hat{U}_A \otimes\hat{U}_B|\psi\rangle|^2. The optimization can be performed explicitly [43, 44], giving the exact form

Fb,k(ψ)=2−b∑j=02k−1(∑s=02b−1λj2b+s)2.(2)F_{b,k}(\psi) = 2^{-b}\sum_{j=0}^{2^k-1}\left(\sum_{s=0}^{2^b-1}\sqrt{\lambda_{j2^b+s}}\right)^2. \tag*{(2)}

See Supplemental Material (SM) for further details. For a prescribed kk, the maximal fidelity over the Bell yields is Fk:=max⁡0≤b≤n−kFb,kF_k := \max_{0\le b\le n-k} F_{b,k}. For a given error tolerance ϵ\epsilon, we define the minimum memory cost as

kϵ:=min⁡{k:Fk≥1−ϵ}.(3)k_\epsilon:= \min\{k:F_k \ge1-\epsilon\}. \tag*{(3)}

At this minimum, the largest Bell yield compatible with the target fidelity is bϵ:=max⁡{0≤b≤n−kϵ:Fb,kϵ≥1−ϵ}b_\epsilon:= \max\{0 \le b \le n-k_\epsilon:F_{b,k_\epsilon} \ge1-\epsilon\}. The pair (bϵ,kϵ)(b_\epsilon,k_\epsilon) therefore describes a reconstruction that first minimizes residual memory, and then maximizes Bell yield while maintaining the desired error. Importantly, the compressibility problem considered here mainly concerns this minimum residual size kϵk_\epsilon, and not the computational cost of finding or implementing the encoders.

Nonlocal stabilizer fidelity. When no residual is allowed in Eq. (2), k=0k=0, the target contains only flat Bell pairs and local pure factors. The optimized compression fidelity then reduces to the known nonlocal stabilizer fidelity [35, 36], FNL=Fk=0=max⁡0≤b≤n2−b(∑i=02b−1λi)2F_{\mathrm{NL}}=F_{k=0}=\max_{0\le b\le n}2^{-b}\left(\sum_{i=0}^{2^b-1}\sqrt{\lambda_i}\right)^2, where D:=−log⁡2FNLD:=-\log_2 F_{\mathrm{NL}} is nonlocal magic. Allowing a residual, one obtains FNL≤Fk≤min⁡{1,(k+1)FNL}F_{\mathrm{NL}}\le F_k\le\min\{1,(k+1)F_{\mathrm{NL}}\}. The upper fidelity bound follows by grouping ordered Schmidt amplitudes into successive powers of two, as described in SM. Crucially, if FNLF_{\mathrm{NL}} tends to zero in the thermodynamic limit, then any memory size growing more slowly than 1/FNL1/F_{\mathrm{NL}} has vanishing fidelity. Further, requiring Fk≥1−ϵF_k\ge1-\epsilon, it holds that

kϵ≥max⁡{0,⌈(1−ϵ)2D−1⌉}.(4)k_\epsilon\ge\max\left\{0,\left\lceil(1-\epsilon)2^D-1\right\rceil\right\}. \tag*{(4)}

Importantly, it follows that if D(n)−log⁡2[k(n)+1]→+∞D(n)-\log_2[k(n)+1]\to+\infty, then Fk(n)→0F_{k(n)}\to0. In particular, if D(n)=βlog⁡2n+O(1)D(n)=\beta\log_2 n+O(1) with β>0\beta>0, every memory size satisfying k(n)/nβ→0k(n)/n^\beta\to0 yields vanishing fidelity. Maintaining fidelity at least f0>0f_0>0 therefore requires k(n)≥f02D(n)−1k(n)\ge f_0 2^{D(n)}-1, so the residual memory must grow at least proportionally to nβn^\beta for this example. Hence, unbounded growth of nonlocal magic constrains not only the best zero-memory approximation, but also the memory required to reach a prescribed reconstruction fidelity. Next, we show that smoothed entanglement spread directly bounds the residual memory required at a prescribed reconstruction accuracy.

Entanglement spread. A complementary and more informative lower bound on the residual memory follows from entanglement spread [18, 19, 42], which also constrains the communication required for entanglement dilution and pure-state preparation from Bell pairs. For ρ^A=Tr⁡B∣ψ⟩⟨ψ∣\hat{\rho}_A=\operatorname{Tr}_B|\psi\rangle\langle\psi|, its unsmoothed form is

Δ0(ψ):=log⁡2(rank⁡ρ^A∥ρ^A∥∞),(5)\Delta_0(\psi):=\log_2(\operatorname{rank}\hat{\rho}_A\|\hat{\rho}_A\|_\infty), \tag*{(5)}

where ∥ρ^A∥∞\|\hat{\rho}_A\|_\infty denotes the largest singular value. We can recast it as Δ0=S0−S∞\Delta_0=S_0-S_\infty, where Sα:=(1−α)−1log⁡2Tr⁡ρ^AαS_\alpha:=(1-\alpha)^{-1}\log_2\operatorname{Tr}\hat{\rho}_A^\alpha is the α\alpha-th Rényi entropy. Δ0\Delta_0 is a measure of nonflatness of the reduced spectrum [33, 45], and is additive under tensor products, so that tensoring flat Bell pairs leaves it unchanged. For a pure stabilizer state the nonzero Schmidt probabilities across every bipartition are equal, and one finds S0=S∞S_0=S_\infty; an exact factorization of the form Eq. (1) then satisfies Δ0(ψ)=Δ0(γ)≤k0\Delta_0(\psi)=\Delta_0(\gamma)\le k_0, meaning all the spread simply lives in the residual state (here k0≡kϵ=0k_0\equiv k_{\epsilon=0}). Crucially, the monotonicity of Rényi entropies also gives 0≤Sα−Sβ≤Δ00\le S_\alpha-S_\beta\le\Delta_0 for 0≤α≤β≤∞0\le\alpha\le\beta\le\infty, upper bounding all orders of Rényi entropy gaps. For power-of-two ranks it also (loosely) holds that D≤S0−S1/2≤S0−S∞=Δ0D\le S_0-S_{1/2}\le S_0-S_\infty=\Delta_0, establishing that bounded spread prevents divergent nonlocal magic. For a full-rank state Δ0=n+log⁡2λmax⁡\Delta_0=n+\log_2\lambda_{\max}.

The raw quantity Δ0\Delta_0 is sensitive to arbitrarily small Schmidt weights and must be "smoothed" to describe a finite-error compression protocol. For the ordered Schmidt probabilities λ\lambda, we can define

Δϵ(λ):=min⁡μrlog⁡2(∣supp⁡μ∣/μmax⁡)∑rλrμr≥1−ϵ,(6)\Delta_\epsilon(\lambda):=\frac{\min_{\mu_r}\log_2(\lvert\operatorname{supp}\mu\rvert/\mu_{\max})}{\sum_r\sqrt{\lambda_r\mu_r}\ge\sqrt{1-\epsilon}}, \tag*{(6)}

where ∑rμr=1\sum_r \mu_r = 1, and μ\mu ranges over normalized, non-increasing probability vectors padded to dimension 2n2^n. supp⁡μ\operatorname{supp}\mu denotes the set of indices with positive entries, and μmax⁡\mu_{\max} is the largest entry. To remain consistent, we use the same squared fidelity convention throughout. Since every admissible factorized target has spread at most kk, the minimum residual memory simply obeys

⌈Δϵ(λ)⌉≤kϵ(ψ).(7)\lceil\Delta_{\epsilon}(\lambda) \rceil\le k_{\epsilon}(\psi). \tag*{(7)}

Physically, the above relation makes explicit that the residual memory ∣γk⟩|\gamma_k\rangle must accommodate the spectral nonflatness that cannot be discarded within the error tolerance. Overall, the spread bound, Eq. (7), can give a stronger quantitative obstruction at a chosen accuracy, while the nonlocal magic bound, Eq. (4), directly relates a resource measure to the full fidelity–memory tradeoff. Given such characterizations, we next study how the spectral compressibility of dynamically generated states evolves in time in two distinct families of random quantum circuits.

Chaotic random circuits. We first consider chaotic random unitary dynamics without conservation laws by evolving the initial state ∣0⟩⊗N|0\rangle^{\otimes N} on a periodic chain divided into equal contiguous halves. A time step consists of independent single-qubit Haar-random rotations followed by even-bond controlled-Z^\hat{Z} gates, diag⁡(1,1,1,−1)\operatorname{diag}(1,1,1,-1), then fresh random rotations followed by odd-bond entangling gates [1, 2, 46].

As displayed in Figure 1(a), the residual memory cost kϵk_{\epsilon} first grows, reaches a size-dependent peak, and then decreases toward a common late-time plateau. Tightening the error tolerance ϵ\epsilon from 0.10 to 0.01 naturally raises the memory requirement while preserving this barrier-like pattern. During the later part of the evolution, the entropy continues to increase even as the remainder shrinks (Figure 1(b)). Entanglement can therefore continue to grow through the flat Bell-pair component even as the required nonflat residual memory decreases. Choosing a fixed register size k=2k=2, as shown in Figure 1(c) and its inset, the transient fidelity minimum decreases with NN, whereas the late-time values are size independent and approach the high-fidelity Haar benchmark. These results give a compression interpretation of transient embezzling spectra associated with diverging nonlocal magic (D→∞D \to\infty) and entanglement capacity [33, 37, 38], casting the “magic barrier” [30, 47] as an operational barrier to entanglement-spectrum compression.

Spectral compression barrier in chaotic dynamics

Figure 1. Spectral compression barrier in chaotic dynamics. (a) Mean minimum residual memory kˉϵ\bar{k}_{\epsilon} at ϵ=0.01\epsilon= 0.01 (solid) and 0.100.10 (faint dashed). (b) Average entropy density Sˉ/n\bar{S}/n, and finite-error Bell yield density bˉϵ/n\bar{b}_{\epsilon}/n at N=26N = 26, with the same error convention. (c) Optimal fixed-kk fidelity F2=F2F_2 = F_2? Inset: its minimum ensemble mean at t=arg⁡min⁡τFˉ2(τ)t = \arg\min_{\tau}\bar{F}_2(\tau) (rose circles) and value at t=Nt = N (blue squares), versus NN. (d) kˉϵ\bar{k}_{\epsilon}, mean smoothed spread Δˉϵ\bar{\Delta}_{\epsilon}, and mean magic bound ℓϵ:=max⁡{0,(1−ϵ)2Dˉ−1}\ell_{\epsilon} := \max\{0,(1-\epsilon)2^{\bar{D}}-1\} at N=26N = 26, ϵ=0.01\epsilon= 0.01, and mean raw spread Δˉ0\bar{\Delta}_0. Optimization precedes averaging, which is done over 50–400 independent circuit realizations. Error bars are of the size of the symbols or smaller.

In Figure 1(d) we also plot the required memory with the lower bounds introduced earlier. At a fixed error, ϵ=0.01\epsilon= 0.01, both the smoothed spread and nonlocal magic bound follow the transient rise and fall, while spread provides a tighter lower bound for these data. The numerical optimization of Δϵ\Delta_{\epsilon} is described in SM. The behavior of Δ0\Delta_0 also admits a simple interpretation. S0S_0 grows linearly until the rank saturates, while S∞=−log⁡2λmax⁡S_{\infty}=-\log_2\lambda_{\max} continues to increase on average, reducing their difference and resulting

in the rise–peak–fall behavior. For the chaotic transient embezzling states, the Schmidt spectra are already full-rank and nondegenerate, giving k0=nk_0=n and k0/Δ0=n(n−S∞)−1k_0/\Delta_0=n(n-S_{\infty})^{-1}. Numerically, k0/Δ0≈1.7k_0/\Delta_0\approx1.7 over the sampled sizes, so the raw spread tracks a roughly fixed fraction of the exact residual memory.

In Figure 2 we resolve the full fidelity–memory tradeoff for the chaotic transient and at the late time t=Nt=N Haar states. The transient embezzling states approximately align under (k+cch)n−βch(k+c_{\mathrm{ch}})n^{-\beta_{\mathrm{ch}}} with βch≈0.45\beta_{\mathrm{ch}}\approx0.45 and cch≈2.30c_{\mathrm{ch}}\approx2.30 (inset of Figure 2(a)) giving an effective and approximate finite-size exponent. The exponent β\beta characterizes the growth of the memory required at fixed fidelity, while the size-independent offset cc accounts phenomenologically for finite-size corrections. The approximate collapse is consistent with algebraic but subextensive growth of the residual memory over the sampled subsystem sizes in the high-fidelity regime. Remarkably, as hinted earlier, the fidelity–memory curves for Haar-random states instead coincide without rescaling the kk axis, approaching a size-independent form shown by the dashed line in Figure 2(b). This limiting curve follows from the Marchenko–Pastur spectral distribution [48, 49, 50] (see SM). Across a balanced cut, a typical large Haar state admits coherent compression into b=n−Oϵ(1)b=n-O_{\epsilon}(1) Bell pairs with a residual register whose size depends only on the target accuracy. In this limit, the infidelity obeys 1−FkH≤144−k1-F_k^{\mathrm{H}}\leq\frac{1}{4}4^{-k} for fixed k≥1k\geq1, so O(log⁡(1/ϵ))O(\log(1/\epsilon)) residual qubits per party suffice (see SM). With no residual memory, the optimum occurs at b=n−1b=n-1, yielding FNLH≈0.84F_{\mathrm{NL}}^{\mathrm{H}}\approx0.84 and DH≈0.24D^{\mathrm{H}}\approx0.24 bits. The raw spread also stays bounded, with Δ0H=log⁡2(2nλmax⁡)→2\Delta_{0}^{\mathrm{H}}=\log_{2}(2^{n}\lambda_{\max})\to2 [51]. The nonzero limiting FNLHF_{\mathrm{NL}}^{\mathrm{H}} thus rules out universal embezzlement despite the near-maximal Bell yield [35, 37].

Compression fidelity–memory tradeoff in chaotic circuits

Figure 2. Compression fidelity–memory tradeoff in chaotic circuits. (a) Chaotic transient states evaluated at t=arg⁡min⁡τF‾0(τ)t = \arg\min_{\tau}\overline{F}_{0}(\tau) for each size. Inset: collapse against (k+cch)n−βch(k+c_{\mathrm{ch}})n^{-\beta_{\mathrm{ch}}}, with cch≈2.30c_{\mathrm{ch}}\approx2.30 and βch≈0.45\beta_{\mathrm{ch}}\approx0.45, where “ch” denotes chaotic. (b) Late-time chaotic-circuit states at t=Nt=N. Black dashed line shows the analytically derived large-NN Haar limit. Each point is the ensemble mean of the individually optimized FkF_k at integer kk. Collapse parameters are fitted over the shaded window 0.90≤F‾k≤0.990.90\leq\overline{F}_k\leq0.99, denoted by the dotted horizontal lines. Data are averages over 50–400 independent circuit realizations and error bars are of the size of the symbols or smaller.

These characterizations also motivate reexamining the accuracy of the entropy-based estimate χ∼2S1\chi\sim2^{S_{1}} for an approximate matrix product state representation [5, 6, 21, 52, 53]. Let χϵ\chi_{\epsilon} be the smallest Schmidt rank that retains total probability at least 1−ϵ1-\epsilon. We define the spectral compression gap δϵ:=log⁡2χϵ−S1\delta_{\epsilon}:=\log_{2}\chi_{\epsilon}-S_{1} to quantify the departure of the required bond dimension across the cut from this estimate. When δϵ>0\delta_{\epsilon}>0, the Schmidt rank required at the chosen fidelity exceeds the entropy estimate 2S12^{S_{1}} by a factor 2δϵ2^{\delta_{\epsilon}}. To connect this gap to residual memory, note that a reconstruction with bb Bell pairs and kk residual qubits per side has Schmidt rank at most 2b+k2^{b+k}. Achieving fidelity at least 1−ϵ1-\epsilon therefore requires b+k≥⌈log⁡2χϵ⌉=⌈S1+δϵ⌉b+k\geq\lceil\log_{2}\chi_{\epsilon}\rceil=\lceil S_{1}+\delta_{\epsilon}\rceil. In particular, the optimal allocation (bϵ,kϵ)(b_{\epsilon},k_{\epsilon}) obeys kϵ≥max⁡{0,⌈S1+δϵ⌉−bϵ}k_{\epsilon}\geq\max\{0,\lceil S_{1}+\delta_{\epsilon}\rceil-b_{\epsilon}\}. This gives a necessary rank constraint on the optimal allocation. For exact reconstruction, ϵ=0\epsilon=0, each Bell pair contributes one bit to both log⁡2χ0\log_{2}\chi_{0} and S1S_{1}. The gap therefore resides entirely in the residual state, δ0(ψ)=δ0(γ)=S0−S1≤Δ0\delta_{0}(\psi)=\delta_{0}(\gamma)=S_{0}-S_{1}\leq\Delta_{0} [54]. If γ\gamma has full Schmidt rank 2k2^{k}, then δ0=k−S1(γ)\delta_{0}=k-S_{1}(\gamma). As shown numerically in the SM, at fixed ϵ\epsilon, the required bond dimension for chaotic transient states exceeds the entropy-based estimate by a factor 2δϵ2^{\delta_{\epsilon}} that increases over the sizes studied. At late times, this factor is approximately size-independent. These results reveal a growing multiplicative overhead associated with multiscale entanglement spectra that entropy alone misses.

Matchgate circuits. We now consider free-fermionic dynamics, where independent nearest-neighbor Gaussian gates act in even and odd layers on a periodic ring, starting from the even-parity vacuum. Such matchgate dynamics admits an efficient (exactly solvable) covariance-matrix description [55, 56]. Each gate is a random SO(4)\mathrm{SO}(4) rotation on the four Majorana modes of its two sites, and the circuit preserves fermion parity but not particle number. Let ΓA\Gamma_A denote the 2n×2n2n\times2n covariance matrix of subsystem AA, defined by (ΓA)rs:=(i/2)⟨[c^r,c^s]⟩(\Gamma_A)_{rs}:=(i/2)\langle[\hat{c}_r,\hat{c}_s]\rangle, where c^r\hat{c}_r are its Majorana operators, normalized by {c^r,c^s}=2δrs\{\hat{c}_r,\hat{c}_s\}=2\delta_{rs}. For a pure bipartite fermionic Gaussian state, the eigenvalues of ΓA\Gamma_A occur in pairs ±iξj\pm i\xi_j, with 0≤ξj≤10\leq\xi_j\leq1. The reduced state of a pure fermionic Gaussian state is unitarily equivalent to a product of single-mode density matrices, ρ^A≃⨂j=1ndiag⁡((1+ξj)/2,(1−ξj)/2)\hat{\rho}_A\simeq\bigotimes_{j=1}^{n}\operatorname{diag}((1+\xi_j)/2,(1-\xi_j)/2) [57, 58, 59, 60, 61, 62]. Each ξj\xi_j describes one entangled mode pair and 0<ξj<10<\xi_j<1 corresponds to a partially mixed pair.

In the invariant Gaussian ensemble, ξj<1\xi_j<1 almost surely, so that rank⁡ρ^A=2n\operatorname{rank}\hat{\rho}_A=2^n and ∥ρ^A∥∞=∏j=1n(1+ξj)/2\|\hat{\rho}_A\|_{\infty}=\prod_{j=1}^{n}(1+\xi_j)/2, giving Δ0=∑j=1nlog⁡2(1+ξj)\Delta_0=\sum_{j=1}^{n}\log_2(1+\xi_j). For an equal bipartition, the normalized single-mode spectral density of the invariant Gaussian ensemble is ρn(x)=(1/n)∑ℓ=0n−1(4ℓ+1)P2ℓ(x)2\rho_n(x)=(1/n)\sum_{\ell=0}^{n-1}(4\ell+1)P_{2\ell}(x)^2, 0≤x≤10\leq x\leq1 [63], where PmP_m denotes the Legendre polynomial of degree mm and ∫01ρn(x) dx=1\int_0^1\rho_n(x)\,dx=1. Using this gives the exact finite-size stationary average, EssΔ0=∑ℓ=0n−1(4ℓ+1)∫01P2ℓ(x)2log⁡2(1+x) dx\mathbb{E}_{\mathrm{ss}}\Delta_0=\sum_{\ell=0}^{n-1}(4\ell+1)\int_0^1P_{2\ell}(x)^2\log_2(1+x)\,dx. Note the steady state EssΔ0\mathbb{E}_{\mathrm{ss}}\Delta_0 is evaluated for each pure state and then averaged over the ensemble. In the thermodynamic limit, the mode density approaches the arcsine density ρ∞(x)=2/[π1−x2]\rho_{\infty}(x)=2/[\pi\sqrt{1-x^2}]. Consequently,

lim⁡n→∞EssΔ0n=2πln⁡2∫01ln⁡(1+x)1−x2 dx=4Gπln⁡2−1=0.682…,(8)\lim_{n\to\infty}\frac{\mathbb{E}_{\mathrm{ss}}\Delta_0}{n} =\frac{2}{\pi\ln2}\int_0^1\frac{\ln(1+x)}{\sqrt{1-x^2}}\,dx =\frac{4G}{\pi\ln2}-1=0.682\ldots, \tag*{(8)}

where G=∑m=0∞(−1)m/(2m+1)2G=\sum_{m=0}^{\infty}(-1)^m/(2m+1)^2 is Catalan’s constant [63]. Thus the stationary unsmoothed spread has the leading behavior EssΔ0≈0.341N\mathbb{E}_{\mathrm{ss}}\Delta_0\approx0.341N. Such states have a nondegenerate Schmidt spectrum almost surely, so k0=nk_0=n and k0/EssΔ0≈1.46k_0/\mathbb{E}_{\mathrm{ss}}\Delta_0\approx1.46.

As shown in Figure 3(a), here the late-time memory remains size dependent for the chosen error tolerances, which can be also verified by independently sampling the global SO(2N)\mathrm{SO}(2N) rotations of the vacuum. The stationary fidelity–memory curves shift toward larger residual registers as the system grows (Figure 3(b)). Similar to the transient chaotic embezzlers, the high-fidelity parts roughly align against (k+cmg)n−βmg(k+c_{\mathrm{mg}})n^{-\beta_{\mathrm{mg}}}, with the numerical estimates βmg≈0.59\beta_{\mathrm{mg}} \approx0.59 and cmg≈2.48c_{\mathrm{mg}} \approx2.48, fitted jointly across the displayed sizes. Noting that the stationary matchgate states are expected to obey the asymptotic scaling D=(1/2)log⁡2n+O(1)D = (1/2)\log_2 n + O(1) [38], the fidelity bound Eq. (4) then implies that the residual memory must grow at least proportionally to n\sqrt{n} to retain a fixed positive fidelity. The fitted exponent βmg\beta_{\mathrm{mg}} is thus consistent with the condition imposed by the growth of nonlocal magic (within the limits of the system sizes studied). Further, as shown in Figure 3(c), Δϵ\Delta_{\epsilon} grows sublinearly alongside kϵk_{\epsilon} and bounds it from below; the linear growth of Δ0\Delta_0 also agrees with the analytical average in Eq. (8). Finally, Figure 3(d) shows the ensemble-averaged Bell-pair fraction bϵ/nb_{\epsilon}/n, total retained fraction (bϵ+kϵ)/n(b_{\epsilon}+k_{\epsilon})/n, and entropy density S1/nS_1/n. In the stationary Gaussian ensemble, both bϵ/nb_{\epsilon}/n and (bϵ+kϵ)/n(b_{\epsilon}+k_{\epsilon})/n are expected to approach the same asymptotic value as the entropy density S1/n≈0.55S_1/n \approx0.55 in the thermodynamic limit N,n→∞N,n \to\infty, consistent with extensive entanglement [63].

Persistent residual memory of stationary Gaussian states

Figure 3. Persistent residual memory of stationary Gaussian states. (a) Mean minimum residual memory kˉϵ\bar{k}_{\epsilon} at ϵ=0.01\epsilon= 0.01 (solid) and 0.10 (faint dashed) as a function of t/N2t/N^2. (b) Stationary fidelity–memory frontiers (i.e. optimized over Bell-pair yield for each kk). Inset: the same data against (k+cmg)n−βmg(k+c_{\mathrm{mg}})n^{-\beta_{\mathrm{mg}}}, with cmg≈2.48c_{\mathrm{mg}} \approx2.48 and βmg≈0.59\beta_{\mathrm{mg}} \approx0.59, where "mg" denotes matchgate. Collapse parameters are fitted over the shaded window 0.90≤F‾k≤0.990.90 \leq\overline{F}_k \leq0.99, denoted by the dotted horizontal lines. (c) Minimum memory and fidelity-smoothed spread at both errors, with the line convention of (a). Open circles show raw spread and the dotted curve is its exact stationary mean, Eq. (8). (d) Bell yield, entropy, and total retained qubits per party, divided by n=N/2n = N/2, with the same error convention. All optimizations precede averaging. Data are averages over 100–200 independent circuit realizations and error bars are of the size of the symbols or smaller.

Conclusions. The exact tradeoff between Bell-pair yield, residual memory, and reconstruction fidelity defines an operational measure of entanglement-spectrum complexity, kϵk_{\epsilon}, quantifying the minimum nonflat residual memory required at a chosen accuracy. Smoothed entanglement spread bounds this cost from below, while diverging nonlocal magic forces it to grow without bound at any fixed fidelity. Thus, the spectral structure enabling universal entanglement embezzlement also obstructs faithful compression into Bell pairs and bounded residual memory. The residual memory cost in chaotic circuit dynamics rises to a peak and then decreases, even as entanglement entropy continues to grow. During this regime, the bond dimension required for fixed-fidelity compression across the cut exceeds the entropy based estimate by a factor whose peak increases over the simulated sizes. A large unrestricted Haar-typical state however asymptotically admits a size-independent remainder at fixed accuracy, and stationary fermionic Gaussian states retain a storage cost that increases with size. The optimal fidelity also provides a benchmark for quantum autoencoders [64] that separate Bell pairs from a residual register using local unitaries. Determining how efficiently such encoders may achieve this fidelity remains an open problem.

Acknowledgments. This work was supported by the European Union and the European Innovation Council through the Horizon Europe Project No. QRC-4-ESP (Grant Agreement No. 101129663), and EU Horizon Europe Quest project (Project No. 101156088). T.O. acknowledges the support by the Finnish Research Council project 362573 and the Finnish quantum flagship program. This work is part of the Finnish Center of Excellence in Quantum Materials (QMAT). GPT-6 family was used as an auxiliary tool for literature discovery, cross-checking scientific reasoning, and assisting with mathematical proofs, code development and automation. All scientific conclusions and interpretations are those of the authors.

References

Supplemental Material for

"The magical memory cost of entanglement spectrum compression"

Matias Karjula, Vilma Nieminen, Teemu Ojanen, Tapio Ala-Nissila, and Moein N. Ivaki

Optimal compression fidelity. Let b′:=2bb' := 2^b, and k′:=2kk' := 2^k, and let μ0≥⋯≥μk′−1\mu_0 \ge\cdots\ge\mu_{k'-1} be the residual Schmidt probabilities. Tensoring with bb Bell pairs repeats each μj/b′\mu_j/b' exactly b′b' times. To maximize the overlap over local bases, we can write the two states ψ,Ω\psi,\Omega in fixed product bases as coefficient matrices CψC_\psi and CΩC_\Omega. These are the matrices of amplitudes in fixed product bases, whose singular values are the Schmidt amplitudes. Local unitaries transform CψC_\psi into UACψUBTU_A C_\psi U_B^{\mathsf T}, and the overlap becomes Tr⁡(CΩ†UACψUBT)\operatorname{Tr}(C_\Omega^\dagger U_A C_\psi U_B^{\mathsf T}), where UA,BU_{A,B} are the matrix representations of U^A,B\hat U_{A,B}. The von Neumann trace inequality bounds the absolute value of this trace by the sum of products of the ordered singular values. Equality can then be achieved by aligning the corresponding left and right singular vectors, which the independent local unitaries allow [43]. Thus the optimal overlap pairs the largest Schmidt amplitude of one state with the largest of the other, and so on. Applying these results, for a fixed residual spectrum μ\mu it gives

max⁡U^A,U^B∣⟨Ωb,k∣(U^A⊗U^B)∣ψ⟩∣=1b′∑j=0k′−1Ajμj.(9)\max_{\hat U_A,\hat U_B} \left|\langle\Omega_{b,k}|(\hat U_A\otimes\hat U_B)|\psi\rangle\right| = \frac{1}{\sqrt{b'}}\sum_{j=0}^{k'-1} A_j\sqrt{\mu_j}. \tag*{(9)}

where Aj:=∑s=0b′−1λjb′+sA_j := \sum_{s=0}^{b'-1}\sqrt{\lambda_{jb'+s}}. Using the Cauchy–Schwarz inequality then gives

(∑jAjμj)2≤(∑jAj2)(∑jμj)=∑jAj2.(10)\left(\sum_j A_j\sqrt{\mu_j}\right)^2 \le\left(\sum_j A_j^2\right)\left(\sum_j\mu_j\right)=\sum_j A_j^2. \tag*{(10)}

Equality is achieved by taking μj\sqrt{\mu_j} proportional to AjA_j, or μj=(Aj2)/∑lAl2\mu_j=(A_j^2)/\sum_l A_l^2, where the denominator normalizes the residual. This choice also respects the ordering, as AjA_j are nonincreasing and the input probabilities are sorted. Local unitary encoders that map the corresponding Schmidt bases to the designated register bases then achieve the equality. Thus Eq. (2) in the main text is an attainable optimum, and not only an upper bound.

It is also useful to note that the optimal reconstruction error separates exactly into a discarded Schmidt weight and a flattening error, where the errors arise from approximating the encoded state by the prescribed Bell-pair-plus-remainder form. We can write 1−Fb,k=δtail+δflat1-F_{b,k}=\delta_{\mathrm{tail}}+\delta_{\mathrm{flat}}, where

δtail:=∑r=b′k′2n−1λr,(11)\delta_{\mathrm{tail}}:=\sum_{r=b'k'}^{2^n-1}\lambda_r, \tag*{(11)}
δflat:=∑j=0k′−1[∑s=0b′−1λjb′+s−1b′(∑s=0b′−1λjb′+s)2].\delta_{\mathrm{flat}}:=\sum_{j=0}^{k'-1}\left[\sum_{s=0}^{b'-1}\lambda_{jb'+s}-\frac{1}{b'}\left(\sum_{s=0}^{b'-1}\sqrt{\lambda_{jb'+s}}\right)^2\right].

Here δtail\delta_{\mathrm{tail}} is the probability weight beyond the retained Schmidt components. The second contribution measures the departure from the equal-weight blocks required by bb Bell pairs. Indeed, defining aˉj:=(1/b′)∑s=0b′−1λjb′+s\bar a_j:=(1/b')\sum_{s=0}^{b'-1}\sqrt{\lambda_{jb'+s}}, one obtains δflat=∑j=0k′−1∑s=0b′−1(λjb′+s−aˉj)2≥0\delta_{\mathrm{flat}}=\sum_{j=0}^{k'-1}\sum_{s=0}^{b'-1}\left(\sqrt{\lambda_{jb'+s}}-\bar a_j\right)^2\ge0. Thus, as one might expect, flattening incurs no error precisely when the probabilities within each retained block are equal. For b=0b=0, each block contains a single probability and δflat=0\delta_{\mathrm{flat}}=0. For b+k=nb+k=n, all Schmidt components are retained and δtail=0\delta_{\mathrm{tail}}=0, although flattening can still incur an error when b>0b>0.

Memory bound from fidelity. Adopting the same notations, Tj:=Aj/b′T_j:=A_j/\sqrt{b'} gives a nonnegative, nonincreasing sequence with Fb,k=∑j=0k′−1Tj2F_{b,k}=\sum_{j=0}^{k'-1}T_j^2. The first term, T02T_0^2, is the fidelity with a rank-b′b' Bell sector and is therefore at most FNLF_{\mathrm{NL}}. The remaining terms can be bounded in groups with indices m,…,2m−1m,\ldots,2m-1, where m=1,2,4,…,2k−1m=1,2,4,\ldots,2^{k-1}. Since the sequence is nonincreasing, every TjT_j in such a group is no larger than the average of the first mm terms. Consequently,

∑j=m2m−1Tj2≤m(1m∑j=0m−1Tj)2=1b′m(∑i=0b′m−1λi)2≤FNL.(12)\begin{aligned} \sum_{j=m}^{2m-1}T_j^2\le m\left(\frac{1}{m}\sum_{j=0}^{m-1}T_j\right)^2 \\ &=\frac{1}{b'm}\left(\sum_{i=0}^{b'm-1}\sqrt{\lambda_i}\right)^2\le F_{\mathrm{NL}}. \tag*{(12)} \end{aligned}

Here the first mm blocks contain exactly the first b′mb'm Schmidt amplitudes, and the resulting expression is therefore a Bell-sector fidelity of rank b′mb'm. The choice of power-of-two group sizes ensures that b′m=2b+log⁡2mb'm = 2^{b+\log_2 m} is an allowed Bell-sector rank. The first term T02T_0^2 plus these kk groups cover all 2k2^k terms, with each contributing at most FNLF_{\mathrm{NL}}. Summing gives Fb,k≤(k+1)FNLF_{b,k} \le(k+1)F_{\mathrm{NL}}, and maximizing over bb proves the upper bound on FkF_k. For the lower bound, we simply note that any zero-memory approximation can also be represented with a kk-qubit residual by moving Bell pairs into the residual register, and filling its remaining qubits with local pure factors. Thus Fk≥FNLF_k \ge F_{\mathrm{NL}}. Finally, requiring Fk≥1−ϵF_k \ge1-\epsilon gives 1−ϵ≤(k+1)FNL1-\epsilon\le(k+1)F_{\mathrm{NL}}. Rearranging and noting that kk is an integer, yields the finite-error memory bound in Eq. (4).

Compression of unrestricted Haar-random states. The key idea here is to eventually express the Haar compression problem as an approximation of a function by its averages over equal intervals. For d=2nd=2^n, the rescaled Schmidt probabilities x:=dλx := d\lambda have density ρ(x):=(2π)−1(4−x)/x\rho(x) := (2\pi)^{-1}\sqrt{(4-x)/x} on 0<x<40 < x < 4 in the balanced (equal contiguous bipartitioning) Haar-typical limit [48, 49]. We order the probabilities from largest to smallest, and let u∈[0,1]u \in[0,1] denote the fraction counted from the largest; x(u)x(u) then gives the rescaled probability at position uu in the ordered spectrum, with u=0u=0 corresponding to the largest values and u=1u=1 to the end of the spectrum. Equivalently, u=∫x(u)4ρ(y) dyu = \int_{x(u)}^4 \rho(y)\,\mathrm{d}y.

The substitution x=4cos⁡2θx=4\cos^2\theta can simplify the square root in the density, giving

x(u)=4cos⁡2θ,u=2θ−sin⁡2θπ,0≤θ≤π2.(13)x(u)=4\cos^2\theta,\qquad u=\frac{2\theta-\sin2\theta}{\pi},\qquad0\le\theta\le\frac{\pi}{2}. \tag*{(13)}

Using du/dθ=4sin⁡2θ/π\mathrm{d}u/\mathrm{d}\theta=4\sin^2\theta/\pi, the cumulative amplitude X(u)=∫0ux(v) dvX(u)=\int_0^u\sqrt{x(v)}\,\mathrm{d}v and cumulative probability P(u)=∫0ux(v) dvP(u)=\int_0^u x(v)\,\mathrm{d}v become

X(u)=83πsin⁡3θ,P(u)=2θ−12sin⁡4θπ.(14)X(u)=\frac{8}{3\pi}\sin^3\theta,\qquad P(u)=\frac{2\theta-\frac{1}{2}\sin4\theta}{\pi}. \tag*{(14)}

Here uu counts eigenvalues, whereas P(u)P(u) counts their probability weight. For example, the largest half of the Schmidt probabilities carries the majority of the total probability, and P(1/2)≈0.8937P(1/2)\approx0.8937.

For an allocation (b,k)(b,k), write r:=n−b−kr:=n-b-k for the number of local pure-state qubits per party. The target has Schmidt rank at most 2b+k2^{b+k}, so its optimal overlap uses the largest 2b+k2^{b+k} input probabilities. These occupy a fraction 2−r2^{-r} of the full spectrum and are grouped into 2k2^k blocks, each containing 2b2^b probabilities. Each block therefore has width h:=2b/d=2−(k+r)h:=2^b/d=2^{-(k+r)} in the variable uu. In the large-dd limit, its amplitude sum becomes ∑blockλi≃d∫blockx(u) du\sum_{\mathrm{block}}\sqrt{\lambda_i}\simeq\sqrt{d}\int_{\mathrm{block}}\sqrt{x(u)}\,\mathrm{d}u. Substituting into Eq. (2), with d/2b=1/hd/2^b=1/h, gives the limiting fidelity at fixed k,rk,r,

Fk,r=1h∑j=02k−1[X((j+1)h)−X(jh)]2.(15)F_{k,r}=\frac{1}{h}\sum_{j=0}^{2^k-1}\left[X((j+1)h)-X(jh)\right]^2. \tag*{(15)}

This expression accounts for both the omitted probability beyond u=2−ru=2^{-r} and the error from approximating the retained amplitudes by equal values within each block. In particular, the fidelity cannot exceed the probability included in the target support, so Fk,r≤P(2−r)F_{k,r}\le P(2^{-r}).

We may now try and optimize the Bell yield explicitly by comparing the numerical values of fidelity for allowed choices of rr. The numerical values can be calculated directly using Python/SciPy or similar libraries. With no residual memory k=0k=0, choosing r=0r=0 and r=1r=1 give F0,0≈0.7205F_{0,0}\approx0.7205 and F0,1≈0.8444F_{0,1}\approx0.8444, respectively. Every r≥2r\ge2 includes at most the largest quarter of the spectrum and therefore has fidelity at most P(1/4)≈0.6229P(1/4)\approx0.6229. Thus r=1r=1, or b=n−1b=n-1, is optimal for k=0k=0. With one residual qubit k=1k=1, retaining the whole spectrum instead gives F1,0≈0.9236F_{1,0}\approx0.9236. This already exceeds P(1/2)≈0.8937P(1/2)\approx0.8937, the upper bound for every allocation with r≥1r\ge1. Increasing kk at r=0r=0 subdivides each block into smaller blocks, allowing a more accurate approximation of the amplitudes and hence a fidelity that cannot decrease. Therefore, r=0r=0, or b=n−kb=n-k, remains optimal for every fixed k≥1k\ge1. Allocations with rr growing without bound as nn increases cannot improve the result either. Their retained probability is at most 2−rdλmax⁡2^{-r}d\lambda_{\max}, which vanishes since dλmax⁡→4d\lambda_{\max}\to4 [51]. Setting r=0r=0, and hence h=2−kh=2^{-k}, in Eq. (15) then gives

FkH=2k∑j=02k−1[X((j+1)2−k)−X(j2−k)]2.F_k^{\mathrm{H}}=2^k\sum_{j=0}^{2^k-1}\left[X((j+1)2^{-k})-X(j2^{-k})\right]^2.

For the optimal choice r=0r=0, no Schmidt components are omitted, so the infidelity arises entirely from flattening the amplitudes within each block. To bound this error, let f(u):=x(u)f(u):=\sqrt{x(u)} and set h=2−kh=2^{-k}. The mean amplitude on block jj is fˉj=[X((j+1)h)−X(jh)]/h\bar f_j=[X((j+1)h)-X(jh)]/h, so FkH=∑jhfˉj2F_k^{\mathrm{H}}=\sum_j h\bar f_j^2. Since ∫01f(u)2 du=1\int_0^1 f(u)^2\,\mathrm{d}u=1, subtracting these squared block means gives

1−FkH=∑j=02k−1∫jh(j+1)h[f(u)−fˉj]2 du.(16)1-F_k^{\mathrm{H}}=\sum_{j=0}^{2^k-1}\int_{jh}^{(j+1)h}\left[f(u)-\bar f_j\right]^2\,\mathrm{d}u. \tag*{(16)}

Thus the infidelity measures how much the amplitudes vary within the blocks. On each interval of width hh, the Poincaré-Wirtinger inequality bounds this variation by h2/π2h^2/\pi^2 times the integral of ∣f′(u)∣2|f'(u)|^2. Summing over all blocks gives

1−FkH≤h2π2∫01∣f′(u)∣2 du=124−k,k≥1.(17)1-F_k^H \le\frac{h^2}{\pi^2}\int_0^1 |f'(u)|^2\,du=\frac{1}{2}4^{-k},\quad k\ge1. \tag*{(17)}

To evaluate the integral, the parametrization above gives f′(u)=−π/(2sin⁡θ)f'(u)=-\pi/(2\sin\theta) and du=4sin⁡2θ dθ/πdu=4\sin^2\theta\,d\theta/\pi. Consequently, ∫01∣f′(u)∣2 du=π∫0π/2dθ=π2/2\int_0^1 |f'(u)|^2\,du=\pi\int_0^{\pi/2}d\theta=\pi^2/2. Although f′(u)f'(u) diverges at u=0u=0, its square is integrable with respect to uu, so the inequality applies. Each additional residual qubit halves the block width and reduces this error bound by a factor of four.

These statements concern the limiting Haar spectrum, with n→∞n\to\infty taken at fixed kk. For any fixed ϵ>0\epsilon>0, one can choose a fixed k≥1k\ge1 such that 124−k<ϵ\frac{1}{2}4^{-k}<\epsilon. The reconstruction fidelity then exceeds 1−ϵ1-\epsilon with probability approaching one as nn increases. As promised, the limiting memory only depends on the chosen accuracy.

Additional results. Figure S1 compares fixed and size-dependent residual-memory budgets in the two circuit models. For chaotic circuits, all budgets are evaluated at the same sampled time minimizing F‾0\overline{F}_0 for each system size. For matchgate circuits, the states are evaluated at t=N2t=N^2. Figure S1,Figure S1) show that the fidelity obtained with a fixed budget k=2k=2 decreases with increasing system size. Allowing the residual register to grow improves reconstruction, but the different budget choices have distinct finite-size behavior. In particular, the logarithmic and square-root budgets also show decreasing fidelity over the larger matchgate sizes, whereas the extensive budgets retain higher fidelities over the sampled range. Steps and overlaps in these curves arise from rounding the budgets to integer numbers of qubits. Figure S1,Figure S1) compare the minimum residual memory directly with the optimized smoothed spread at the same error tolerance. The chaotic transient and stationary matchgate data follow similar curves over their overlapping range, while satisfying the spread lower bound kϵ≥Δϵk_\epsilon\ge\Delta_\epsilon.

Four-panel plots showing mean fidelity and memory versus spread

Figure S1. Different memory budgets and the spread lower bound. (a,b) Mean fidelity for k=2k=2, k=⌈log⁡2n⌉k=\lceil\log_2 n\rceil, k=⌈n⌉k=\lceil\sqrt{n}\rceil, k=⌈n/4⌉k=\lceil n/4\rceil, and k=⌈n/2⌉k=\lceil n/2\rceil, with N=2nN=2n. (a) Chaotic states at a common pilot-selected time t=arg⁡min⁡τF‾0(τ)t=\arg\min_{\tau}\overline{F}_0(\tau) for each size. (b) Independent stationary Gaussian states. Bell yield is optimized per state and budget. Dotted lines mark 0.90 and 0.99; error bars give pointwise 95% state-bootstrap intervals, conditional on the selected time in (a). Open markers overlap when integer budgets coincide. (c,d) Mean minimum memory versus mean optimized smoothed spread at ϵ=0.01\epsilon=0.01 and 0.10, respectively. Orange diamonds: chaotic transient; blue circles: stationary matchgate. Each point is one size, with both quantities computed per state before averaging. Dotted lines give the unrounded lower bound k‾ϵ≥Δ‾ϵ\overline{k}_{\epsilon}\geq\overline{\Delta}_{\epsilon}.

We further examine the accuracy of entropy-based estimates of spectral compressibility, following the discussions in Ref. [21]. For the ordered Schmidt probabilities, let χϵ=min⁡{χ∣∑r=0χ−1λr≥1−ϵ}\chi_\epsilon=\min\{\chi\mid\sum_{r=0}^{\chi-1}\lambda_r\ge1-\epsilon\} and define the spectral compression gap as δϵ=log⁡2χϵ−S(ρ^A)\delta_\epsilon=\log_2\chi_\epsilon-S(\hat{\rho}_A). The retained weight is the squared fidelity of the optimal normalized Schmidt-truncated state, so ϵ\epsilon uses the same reconstruction-error convention as the compression task in the main text. Since χϵ=2S+δϵ\chi_\epsilon=2^{S+\delta_\epsilon}, a positive gap quantifies the multiplicative excess 2δϵ2^{\delta_\epsilon} of the required Schmidt rank over the entropy estimate 2S2^S. Thus δϵ\delta_\epsilon measures the correction beyond entropy to the total retained Schmidt space, and therefore determines the minimum bond dimension required at this cut in an open-boundary matrix-product-state representation. Such an overhead can grow when a fixed fraction of Schmidt probability, exceeding the error tolerance ϵ\epsilon, lies at entanglement energies −log⁡2λr-\log_2\lambda_r increasingly far above their mean S1S_1. Retaining this weight requires an increasing number of individually small Schmidt probabilities.

Figure S2 shows a transient peak in δϵ\delta_\epsilon that increases over the sampled sizes and recedes toward the size-independent Haar limits. We compute χϵ\chi_\epsilon from the state evolved on the periodic ring. In an open-boundary matrix product state ordered along the chain, χϵ\chi_\epsilon is the minimum dimension of the single bond crossing the half-chain bipartition at the prescribed fidelity, with all other bonds unrestricted. The corresponding ratio χϵ/2S1=2δϵ\chi_\epsilon/2^{S_1}=2^{\delta_\epsilon} also grows with size at the peak, suggesting a stretched-exponential overhead relative to the entropy-based estimate. Tightening the error tolerance from 0.010.01 to 0.0010.001 increases the gap, as more of the small Schmidt weights must be retained. Overall, these results indicate that entropy becomes a progressively poorer predictor of the rank required at fixed accuracy around the transient regime, while its relative accuracy improves in the Haar-typical regime t=Nt=N. For matchgate circuits, the overhead remains size dependent at t=N2t=N^2. Since these circuits remain efficiently simulable using covariance matrices, a large matrix-product-state representation cost does not by itself imply classical computational hardness.

Four-panel plots showing the spectral compression gap

Figure S2. Spectral compression gap in chaotic and matchgate dynamics. (a) Ensemble mean of δϵ=log⁡2χϵ−S\delta_{\epsilon}=\log_2\chi_{\epsilon}-S during chaotic circuit evolution, where χϵ\chi_{\epsilon} is the smallest Schmidt rank retaining total probability at least 1−ϵ1-\epsilon across the contiguous equal-half cut. (b) Mean of the statewise rank ratio χϵ/2S(ρA)=2δϵ\chi_{\epsilon}/2^{S(\rho_A)}=2^{\delta_{\epsilon}} at the sampled maximum-gap time t=arg⁡max⁡τδϵ(τ)t=\arg\max_{\tau}\delta_{\epsilon}(\tau), selected separately for each system size and error tolerance, and at late time t=Nt=N. (c) Corresponding ensemble-mean gap during matchgate circuit evolution, plotted against t/N2t/N^2. (d) Mean statewise rank ratio for matchgate states at the measured late-time checkpoint t=N2t=N^2, as a function of NN. Solid curves show ϵ=0.01\epsilon=0.01, and faint dashed curves show ϵ=0.001\epsilon=0.001 in all panels.

Numerical evaluation of the fidelity-smoothed spread. For each simulated pure state, our goal is to find the nearby Schmidt spectrum with the smallest spread. We denote the fixed Schmidt probabilities of the simulated state by λi\lambda_i, ordered as λ1≥⋯≥λDλ>0\lambda_1\ge\cdots\ge\lambda_{D_\lambda}>0, with ∑iλi=1\sum_i\lambda_i=1. The probabilities μi\mu_i describe a trial pure state and are varied in the optimization. The Schmidt bases are aligned and both spectra are ordered, their squared fidelity is F(λ,μ)=(∑iλiμi)2F(\lambda,\mu)=(\sum_i\sqrt{\lambda_i\mu_i})^2. For a prescribed minimum fidelity f=1−ϵf=1-\epsilon, we compute

Δϵ(λ)=min⁡μi>0,∑iμi=1F(λ,μ)≥flog⁡2(∣supp⁡μ∣max⁡iμi).(18)\Delta_\epsilon(\lambda)=\underset{\substack{\mu_i>0,\,\sum_i\mu_i=1\\F(\lambda,\mu)\ge f}}{\min}\log_2\left(\lvert\operatorname{supp}\mu\rvert\max_i\mu_i\right). \tag*{(18)}

The two factors in the objective are the number of positive μi\mu_i and the largest μi\mu_i. Retaining more components can reduce the largest μi\mu_i, so both factors must be optimized.

Fidelity-maximizing spectrum at a given rank and cap. Consider a candidate rank R=∣supp⁡μ∣R=|\operatorname{supp}\mu| and impose a common upper bound cc on the trial probabilities, μi≤c\mu_i\leq c. Normalization requires c≥1/Rc\geq1/R. To determine whether this pair (R,c)(R,c) can attain the required fidelity, we find the largest fidelity achievable under these constraints. The RR positive μi\mu_i can be aligned with the RR largest λi\lambda_i, since pairing a positive μi\mu_i with a larger λi\lambda_i increases its contribution to the overlap. Maximizing the fidelity is equivalent to maximizing the overlap amplitude H(μ)=∑i=1RλiμiH(\mu)=\sum_{i=1}^{R}\sqrt{\lambda_i\mu_i}. Consider two trial probabilities below the cap. A small amount of probability can be transferred between them in either direction while preserving normalization. At the maximum, their derivatives ∂H/∂μi=λi/(2μi)\partial H/\partial\mu_i=\sqrt{\lambda_i}/(2\sqrt{\mu_i}) must therefore be equal. This gives μi=αλi\mu_i=\alpha\lambda_i with the same α\alpha for every uncapped μi\mu_i. For a capped component, μi=c\mu_i=c, probability can still be transferred from it to an uncapped component. Such a transfer would improve the overlap if the derivative at the capped component were smaller than the common derivative at the uncapped components. The latter derivative is 1/(2α)1/(2\sqrt{\alpha}), so optimality requires λi/(2c)≥1/(2α)\sqrt{\lambda_i}/(2\sqrt{c})\geq1/(2\sqrt{\alpha}), or αλi≥c\alpha\lambda_i\geq c. Combining the two cases gives

μi(R,c)={min⁡{c,αλi},i≤R,0,i>R,∑i=1Rmin⁡{c,αλi}=1.(19)\mu_i^{(R,c)}= \begin{cases} \min\{c,\alpha\lambda_i\}, & i\leq R,\\ 0, & i>R, \end{cases} \qquad \sum_{i=1}^{R}\min\{c,\alpha\lambda_i\}=1. \tag*{(19)}

The normalization equation determines how much the uncapped λi\lambda_i must be scaled after truncation and capping. Concavity of HH makes these conditions sufficient for a global maximum. At c=1/Rc=1/R, normalization directly gives μi=1/R\mu_i = 1/R for all i≤Ri \le R. We write Fmax⁡(R,c)=F(λ,μ(R,c))F_{\max}(R,c) = F(\lambda,\mu^{(R,c)}) for the fidelity of the maximizing spectrum.

Computing the smoothed spread. We first determine which trial ranks RR can reach the fidelity threshold. Let PR=∑i=1RλiP_R = \sum_{i=1}^{R}\lambda_i be the mass of the RR largest original Schmidt probabilities. Cauchy–Schwarz gives F(λ,μ)≤PRF(\lambda,\mu) \le P_R for every normalized μ\mu on these indices, with equality for μi=λi/PR\mu_i = \lambda_i/P_R. The first feasible rank is therefore Rmin⁡=min⁡{R:PR≥f}R_{\min} = \min\{R:P_R \ge f\}. For each feasible RR, let mRm_R denote the smallest cap for which Fmax⁡(R,mR)≥fF_{\max}(R,m_R) \ge f. Then

Δϵ(λ)=log⁡2min⁡Rmin⁡≤R≤DλRmR.(20)\Delta_{\epsilon}(\lambda) = \log_2 \min_{R_{\min}\le R\le D_\lambda} Rm_R. \tag*{(20)}

Every integer rank in this range is eligible. The rank that minimizes RmRRm_R determines the optimal trial spectrum through Eq. (S11). To compute mRm_R, we first test the uniform trial spectrum μi=1/R\mu_i = 1/R for i≤Ri \le R. Its fidelity is (∑i=1Rλi)2/R(\sum_{i=1}^{R}\sqrt{\lambda_i})^2/R. If this reaches ff, then mR=1/Rm_R = 1/R and the spread is zero. Otherwise the smallest feasible cap satisfies Fmax⁡(R,mR)=fF_{\max}(R,m_R) = f. We find this cap by identifying how many trial probabilities μi\mu_i are capped when the fidelity reaches ff. Denote this number by jj. Since the λi\lambda_i are ordered, the first jj trial probabilities equal cc, while the remaining R−jR-j satisfy μi=αλi\mu_i = \alpha\lambda_i. Define Aj=∑i=1jλiA_j = \sum_{i=1}^{j}\sqrt{\lambda_i} and Tj,R=∑i=j+1RλiT_{j,R} = \sum_{i=j+1}^{R}\lambda_i. The first jj trial probabilities carry mass jcjc, so normalization of μ\mu gives αTj,R=1−jc\alpha T_{j,R} = 1-jc. The two groups then contribute to the overlap as

Fmax⁡(R,c)=Ajc+αTj,R=Ajc+Tj,R(1−jc).(21)\sqrt{F_{\max}(R,c)} = A_j\sqrt{c}+\sqrt{\alpha}T_{j,R} = A_j\sqrt{c}+\sqrt{T_{j,R}(1-jc)}. \tag*{(21)}

To identify jj, consider the boundary at which the jjth trial probability first reaches the cap as cc is lowered. At this boundary, c=αλjc=\alpha\lambda_j. Substituting into the normalization condition gives the boundary cap ηj(R)\eta_j^{(R)} and its overlap amplitude gj(R)g_j^{(R)}:

ηj(R)=λjjλj+Tj,R,gj(R)=ηj(R)(Aj+Tj,Rλj).(22)\eta_j^{(R)} = \frac{\lambda_j}{j\lambda_j+T_{j,R}},\qquad g_j^{(R)} = \sqrt{\eta_j^{(R)}}\left(A_j+\frac{T_{j,R}}{\sqrt{\lambda_j}}\right). \tag*{(22)}

Lowering the cap restricts the allowed μi\mu_i, so the boundary overlaps are nonincreasing. The first boundary has g1(R)=PRg_1^{(R)}=\sqrt{P_R}, while the last has gR(R)=AR/Rg_R^{(R)}=A_R/\sqrt{R}, the uniform-spectrum overlap. After the uniform test, a binary search over the integer jj therefore finds gj(R)≥f>gj+1(R)g_j^{(R)} \ge\sqrt{f} > g_{j+1}^{(R)}. The desired cap lies between these two boundaries. With jj determined, we set Eq. (S13) equal to f\sqrt{f} and solve directly for the cap. Writing x=cx=\sqrt{c} gives the quadratic equation

(Aj2+jTj,R)x2−2Ajf x+f−Tj,R=0.(23)(A_j^2+jT_{j,R})x^2-2A_j\sqrt{f}\,x+f-T_{j,R}=0. \tag*{(23)}

The smaller admissible root gives mR=x2m_R=x^2 in the interval [ηj+1(R),ηj(R)][\eta_{j+1}^{(R)},\eta_j^{(R)}]. Thus the integer search determines how many μi\mu_i are capped, and the quadratic determines the cap itself. For chaotic states, the complete ordered λ\lambda spectrum is available. We precompute AjA_j and the suffix masses Qj=∑i=j+1DλλiQ_j=\sum_{i=j+1}^{D_\lambda}\lambda_i, accumulating the latter from the smallest λi\lambda_i. Each boundary evaluation then uses Tj,R=Qj−QRT_{j,R}=Q_j-Q_R. We evaluate all feasible ranks in batches and take the minimum in Eq. (S12).

Additional optimizations for matchgate states. For matchgate states, we perform the same optimization over μ\mu. The Gaussian structure of the original state allows us to access its λi\lambda_i without first constructing the full spectrum. After removing exactly pure modes, each of the qq remaining modes contributes two probabilities pa±=(1±ξa)/2p_a^{\pm}=(1\pm\xi_a)/2. The 2q2^q Schmidt probabilities λi\lambda_i are the products obtained by choosing one factor pa+p_a^+ or pa−p_a^- from each mode.

We first use the mode factors to test whether a rank-one or full-support uniform trial spectrum μ\mu already reaches ff. Either case gives zero spread. Otherwise, we split the modes into two groups of approximately equal size and form the spectrum of each group. Each λi\lambda_i is then a product of one entry from each list. By counting products above a threshold in the sorted lists, we select the KK largest λi\lambda_i. These entries retain their original probabilities, so PK=∑i=1KλiP_K=\sum_{i=1}^{K}\lambda_i is their mass in the full spectrum. Here KK counts the original probabilities λi\lambda_i available for the calculation, whereas RR counts the positive trial probabilities μi\mu_i. If PK<fP_K<f, even rank KK cannot reach the target and we enlarge KK. Once PK≥fP_K\ge f, we evaluate mRm_R on an initial sparse set of feasible ranks up to KK. Let BB be the smallest value of RmRRm_R found so far. Two bounds allow us to exclude further ranks.

First, the minimum feasible cap mRm_R is nondecreasing with the allowed support size. For two evaluated ranks L<UL<U, every intermediate rank satisfies

RmR≥(L+1)mU,L<R<U.(24)Rm_R \ge(L+1)m_U,\qquad L<R<U. \tag*{(24)}

If this lower bound exceeds BB, no rank in the interval can improve the result. Otherwise we evaluate the midpoint and test the two resulting intervals. Second, a rank R>KR>K could still give a smaller product by allowing a sufficiently small cap. To bound this possibility, consider the first LL indices. Their original mass is PL=∑i=1LλiP_L = \sum_{i=1}^{L} \lambda_i, and their trial mass is ML=∑i=1LμiM_L = \sum_{i=1}^{L} \mu_i. Applying Cauchy–Schwarz separately to these indices and their complement gives f≤PLML+(1−PL)(1−ML)\sqrt{f} \le\sqrt{P_L M_L} + \sqrt{(1-P_L)(1-M_L)} for every fidelity-feasible μ\mu. Solving this inequality for the minimum possible MLM_L and using ML≤Lmax⁡iμiM_L \le L\max_i \mu_i yields the global cap lower bound

mLB=max⁡1≤L≤min⁡(K,Lmax⁡)[PLf−(1−PL)(1−f)]+2L,[x]+=max⁡{0,x}.(25)m_{\mathrm{LB}} = \max_{1 \le L \le\min(K,L_{\max})} \frac{\left[\sqrt{P_L f} - \sqrt{(1-P_L)(1-f)}\right]_{+}^{2}}{L}, \qquad[x]_{+} = \max\{0,x\}. \tag*{(25)}

Here Lmax⁡L_{\max} limits the number of leading sets used to compute the bound. Every unseen rank R>KR > K has objective RmR≥(K+1)mLBR m_R \ge(K+1)m_{\mathrm{LB}}. If this exceeds BB, the current minimum is global. Otherwise we double KK and repeat until the remaining ranks are excluded or the full spectrum has been evaluated.

References

  1. [1]M. P. Fisher, V. Khemani, A. Nahum, and S. Vijay, Random quantum circuits, Annual Review of Condensed Matter Physics 14, 335 (2023).DOI
  2. [2]A. Nahum, J. Ruhman, S. Vijay, and J. Haah, Quantum entanglement growth under random unitary dynamics, Physical Review X 7, 031016 (2017).
  3. [3]R. Horodecki, P. Horodecki, M. Horodecki, and K. Horodecki, Quantum entanglement, Rev. Mod. Phys. 81, 865 (2009).DOI
  4. [4]H. Li and F. D. M. Haldane, Entanglement spectrum as a generalization of entanglement entropy: Identification of topological order in non-abelian fractional quantum Hall effect states, Phys. Rev. Lett. 101, 010504 (2008).DOI
  5. [5]G. Vidal, Efficient simulation of one-dimensional quantum many-body systems, Phys. Rev. Lett. 93, 040502 (2004).pubmed.ncbi.nlm.nih.gov/15323740
  6. [6]N. Schuch, M. M. Wolf, F. Verstraete, and J. I. Cirac, Entropy scaling and simulability by matrix product states, Phys. Rev. Lett. 100, 030504 (2008).
  7. [7]U. Schollwöck, The density-matrix renormalization group in the age of matrix product states, Annals of Physics 326, 96 (2011).DOI
  8. [8]E. Chitambar and G. Gour, Quantum resource theories, Reviews of Modern Physics 91, 025001 (2019).DOI
  9. [9]P. Lipka-Bartosik, H. Wilming, and N. H. Y. Ng, Catalysis in quantum information theory, Reviews of Modern Physics 96, 025005 (2024).
  10. [10]M. A. Nielsen, Conditions for a class of entanglement transformations, Physical Review Letters 83, 436 (1999).DOI
  11. [11]G. Vidal, Entanglement of pure states for a single copy, Physical Review Letters 83, 1046 (1999).DOI
  12. [12]W. van Dam and P. Hayden, Universal entanglement transformations without communication, Physical Review A 67, 060302 (2003).DOI
  13. [13]D. Jonathan and M. B. Plenio, Entanglement-assisted local manipulation of pure quantum states, Physical Review Letters 83, 3566 (1999).DOI
  14. [14]R. Jozsa and B. Schumacher, A new proof of the quantum noiseless coding theorem, Journal of Modern Optics 41, 2343 (1994).
  15. [15]B. Schumacher, Quantum coding, Physical Review A 51, 2738 (1995).pubmed.ncbi.nlm.nih.gov/9911903
  16. [16]C. H. Bennett, H. J. Bernstein, S. Popescu, and B. Schumacher, Concentrating partially entangled states by local operations, Physical Review A 53, 2046 (1996).
  17. [17]H.-K. Lo and S. Popescu, Classical communication cost of entanglement manipulation: Is entanglement an inter-convertible resource?, Physical Review Letters 83, 1459 (1999).DOI
  18. [18]P. Hayden and A. Winter, Communication cost of entanglement transformations, Physical Review A 67, 012326 (2003).DOI
  19. [19]A. W. Harrow, Entanglement spread and clean resource inequalities, in XVIth International Congress on Mathematical Physics (World Scientific, 2010) pp. 536–540.
  20. [20]A. W. Harrow and P. W. Shor, Time reversal and exchange symmetries of unitary gate capacities, IEEE Transactions on Information Theory 56, 462 (2010).arxiv.org/abs/quant-ph/0511219
  21. [21]C. Cao, G. Cheng, A. Hamma, L. Leone, W. Munizzi, and S. F. Oliviero, Gravitational backreaction is magical, PRX Quantum 6, 040375 (2025).
  22. [22]A. Gu, S. F. Oliviero, and L. Leone, Magic-induced computational separation in entanglement theory, PRX Quantum 6, 020324 (2025).DOI
  23. [23]D. Gottesman, The heisenberg representation of quantum computers, arXiv preprint quant-ph/9807006 (1998).arxiv.org/abs/quant-ph/9807006
  24. [24]S. Bravyi and A. Kitaev, Universal quantum computation with ideal Clifford gates and noisy ancillas, Phys. Rev. A 71, 022316 (2005).DOI
  25. [25]V. Veitch, S. A. Hamed Moussavian, D. Gottesman, and J. Emerson, The resource theory of stabilizer quantum computation, New Journal of Physics 16, 013009 (2014).DOI
  26. [26]M. Howard and E. Campbell, Application of a resource theory for magic states to fault-tolerant quantum computing, Physical Review Letters 118, 090501 (2017).DOI
  27. [27]L. Leone, S. F. E. Oliviero, and A. Hamma, Stabilizer Rényi entropy, Physical Review Letters 128, 050402 (2022).DOI
  28. [28]C. D. White, C. Cao, and B. Swingle, Conformal field theories are magical, Physical Review B 103, 075145 (2021).DOI
  29. [29]Z.-W. Liu and A. Winter, Many-body quantum magic, PRX Quantum 3, 020333 (2022).DOI
  30. [30]J. J. Minar, A. Schäfer, L. Schmotzer, and C. Seidl, The magic barrier: A universal feature of quantum thermalization, arXiv:2610.20768 (2026).arxiv.org/abs/2610.20768
  31. [31]J. Odavić, M. Viscardi, and A. Hamma, Stabilizer entropy in nonintegrable quantum evolutions, Physical Review B 112, 104301 (2025).DOI
  32. [32]L. Ebner, G. Cataldi, K. Xu, J. C. Halimeh, and E. Tirrito, Global and nonlocal magic of quantum many-body scars, arXiv:2610.00129 (2026).arxiv.org/abs/2610.00129
  33. [33]M. Karjula, T. Ojanen, T. Ala-Nissila, and M. N. Ivaki, The ebbs and flows of quantum learning and sensing, arXiv:2608.20155 (2026).arxiv.org/abs/2608.20155
  34. [34]D. Fattal, T. S. Cubitt, Y. Yamamoto, S. Bravyi, and I. L. Chuang, Entanglement in the stabilizer formalism, arXiv:quant-ph/0406168 (2004).DOI
  35. [35]P. Sierant, Exact quantification of nonlocal magic, arXiv:2608.28563 (2026).arxiv.org/abs/2608.28563
  36. [36]M. Viscardi, L. Leone, and A. Hamma, Non-local magic: closed-form solution and equivalence with magic of purification, arXiv:2609.04119 (2026).arxiv.org/abs/2609.04119
  37. [37]M. Karjula, T. Ojanen, K. Pöyhönen, T. Ala-Nissila, and M. N. Ivaki, Universal entanglement embezzlement and divergent nonlocal magic from generic local chaotic quantum evolution, arXiv:2609.18691 (2026).arxiv.org/abs/2609.18691
  38. [38]S. Aditya, P. Sierant, and X. Turkeshi, Nonlocal magic spreading in many-body quantum dynamics: From chaotic evolution to quasi-particle picture in integrable models, arXiv:2609.20951 (2026).
  39. [39]S.-X. Zhang, S. Liu, and Y.-Q. Chen, Entanglement embezzlement from diffusive hydrodynamics, arXiv:2609.26362 (2026).arxiv.org/abs/2609.26362
  40. [40]E. Zanoni, T. Theurer, and G. Gour, Complete characterization of entanglement embezzlement, Quantum 8, 1368 (2024).DOI
  41. [41]C. H. Bennett, G. Brassard, C. Crépeau, R. Jozsa, A. Peres, and W. K. Wootters, Teleporting an unknown quantum state via dual classical and Einstein-Podolsky-Rosen channels, Physical Review Letters 70, 1895 (1993).pubmed.ncbi.nlm.nih.gov/10053414
  42. [42]A. Anshu, A. W. Harrow, and M. Soleimanifar, Entanglement spread area law in gapped ground states, Nature Physics 18, 1362 (2022).DOI
  43. [43]G. Vidal, D. Jonathan, and M. A. Nielsen, Approximate transformations and robust manipulation of bipartite pure-state entanglement, Physical Review A 62, 012304 (2000).DOI
  44. [44]I. George and E. Chitambar, Reexamination of quantum state transformations with zero communication, Physical Review A 109, 062418 (2024).
  45. [45]E. Tirrito, P. S. Tarabunga, G. Lami, T. Chanda, L. Leone, S. F. E. Oliviero, M. Dalmonte, M. Collura, and A. Hamma, Quantifying nonstabilizerness through entanglement spectrum flatness, Phys. Rev. A 109, L040401 (2024).arxiv.org/abs/2304.01175
  46. [46]P.-Y. Chang, X. Chen, S. Gopalakrishnan, and J. H. Pixley, Evolution of entanglement spectra under generic quantum dynamics, Physical Review Letters 123, 190602 (2019).DOI
  47. [47]L. Zhang, S.-X. Zhang, H. Fan, and S. Liu, Revealing entanglement-growth mechanisms through the magic barrier, arXiv:2607.09875 (2026).arxiv.org/abs/2607.09875
  48. [48]V. A. Marchenko and L. A. Pastur, Distribution of eigenvalues for some sets of random matrices, Mathematics of the USSR-Sbornik 1, 457 (1967).
  49. [49]K. Zyczkowski and H.-J. Sommers, Induced measures in the space of mixed quantum states, Journal of Physics A: Mathematical and General 34, 7111 (2001).DOI
  50. [50]I. Dumitriu and A. Edelman, Matrix models for beta ensembles, Journal of Mathematical Physics 43, 5830 (2002).
  51. [51]I. Nechita, Asymptotics of random density matrices, Annales Henri Poincaré 8, 1521 (2007).DOI
  52. [52]N. Bao, G. Penington, J. Sorce, and A. C. Wall, Beyond toy models: distilling tensor networks in full ads/cft, Journal of High Energy Physics 2019, 69 (2019).DOI
  53. [53]A. L. Shaw, Z. Chen, J. Choi, D. K. Mark, P. Scholl, R. Finkelstein, A. Elben, S. Choi, and M. Endres, Benchmarking highly entangled states on a 60-atom analogue quantum simulator, Nature 628, 71 (2024).DOI
  54. [54]For power-of-two Schmidt ranks, one also has D ≤ S_0 − S_{1/2} ≤ δ_0 ≤ Δ_0.
  55. [55]R. Jozsa and A. Miyake, Matchgates and classical simulation of quantum circuits, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 464, 3089 (2008).
  56. [56]B. M. Terhal and D. P. DiVincenzo, Classical simulation of noninteracting-fermion quantum circuits, Physical Review A 65, 032325 (2002).DOI
  57. [57]I. Peschel, Calculation of reduced density matrices from correlation functions, Journal of Physics A: Mathematical and General 36, L205 (2003).DOI
  58. [58]A. Botero and B. Reznik, BCS-like modewise entanglement of fermion Gaussian states, Physics Letters A 331, 39 (2004).DOI
  59. [59]J. Surace and L. Tagliacozzo, Fermionic Gaussian states: an introduction to numerical approaches, SciPost Physics Lecture Notes, 54 (2022).DOI
  60. [60]M. Collura, B. Béri, and E. Tirrito, Nonlocal nonstabilizerness in free fermion models, arXiv:2604.27055 (2026).arxiv.org/abs/2604.27055
  61. [61]D. Iannotti, B. Magni, R. Cioli, A. Hamma, and X. Turkesh, Non-local magic resources for fermionic Gaussian states, arXiv:2604.27049 (2026).arxiv.org/abs/2604.27049
  62. [62]B. Magni, D. Iannotti, and R. Travaglino, Entanglement spectrum and magic in higher-dimensional free fermionic systems, arXiv:2609.36008 (2026).arxiv.org/abs/2609.36008
  63. [63]E. Bianchi, L. Hackl, and M. Kieburg, Page curve for fermionic Gaussian states, Physical Review B 103, L241118 (2021).DOI
  64. [64]J. Romero, J. P. Olson, and A. Aspuru-Guzik, Quantum autoencoders for efficient compression of quantum data, Quantum Science and Technology 2, 045001 (2017).DOI

Paper details

vlnc:W_3cfsdetz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

GPT-6 family

View statement
Looking for related papers…
Contents