Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

The paper claims an end-to-end quantum algorithm that estimates a nonlinear plasma observable with superquadratic speedup and exponential memory savings over Fourier–Hermite spectral methods, with rigorous convergence in a certified weakly

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 02:30 UTC pith:GSWHPMAZ

load-bearing objection Serious, honestly scoped quantum algorithm for weakly nonlinear plasma, but Theorem 1 as stated is proven for a real-space-sampled, regularized variant of the nonlinearity, and that gap is the main thing to referee. the 3 major comments →

arxiv 2607.14308 v2 pith:GSWHPMAZ submitted 2026-07-15 quant-ph physics.plasm-ph

An end-to-end quantum algorithm for weakly nonlinear plasma physics with superquadratic speedup

Bjorn K. Berntson , David Jennings , Matteo Lostaglio , Scott Parker This is my paper
classification quant-ph physics.plasm-ph MSC 81P6835Q8365M70
keywords quantum algorithmVlasov-PoissonCarleman linearizationblock-encodingplasma simulationLyapunov stabilitykinetic energy estimationnonlinear ODE
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to prove that a fully nonlinear kinetic plasma model—the three-dimensional Vlasov–Poisson system with adiabatic electrons and Krook collisions—can, in a certified weakly nonlinear regime, be simulated end-to-end on a quantum computer with exponentially less memory and a superquadratic time improvement over the standard Fourier–Hermite spectral method. The central move is to rephrase the nonlinear dynamics as a high-dimensional quadratic ODE, then embed it in a much larger linear system using a Carleman expansion. The embedding is made convergent by rescaling coordinates with a Lyapunov transform derived from the plasma free energy. The paper also develops a hierarchical block-encoding that loads the dense screened-field interaction without destroying the speedup, and an extraction step that reads out the time-averaged kinetic energy from the Carleman history state. A sympathetic reader would take the paper's claim to be: this is the first controlled nonlinear plasma benchmark in which every quantum bottleneck—nonlinearity, dense data loading, and information extraction—has a rigorous end-to-end solution.

Core claim

For the Galerkin-truncated system, the paper constructs a quantum algorithm that estimates the spacetime-averaged kinetic energy to epsilon-additive error with O~(N_F sqrt(N_H) polylog(T/epsilon)/epsilon) gates and O~(log(N_F sqrt(N_H) T) log(1/epsilon)) qubits. This follows from three linked ingredients: a plasma free energy that acts as a Lyapunov functional, which after a non-unitary rescaling makes the linearized dynamics strictly dissipative and forces exponential Carleman convergence when the initial nonlinearity satisfies an explicit bound; a hierarchical block-encoding of the dense nonlinear interaction that exploits the real-space decay of the screened Yukawa-like field to achieve n

What carries the argument

The load-bearing object is the pair (P, Gbar_2): a Lyapunov matrix P, constructed from the quadratic part of the plasma free energy, which symmetrizes the linearized Vlasov–Poisson dynamics so that its logarithmic norm equals -nu_bar; and the regulated real-space-sampled nonlinear interaction Gbar_2, which replaces the original Fourier–Hermite quadratic term Fbar_2 after a hard-core cutoff and Nyquist sampling. The Carleman matrix Abar built from these objects is the central mechanism: because its log-norm is bounded away from zero, the truncated linear system can be solved and fast-forwarded by a quantum ODE solver at polylog(T/epsilon) cost. The hierarchical block-encoding of the two decay

Load-bearing premise

Theorem 1 is a theorem about the real-space-sampled, singularity-regulated operator Gbar_2, which the paper substitutes for the Fourier-truncated Fbar_2 of Problem 1 ('we simply take Fbar_2 = Gbar_2'); the error between the two is asserted to vanish as N_F grows but is not bounded explicitly, so if the aliasing/regulation gap is not negligible at finite resolution, the proven speedup is for a different model than the one stated.

What would settle it

Simulate (or on a small scale, exactly diagonalize) the truncated Vlasov–Poisson Galerkin system with the original Fbar_2 and with the regulated, real-space-sampled Gbar_2 at the same (N_F, N_H, tau, nu_bar, L); compute the operator-norm gap ||Fbar_2 - Gbar_2|| and the resulting differences in trajectories and in the time-averaged kinetic energy. If the gap is not below (or comparable to) epsilon at the targeted resolution, the proof of Theorem 1 does not transfer to Problem 1. A cheaper classical check: evaluate ||Fbar_2 - Gbar_2|| as a function of N_F at fixed physical parameters and look fo

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Any initial datum in the certified window (condition (10) or its Gaussian analogue) can be simulated with the stated polylogarithmic-in-T gate and qubit counts, yielding exponential memory compression (from O(N_H^3 N_F^3) to O(log(N_F sqrt(N_H) T) log(1/epsilon))) against the Fourier–Hermite spectral baseline.
  • The quantum gate complexity beats the explicit Fourier–Hermite spectral solver by a quartic factor in N_F and a seventh-power factor in N_H (superquadratic overall), at the price of a worse dependence on epsilon; expressing the epsilon-scaling via Sobolev regularity gives Q_op = O~(epsilon^{-3/2}) when the Hermite regularity is s_H = 1.
  • The Carleman truncation order need only be logarithmic in 1/epsilon within the certified regime, so the linear-embedding overhead is not an asymptotic blocker.
  • The same hierarchical block-encoding theorem extends to pure Coulomb kernels, screened plasma interactions, and square kernels, recovering the threshold decay p = 3/2 for convolution-type interactions and p = 3 for square kernels.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the aliasing/regulation gap between Gbar_2 and Fbar_2 is not tiny at the resolutions used in practice, Theorem 1 would prove a speedup for a different discrete model; a direct numerical comparison of the spectra (or trajectories) of Fbar_2 and Gbar_2 at finite N_F would settle this without any quantum hardware.
  • The certified nonlinearity bound scales as (N_F N_H)^(-1/2), so the a priori guarantee recedes as resolution grows; whether the practical convergence threshold tracks the much larger weakly nonlinear regimes shown in the paper's Fig. 1 is a testable question that classical Carleman-convergence studies on the truncated ODE can answer.
  • The 'integrator' extraction idea—using the Carleman generator Abar to build higher-order quadrature from the history state—should apply to any dissipative quadratic ODE whose quantum solver outputs a Taylor-history state, not only to Vlasov–Poisson; observable estimation beyond kinetic energy (e.g., flux functionals) could be attacked with the same overlap-with-R_k(Abar h) recipe.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper presents an end-to-end quantum algorithm for a weakly nonlinear 3D Vlasov–Poisson–Krook plasma model with adiabatic electrons. After Fourier–Hermite truncation, the dynamics is a quadratic ODE. The authors use a plasma free energy to construct a Lyapunov transform, apply a Carleman linearization with an R-number convergence criterion, develop a hierarchical block-encoding for the dense screened nonlinear interaction, and extract the spacetime-averaged kinetic energy from a Carleman history state. The central claim (Theorem 1) is that Problem 1 can be solved with \tilde O(N_F N_H^{1/2} polylog(T/\epsilon)/\epsilon) gates and \tilde O(\log(N_F N_H^{1/2} T)\log(1/\epsilon)) qubits, whenever the initial nonlinearity satisfies Eq. (10), yielding exponential memory savings and a superquadratic time improvement over an explicit Fourier–Hermite spectral solver. The Carleman convergence machinery, the norm estimates (Props. 3–4), and the hierarchical block-encoding theorem (Theorem 4) are coherent and checkable as far as they go. However, a load-bearing substitution between the Fourier-truncated nonlinearity F2 stated in Problem 1 and the real-space-sampled, regularized operator G2 used in the block-encoding proof is asserted but never quantified. There is also an apparent mismatch between the Lyapunov scaling used in the Carleman analysis and the scaling used in the observable-extraction step.

Significance. If the missing quantitative bounds are supplied, this would be a landmark result: the first end-to-end quantum algorithm for a nonlinear kinetic plasma benchmark with a priori Carleman convergence guarantees, dense-field block-encoding, and observable extraction. The technical contributions are substantial and independently useful: the connection between the plasma free energy and the Lyapunov R-number criterion, the explicit norm estimates for the Fourier–Hermite operators, and the general two-kernel hierarchical block-encoding theorem. The paper also honestly states the certified nonlinearity window is very small (\phi_max ~ 10^{-8}–10^{-5} in the representative cases) and does not overclaim practical applicability. The value of the paper, however, depends on resolving the gap between the stated problem and the model for which the quantum speedup is proven.

major comments (3)
  1. [Appendix E3–E4, Eqs. (E59)–(E75); pp. 37 and 45] Theorem 1 is stated for Problem 1, whose quadratic interaction is F2 defined in Eqs. (C14)/(C18)–(C19). The efficient block-encoding with alpha=O(sqrt(NF NH)), which drives the claimed speedup, is proven for G2, the real-space-sampled regularized kernel: Eq. (E72) defines G2, and p. 45 states 'we simply take F2 = G2 in what follows'. The text also notes that the original F2 'does not have the required properties to directly apply hierarchical block-encoding methods' (p. 37). No explicit bound is given for ||G2 - (Pi_NF \otimes I) F2 (Pi_NF \otimes I)^{\otimes 2}|| as a function of NF, the cutoff scale a, and L. The hard-core regulation gives only O(q^{-3}) Fourier decay (p. 43), so the gap need not be exponentially small. This is load-bearing because both the Carleman R-number analysis and the gate count use the norm and block-encoding normalization of the nonlinear term. Without a quant
  2. [II C 1, Eqs. (107)–(110) vs. II D 4, Eq. (176)] The Carleman analysis uses the Lyapunov transform \bar u = (1+R_thr)/(2||\tilde u0||) \tilde u, with \tilde u0 = P^{1/2}u0, leading to \bar F2 = (2||\tilde u0||/(1+R_thr))\tilde F2 and ||\bar u0|| = (1+R_thr)/2 (Eqs. (107)–(110)). In contrast, Eq. (176) defines \bar u = \gamma P^{1/2} u with \gamma = (1+R_thr)/(2||u0|| sqrt(1+1/tau)). Since ||P^{1/2}u0|| <= sqrt(1+1/tau)||u0||, these two definitions coincide only in the worst-case equality. In general, the Eq. (176) scaling yields a smaller ||\bar u0|| and a correspondingly larger \bar F2. The proof that mu(\bar A) < 0 in Eq. (120) uses Eq. (119), which is derived from the exact scaling of Eq. (109); this does not apply to the Eq. (176) scaling. Moreover, the information-extraction step uses \bar\ell_K = \gamma^{-1} P^{-1/2}\ell_K; if the simulated history state uses the Eq. (107) scaling, this introduces a multiplicative bias in the est
  3. [Appendix F, Theorem 4 and Appendix F.6] The end-to-end gate count in Theorem 1 assumes access to oracles enumerating the hierarchical supports of the spatial kernels and returning b-bit values of sigma, K^a, and W^a. Theorem 4 bounds the block-encoding cost in terms of these oracles, but their implementation cost is not included. For the plasma nonlinearity, the matrix entries are defined through integrals over the Feynman parameter (Eq. (F83)); the text says they can be realized via coherent arithmetic with quadrature, but adds that 'a detailed analysis would need to be done' (Appendix F.6). Since Qop in Theorem 1 is a claimed gate count, the missing oracle cost leaves the end-to-end complexity claim incomplete. The authors should provide explicit polylog(N) bounds for the support and value oracles, including the quadrature depth and precision, or state Theorem 1 under an oracle model whose cost is explicitly separated from t
minor comments (4)
  1. [Eq. (E65)] The diagonal value of sigma'_{j,j''} at j=j'' appears to be off by a factor of pi: evaluating \int_0^1 \sqrt{s/(1-s)} ds = pi/2 gives 1/(8\Delta x), not 1/(8\pi\Delta x). Please check and correct.
  2. [References] Reference [43] has an arXiv identifier typo: '1806.018384' should presumably be '1806.01838'.
  3. [Eq. (145)] The statement ||\tilde F1||_max = (N_F/L \sqrt{N_H} + \bar\nu) drops constant terms and the extra \tau-dependent contributions from Eq. (C16). Clarify whether this is an asymptotic statement or an equality up to constants.
  4. [General] The paper uses both '\widetilde O' and 'eO' notation inconsistently; standardize to a single asymptotic notation. Also, the figures' axes labels are too small to read in the printed version; please enlarge them.

Circularity Check

1 steps flagged

Theorem 1's claim to solve Problem 1 reduces to the stipulated identity F̃2 = G̃2; the speedup is proven for the substituted sampled operator, not for the stated Fourier–Hermite nonlinearity.

specific steps
  1. self definitional [Appendix E.4, Eq. (E74); Appendix E intro, p.37; used in Theorem 1 (Eqs. (11)–(12))]
    "we simply take $\tilde{F}_2 = \tilde{G}_2$ in what follows. This implies that a unitary block-encoding $U_K$ of $K$ with scale factor $\alpha_K$ gives a unitary block-encoding $U_{\tilde{F}_2}$ of $\tilde{F}_2$ with scale factor $\alpha_{\tilde{F}_2} = \Delta_x^{9/2} \alpha_K$"

    Problem 1's quadratic term is the Fourier–Hermite $\tilde{F}_2$ of Eq. (C14)/(74). The hierarchical block-encoding with the claimed optimal $\alpha_{\tilde{F}_2}=O(\sqrt{N_FN_H})$ is proven only for $\tilde{G}_2$, the discrete Fourier transform of the real-space-sampled, hard-core-regulated kernel; the text even notes $\tilde{F}_2$ 'does not have the required properties to directly apply hierarchical block-encoding methods'. Setting $\tilde{F}_2=\tilde{G}_2$ by fiat, with only an asymptotic assertion that aliasing errors decrease, makes Theorem 1's 'Problem 1 solved' claim true by construction for the substituted operator, not by a bound on the gap. The speedup and memory claims therefore reduce, as stated, to the asserted equality rather than to a derived property of the original $\tilde{

full rationale

The main derivation has substantial independent content: the Lyapunov transform P, the norm bounds, the Carleman embedding, the hierarchical block-encoding theorem, and the observable-extraction routine are all derived in the paper and do not fit parameters to data. The φmax thresholds are analytic sufficient conditions, and the R_P condition is an input assumption, not an output of fitting. The self-citations to Refs. [6,37] are general convergence criteria applied to this system, so by themselves they are not circular. However, the central theorem as stated solves Problem 1, whose nonlinearity is the original Fourier–Hermite operator F̃2, while the rigorously analyzed and block-encoded operator is G̃2, the sampled and regulated kernel. The text collapses the two by 'we simply take F̃2 = G̃2', with no explicit bound on the aliasing gap, and explicitly concedes that F̃2 itself lacks the properties needed for the hierarchical block-encoding. Because the advertised α=O(√(NF NH)) normalization and all downstream complexity claims are proven for G̃2, the end-to-end theorem's claim about Problem 1 reduces to this stipulated equality unless the missing error bound is supplied. The paper's alternative—redefining the benchmark as (F̃1, G̃2, g(0))—would remove the circularity but would change the problem that Theorem 1 states. Hence the partial circularity score of 6.

Axiom & Free-Parameter Ledger

2 free parameters · 6 axioms · 0 invented entities

Everything the central claim rests on that the reader didn't pay for upstream: the R-number machinery from [6], the ODE-solver bound from [37], the real-space-sampling substitution (the most fragile), the regularity assumptions, and the oracle model. The paper's own derived objects (P, µ(F̄_1)=−ν̄, norm bounds, mass conservation) are proven in-appendices and not counted as axioms. No new physical entities (particles, forces, dimensions) are postulated; the Lyapunov-transformed fields ϕ_L, σ_L are derived mathematical kernels, hence the empty invented_entities list.

free parameters (2)
  • Lyapunov rescaling constant R_thr
    Free constant in (0,1) used to set the Lyapunov transform (31) and initial-norm bound (33). Any value <1 yields the R_P<1 criterion; the choice does not affect asymptotics, and no sensitivity analysis is given.
  • Hard-core regulation scale a (and boundary regulation 1/|j−j'|→1 at j=j')
    The real-space kernel singularity is regulated by a cutoff at ||x|| ≤ a (App. E3, E4). The scale is not specified; the analysis needs a ≈ Δx so that Ḡ_2 → F̄_2 as N_F → ∞, but the error is asserted, not bounded. Different regulation changes the Fourier decay rate O(1/q^m) and hence the aliasing gap.
axioms (6)
  • domain assumption Lyapunov R-number criterion of [6]: R_P < 1 ⟹ exponential Carleman convergence and µ(Ā)<0 (Eqs. (38), (120), (124))
    Imported from [6] (co-authored by two present authors); the theorem is not reproduced and its hypotheses are only quoted. It is the load-bearing convergence guarantee of the paper.
  • domain assumption Log-norm estimate µ(Ā) < µ(F̄_1)/2 = −ν̄/2 (Eq. (120), from [6, Eq. B.39] and [34, Lemma 15])
    Gives the T-independent ||L^{-1}|| bound in Appendix I, hence the fast-forwarding in T; cited, not derived.
  • standard math Lemma 10, the rectangular-tensor norm bound adapted from [59]
    Used in the proof of Proposition 4 to bound ||F̃_2||; cited, proof not reproduced.
  • ad hoc to paper Real-space sampling substitution: Ḡ_2 ≈ F̄_2 up to negligible aliasing under smooth regularization; hard-core cutoff at scale a ≈ Δx
    Appendix E3–E4 ('we simply take F̄_2 = Ḡ_2 in what follows'). The HBE normalization is proven for Ḡ_2; closeness to the Problem-1 interaction is asserted, not bounded.
  • domain assumption Assumption 1: uniform Fourier–Hermite Sobolev regularity of the solution on [0,T] (App. H5)
    Controls the Galerkin truncation error (Prop. 20); standard for spectral methods but not proven for this model.
  • domain assumption Oracle model: coherent arithmetic for Feynman-parameter kernel integrals and hierarchical sparse-access unitaries cost polylog(N)
    Required for the Õ(·) gate counts; the resource analysis is deferred ('no obstacle... a detailed analysis would need to be done', App. F6, p. 58).

pith-pipeline@v1.3.0-alltime-deepseek · 67446 in / 27169 out tokens · 248777 ms · 2026-08-02T02:30:34.168718+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of An end-to-end quantum algorithm for weakly nonlinear plasma physics with superquadratic speedup." pith.science (2026). https://pith.science/paper/GSWHPMAZ

@misc{pith2026260714308,
  author       = {Pith},
  title        = {Pith review of: An end-to-end quantum algorithm for weakly nonlinear plasma physics with superquadratic speedup},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GSWHPMAZ}},
  note         = {Machine review of arXiv:2607.14308}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Nonlinear kinetic plasma simulation is high-dimensional and classically demanding, while quantum algorithms face different bottlenecks: embedding nonlinear dynamics into a linear computation, loading dense field-interaction data, and efficiently extracting information. We present an end-to-end quantum algorithm, with rigorous convergence guarantees, for a weakly nonlinear kinetic plasma model. The system describes a 3D electron-ion plasma with adiabatic electrons, kinetic ions, Debye screening, and Krook relaxation. After Fourier-Hermite truncation, the dynamics reduces to a high-dimensional quadratic ordinary differential equation. To tackle quantum bottlenecks we combine three key ingredients. First, we use a plasma free energy to identify a Lyapunov transform under which a Carleman linear embedding converges exponentially in the truncation order within a certified weakly nonlinear regime. Second, we develop a hierarchical block-encoding protocol for dense matrices, exploiting the spatial decay of the field to avoid polynomial overhead from sparse access encodings. Third, we introduce a subroutine for information extraction that exploits nonlinear components encoded in the full Carleman history state to improve the estimation of linear observables. We construct a quantum algorithm to estimate the spacetime-averaged kinetic energy using $\widetilde{O}\!\left( N_F N_H^{1/2} \operatorname{polylog}\!\left(\frac{T}{\epsilon}\right)\frac{1}{\epsilon}\right)$ gates and $\widetilde{O}\!\left(\log\!\left(N_F N_H^{1/2}T\right)\log\!\left(\frac{1}{\epsilon}\right)\right)$ qubits, where $N_F$ and $N_H$ are the Fourier and Hermite cutoffs. Relative to a Fourier-Hermite spectral solver, this yields exponential memory savings and superquadratic improvements in time. Together, these results establish a controlled nonlinear plasma benchmark for quantum simulation.

Figures

Figures reproduced from arXiv: 2607.14308 by Bjorn K. Berntson, David Jennings, Matteo Lostaglio, Scott Parker.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p049_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Improved Convergence of Carleman-Embedded Quantum Algorithm for the Vlasov-Poisson System

    quant-ph 2026-07 conditional novelty 6.0

    Shifting the Vlasov–Poisson distribution by a Maxwellian and minimizing over Lyapunov matrices extends Carleman embedding convergence to physically allowed collision frequencies for Landau-damping-type initial data.

Reference graph

Works this paper leans on

31 extracted references · cited by 1 Pith paper

  1. [1]

    Analysis of Lyapunov-transformed electrostatic field We start our analysis from theF 2 after Hermite truncation but before Fourier truncation (˜F2)n,α|n′,α′,n′′,α′′ =δ n,n′+n′′ (˜F′ 2)α|n′,α′,n′′,α′′ ,α,α ′,α ′′ ∈ NNF ,n,n ′n′′ ∈Z 3.(E1) Define the kernel as KL α|α′,α′′ (x|x′,x ′′) = 1 L9/2 X n,n′,n′′∈Z3 ei(ξn·x−ξn′ ·x′−ξn′′ ·x′′)(˜F′ 2)α|n′,α′,n′′,α′′δn,...

  2. [2]

    The precise statement is now given

    The kernelKto be block-encoded The form of the distributionsϕ L(x),σ L(x) on the 3-torus can be related to that of the unboundedϕ(x),σ(x) via periodization by images. The precise statement is now given. Proposition 13(Periodization ofϕ(x) andσ(x)).The distributionsϕ L andσ L onT 3 L are, in the sense of distribu- tions, the periodizations ofϕandσ, respect...

  3. [3]

    We make the natural choice of a hard-core cut-off for∥x∥ ≤a, whereais the cut-off scale

    Regulation of the real-space singularity The electrostatic field in the Lyapunov picture blows up forx→0, and so we must regulate it. We make the natural choice of a hard-core cut-off for∥x∥ ≤a, whereais the cut-off scale. This has the consequence of distorting the spectral data for ˜F2 in Fourier space, and thus the nonlinear interaction in the plasma mo...

  4. [4]

    , NF}3, each spatial variable is represented by a finite Fourier series

    Real-space sampling Since the spatial dependence has been truncated to the Fourier modesZ 3 NF ={−N F, . . . , NF}3, each spatial variable is represented by a finite Fourier series. We therefore evaluate the kernel (E46) on the uniform lattice xj = L Nx j,j∈Z 3 Nx , N x = 2NF + 1,∆ x := L Nx (E58) On this lattice, the exponentials{e 2πin·j/Nx }n∈Z 3 NF , ...

  5. [5]

    We assume for simplicity that a given matrix transforms definite numbers of qubits, so the dimensions are power of 2

    Block-encoding of rectangular matrices We begin by generalizing the notation of block-encodings to rectangular matrices. We assume for simplicity that a given matrix transforms definite numbers of qubits, so the dimensions are power of 2. In what follows, we shall use the notation thatunprimedindices or variables refer to therowsof a matrixA, andprimedind...

  6. [6]

    W eighted sparse encodings We now give relevant block-encodings for sparse rectangular matrices. We say a rectangular matrixA∈C 2s×2s′ is dr row-sparse if it has at mostd r non-zero elements in each row, and say it isd c column sparse if it has at mostd c non-zero elements in each column. I.e., ifA= (a mn′), then row-sparsity ofd r means that for fixedmth...

  7. [7]

    blocks” associated to well-separated points and small “blocks

    Hierarchical decomposition of the nonlinear interaction We now apply this theory to the nonlinear term ˜F2 in the plasma problem. Recall that this is in the Fourier-Hermite basis, but with a discrete Fourier transform we obtain a matrix with spatial indices that have a decaying profile in terms of the separation of points. The hierarchical decomposition w...

  8. [8]

    Split the cubic lattice inl-clusters

  9. [9]

    49 referenceadjacent(8 clusters)𝑙=4(27 clusters)𝑙=3(27 clusters)𝑙=2(7 clusters) FIG

    Select the admissible blocks inK. 49 referenceadjacent(8 clusters)𝑙=4(27 clusters)𝑙=3(27 clusters)𝑙=2(7 clusters) FIG. 4.Hierarchical decomposition of a 2-d kernel: The combinatorics of the hierarchical decomposition for a 2-d kernel, withN= 16,ℓ max = log 2 16 = 4. The red square is the reference, and the selected 2-clusters, 3-clusters and 4-clusters ha...

  10. [10]

    Remove from matrixKthe elementsK j,j ′ with indices (j,j ′) selected in step 3 and put the elements into a matrixK (l)

  11. [11]

    Increasel7→l+ 1 and go back to step 2, repeating the process on the resulting sublattice

  12. [12]

    Repeat up tol=ℓ max when the filtering process terminates

  13. [13]

    clusters

    Collect the remaining points not filtered into a single matrixK ad. For example, Fig. 4 shows the result of applying this algorithm in a 16×16 lattice. A decomposition of the kernel results by repeating the above construction for every reference particle, and defining matricesK (l) that only include the interactions from clusters selected at levell. We th...

  14. [14]

    The number of elements in a level–lcluster is|τ l jl |= 2 3(ℓmax−l) for any indexj l

  15. [15]

    The largest separation in theL ∞(N)norm between two points insideτ l jl is2 ℓmax−l −1, namely the diameter of any cluster at levell

  16. [16]

    Any pointj∈I 3 belongs to a unique clusterτ l jl at levell, where j= 2 ℓmax−ljl +j ¯l,(F24) Therefore, for anyk= 1,2,3writing components as binary strings, we have thatj k =j l kj¯l k, namelyj l k in binary is given by the leadinglbits ofj k, andj ¯l k is the remaining string ofℓ max −lbits ofj k in binary notation

  17. [17]

    The number of distinct clusters at level-lis2 3l

  18. [18]

    ,2l −1, and all clusters are disjoint

    The clusters at each levelldefine a partition ofI 3: namely,∪ jl τ l jl =I 3, where the union is over allj l = (jl 1, jl 2, jl 3)withj l k = 0, . . . ,2l −1, and all clusters are disjoint

  19. [19]

    internal

    Subsequent levels of the hierarchy have the following nesting relation for clusters: [ b:b k∈{0,1} τ l 2jl−1+b =τ l−1 jl−1 ,(F25) and hence the set of clusters over all levels forms an octree in3–dimensions, with an edge denoting set-inclusion. Proof.To see that (1) is true, note that the cluster is obtained by rangingj ¯l over all possible values. Since ...

  20. [20]

    To handle this neatly we give a theorem that is sufficiently flexible to accommodate all cases of interest here

    General HBE for two-kernel, dense rectangular matrices Note that theKthat generates our nonlinear term is a dense rectangular matrix, and the terms appearing inKcan be slightly awkward. To handle this neatly we give a theorem that is sufficiently flexible to accommodate all cases of interest here. Theorem 4(Hierarchical block-encoding for a two-kernel rec...

  21. [21]

    Applications of the hierarchical block-encoding theorem We now discuss some applications of the above theorem. We first unpack the scaling behaviour as a function of decay profiles, then specialize to a natural convolutional case that arises for quadratic nonlinear interactions, and then recover previous square kernel results. a. Scaling as a function of ...

  22. [22]

    We explained how this can be obtained from a block-encodingU K of the real-space dataKdefined earlier, through controlled discrete Fourier transforms

    Block-encoding the Lyapunov-transformed nonlinear term Recall that we wish to construct a block-encodingU ˜F2 of the Lyapunov-transformed nonlinear term. We explained how this can be obtained from a block-encodingU K of the real-space dataKdefined earlier, through controlled discrete Fourier transforms. Moreover, ifU K has scale factorα K thenU ˜F2 has sc...

  23. [23]

    Then, d dt M(t) =−νM(t).(G1) Consequently, M(t) = e −νtM(0).(G2) In particular, ifM(0) = 0, thenM(t) = 0for allt∈[0, T]

    Conservation of mass Proposition 17(Preservation of zero perturbation mass).Lethbe a sufficiently regular solution of(50)onT 3 L × R3 ×[0, T], with periodic boundary conditions inxand sufficient decay inv. Then, d dt M(t) =−νM(t).(G1) Consequently, M(t) = e −νtM(0).(G2) In particular, ifM(0) = 0, thenM(t) = 0for allt∈[0, T]. Proof.From (51) and (50), we c...

  24. [24]

    Quadratic Lyapunov functional We next identify the Lyapunov matrixPfrom the quadratic part of the truncated plasma free energy. Proposition 18(Lyapunov matrix as the truncated quadratic free energy).LetF (2) denote the quadratic part of the continuum free energy in Eq.(24), written in the dimensionless variables of Eq.(60)and withϕconstrained by Eq.(67). ...

  25. [25]

    Proof of Proposition 6 By Proposition 18, 1 2 ∥u0∥2 2,P = 1 2 Z T3 L Z R3 |g(x,v,0)| 2w(v) d3vd 3x+ 1 2 Z T3 L |∇xϕ(x,0)| 2 +τ|ϕ(x,0)| 2 d3x ≥ τ 2 Z T3 L |ϕ(x,0)| 2 d3x= 1 2τ Z T3 L |φ(x,0)| 2 d3x.(H1) 61 Hence, ∥u0∥2 2,P ≥ 1 τ Z T3 L |φ(x,0)| 2 d3x.(H2) By the Cauchy–Schwarz inequality, φ2 avg = 1 L6 Z T3 L |φ(x,0)|d 3x !2 ≤ 1 L3 Z T3 L |φ(x,0)| 2 d3x.(H...

  26. [26]

    Moreover, the initial data (8) satisfies (67) att= 0

    Proof of Proposition 7 Observe that the initial datum (8) verifies (129) as cos(ξ n0 ·x) = 1 is attained atx=0∈T 3 L. Moreover, the initial data (8) satisfies (67) att= 0. Hence the initial data (8) is consistent. We compute the Fourier–Hermite coefficients ofg(x,v,0) as gn,α(0) = φmaxκn0 L3/2 2τ δn,n0 +δ n,−n0 δα,0.(H6) Therefore, ∥u(0)∥2 2,P = X n∈Z 3 N...

  27. [27]

    Proof of Proposition 8 We first show that the initial data (134)–(135) are consistent. Writing g(x,v,0) = C L3 X n∈Z3\{0} exp − σ2 2 |ξn|2 exp(iξn ·x) (H9) and applying the Helmholtz operator to (135), we obtain (−∇2 x +τ)ϕ(x,0) = C L3 X n∈Z3\{0} exp − σ2 2 |ξn|2 exp(iξn ·x).(H10) 62 Comparing (H10) with (H9) gives the desired result. We next show that φm...

  28. [28]

    Since the partial sums (H38) are monotone increasing inN F andN H, we also obtain sup (NF,NH)∈N2 ∥u0∥2 2 = X n∈Z3 α∈N3 0 |gn,α(0)|2.(H39) Combining (H36) and (H39) gives (H35)

    Norm of the initial data Proposition 19(Weighted square integrability implies∥u 0∥2 =O(1)).For each pair(N F, NH)∈N 2, let u0 := gn,α(0) n∈Z 3 NF α∈N 3 NH .(H32) Assume that g(·,·,0)∈L 2(T3 L ×R 3, w(v) d3vd 3x).(H33) Then, sup (NF,NH)∈N2 ∥u0∥2 <∞.(H34) and, moreover, sup (NF,NH)∈N2 ∥u0∥2 2 = X n∈Z3 α∈N3 0 |gn,α(0)|2 = Z T3 L Z R3 |g(x,v,0)| 2w(v) d3vd 3x...

  29. [29]

    For any subset Ω⊆Z 3×N3 0, define the truncation operatorT Ω through its Fourier-Hermite coefficients, by (TΩg)n,α := ( gn,α (n,α)∈Ω, 0 (n,α)/∈Ω

    Sobolev regularity Letg(x,v, t) be a solution of (4), and letg n,α(t) denote its Fourier-Hermite coefficients. For any subset Ω⊆Z 3×N3 0, define the truncation operatorT Ω through its Fourier-Hermite coefficients, by (TΩg)n,α := ( gn,α (n,α)∈Ω, 0 (n,α)/∈Ω. (H40) For the chosen Fourier and Hermite cutoffsN F, NH, the corresponding Fourier-Hermite truncatio...

  30. [30]

    Averaged kinetic energy moment Let us now see how to extract the spacetime averaged kinetic energy moment of the perturbation. The spatial averaging selects only the zero Fourier mode: 1 L3 Z T3 L g(x,v, t) d3x= 1 L3 Z T3 L 1 L3/2 X n∈Z3 ˆgn(v, t)ein·x d3x = 1 L3/2 ˆg0(v, t).(J13) Therefore ⟨K⟩= 1 T L3/2 Z T 0 dt Z R3 |v|2 2 w(v)ˆg0(v, t) d3v.(J14) We now...

  31. [31]

    We must therefore construct a state preparation routine for the normalized state| ¯ℓK,k⟩

    State preparation of| ¯ℓK,k⟩and state overlap estimate We estimate the time averaged kinetic energy by overlapping the history state with a normalized state vector| Kk⟩ that involves a factor| ¯ℓK,k⟩= ¯ℓK,k/∥¯ℓK,k∥. We must therefore construct a state preparation routine for the normalized state| ¯ℓK,k⟩. Recall, that the vector ¯ℓK,k is obtained by acting...