Quantum Koopman Algorithms
Pith reviewed 2026-05-20 10:23 UTC · model grok-4.3
The pith
Quantum Koopman Algorithms encode observable evolution to simulate free-fermion baths and nonlinear classical dynamics with polylogarithmic quantum gate costs.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Quantum Koopman Algorithms rest on approximately closed sets of observables together with efficient coherent encodings of their evolution under the Koopman operator; the framework splits into Dynamic-QKA for initial-value problems and Spectral-QKA for eigenvalue extraction, delivering O(polylog(N)) algorithms for free-fermion baths and new perturbative and spectral tools for nonlinear classical flows.
What carries the argument
Approximately closed sets of observables whose Koopman-driven evolution is coherently encoded on a quantum computer, enabling both dynamic simulation and spectral analysis.
If this is right
- Simulations of N-particle open fermionic systems become feasible on quantum hardware whose resources grow only logarithmically with N.
- Heat currents and relaxation rates can be read out directly from the encoded observable trajectories.
- Perturbative expansions for nonlinear classical dynamics can be centered on exactly solvable reference flows rather than only linear ones.
- Late-time frequencies of nonlinear oscillators can be extracted by a windowed quantum ODE solver without full long-time integration.
Where Pith is reading between the lines
- The same observable encoding may apply to other open many-body systems whose relevant operators close under commutation with the Liouvillian.
- Combining the spectral-QKA with variational methods could yield hybrid algorithms for finding steady states of driven nonlinear systems.
- The polylog scaling suggests a route to quantum advantage in classical fluid or plasma simulations that admit low-dimensional observable closures.
Load-bearing premise
The selected observables remain approximately closed under the dynamics and admit an efficient coherent quantum encoding.
What would settle it
An explicit scaling test on free fermions where the number of observables needed for fixed accuracy grows exponentially with N, forcing gate cost to exceed polylog(N).
Figures
read the original abstract
We define an observable-space framework of Quantum Koopman Algorithms (QKAs) for simulating the dynamics of both linear quantum and nonlinear classical systems, based on approximately closed sets of observables and efficient coherent encodings of their Koopman-driven evolution. QKAs have two strands: Dynamic-QKA for the initial-value problem of observables dynamics, and Spectral-QKA for the eigenvalue analysis of the Koopman operator. We demonstrate the scope of the framework through several applications. First, for classes of $N$ free fermions linearly coupled to a bath, we construct quantum algorithms with gate cost $O(\mathrm{polylog}(N))$, an exponential improvement over classical methods, and use them to reconstruct heat flows and decay rates. Second, for nonlinear classical dynamics, we introduce a novel nonlinear interaction-picture quantum algorithm that enables perturbative expansions around solvable nonlinear reference flows, going beyond existing approaches that only apply to weakly nonlinear systems. Third, we develop spectral methods for extracting eigen-frequencies of late-time nonlinear dynamics, introducing a windowed quantum ODE-solver. Our results identify the Koopman-quantum interface as a natural setting in which quantum algorithms can exploit observable-space structure to simulate both classical and quantum dynamics.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper defines an observable-space framework of Quantum Koopman Algorithms (QKAs) for simulating dynamics of both linear quantum and nonlinear classical systems, based on approximately closed sets of observables and efficient coherent encodings of their Koopman-driven evolution. It distinguishes Dynamic-QKA for the initial-value problem of observable dynamics and Spectral-QKA for eigenvalue analysis of the Koopman operator. Applications include O(polylog(N)) gate-cost quantum algorithms for N free fermions linearly coupled to a bath to reconstruct heat flows and decay rates (exponential improvement over classical methods), a nonlinear interaction-picture quantum algorithm enabling perturbative expansions around solvable nonlinear reference flows, and spectral methods with a windowed quantum ODE-solver for extracting eigen-frequencies of late-time nonlinear dynamics.
Significance. If the central constructions hold, the work is significant for identifying the Koopman-quantum interface as a setting where quantum algorithms can exploit observable-space structure to achieve exponential speedups in open quantum systems and to extend beyond weakly nonlinear regimes in classical dynamics. The paper introduces novel algorithmic strands and claims of efficient coherent encodings, which, if substantiated with explicit derivations and error bounds, would strengthen the case for observable-space approaches in quantum simulation.
major comments (2)
- [Abstract and § on free-fermion QKA construction] Abstract and fermion application: The headline claim of O(polylog(N)) gate cost for classes of N free fermions linearly coupled to a bath requires that a polynomially sized set of observables remains approximately closed under the linear bath coupling and that the resulting Koopman generator admits an efficient coherent encoding. The manuscript invokes an implicit truncation of the observable hierarchy but does not provide an explicit N-dependent error bound or show that the truncation cost remains polylog(N) for general bath spectral densities; this is load-bearing for the asserted exponential separation from classical methods.
- [Nonlinear dynamics application] Nonlinear classical dynamics section: The novel nonlinear interaction-picture quantum algorithm is presented as enabling perturbative expansions around solvable nonlinear reference flows and going beyond existing weakly nonlinear approaches, yet the manuscript does not include a concrete comparison of perturbative orders, convergence radii, or gate-cost scaling relative to prior interaction-picture methods; without this, the scope of the improvement remains unclear.
minor comments (2)
- [Introduction/Definition of QKAs] The opening definition of QKAs would benefit from an explicit diagram or pseudocode distinguishing the Dynamic-QKA and Spectral-QKA strands and their shared reliance on approximately closed observable sets.
- [Coherent encoding paragraphs] Ensure that all claims of 'efficient coherent encodings' are accompanied by at least a high-level circuit construction or reference to a standard quantum linear-systems subroutine.
Simulated Author's Rebuttal
We thank the referee for their careful reading of the manuscript and for identifying points that require clarification. We address each major comment below and indicate the changes we will make in the revised version.
read point-by-point responses
-
Referee: [Abstract and § on free-fermion QKA construction] Abstract and fermion application: The headline claim of O(polylog(N)) gate cost for classes of N free fermions linearly coupled to a bath requires that a polynomially sized set of observables remains approximately closed under the linear bath coupling and that the resulting Koopman generator admits an efficient coherent encoding. The manuscript invokes an implicit truncation of the observable hierarchy but does not provide an explicit N-dependent error bound or show that the truncation cost remains polylog(N) for general bath spectral densities; this is load-bearing for the asserted exponential separation from classical methods.
Authors: We agree that an explicit error bound is necessary to fully substantiate the polylog(N) gate-cost claim and the exponential separation. In the revised manuscript we will add a dedicated subsection deriving an N-dependent truncation error bound for the observable hierarchy under linear bath coupling. For the specific classes of free-fermion models considered (those whose bath spectral densities admit finite moments or are sufficiently smooth), we will show that the truncation error decays as O(1/poly(N)) while the coherent encoding of the truncated Koopman generator remains efficient, preserving the overall polylog(N) scaling. We note that the original construction already selects an approximately closed observable set whose size is polynomial in N for these models; the revision will make the error analysis fully rigorous rather than implicit. revision: yes
-
Referee: [Nonlinear dynamics application] Nonlinear classical dynamics section: The novel nonlinear interaction-picture quantum algorithm is presented as enabling perturbative expansions around solvable nonlinear reference flows and going beyond existing weakly nonlinear approaches, yet the manuscript does not include a concrete comparison of perturbative orders, convergence radii, or gate-cost scaling relative to prior interaction-picture methods; without this, the scope of the improvement remains unclear.
Authors: We accept that a side-by-side comparison would strengthen the presentation. In the revision we will insert a new paragraph (or short subsection) that explicitly compares (i) the perturbative orders reachable around a solvable nonlinear reference flow versus standard weakly nonlinear interaction pictures, (ii) the associated convergence radii for representative nonlinear systems, and (iii) the resulting gate-cost scalings. This comparison will use concrete model examples to illustrate the extension beyond the weakly nonlinear regime while retaining the same asymptotic complexity class. revision: yes
Circularity Check
No significant circularity; derivations rest on independent framework definitions and algorithmic constructions
full rationale
The paper defines the QKA framework from first principles using approximately closed observable sets and coherent encodings of Koopman evolution, then constructs explicit algorithms for free-fermion bath models and nonlinear classical dynamics. No load-bearing step reduces by construction to a fitted parameter, self-citation chain, or renamed input; the polylog(N) claims are presented as outputs of the new encoding methods rather than tautological re-expressions of the inputs. The derivation chain is therefore self-contained against the stated assumptions.
Axiom & Free-Parameter Ledger
axioms (1)
- standard math Quantum mechanics and the Koopman operator theory for linearizing dynamics in observable space
invented entities (1)
-
Quantum Koopman Algorithms (QKAs)
no independent evidence
Lean theorems connected to this paper
-
IndisputableMonolith/Cost/FunctionalEquationwashburn_uniqueness_aczel unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
We define an observable-space framework of Quantum Koopman Algorithms (QKAs) ... based on approximately closed sets of observables and efficient coherent encodings of their Koopman-driven evolution.
-
IndisputableMonolith/Foundation/RealityFromDistinctionreality_from_one_distinction unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
For classes of N free fermions linearly coupled to a bath, we construct quantum algorithms with gate cost O(polylog(N))
What do these tags mean?
- matches
- The paper's claim is directly supported by a theorem in the formal canon.
- supports
- The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
- extends
- The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
- uses
- The paper appears to rely on the theorem as machinery.
- contradicts
- The paper's claim conflicts with a theorem or certificate in the canon.
- unclear
- Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.
Reference graph
Works this paper leans on
-
[1]
D. W. Berry, J. Phys. A47, 105301 (2014)
work page 2014
-
[2]
D. W. Berry, A. M. Childs, A. Ostrander, and G. Wang, Commun. Math. Phys.356, 1057 (2017)
work page 2017
-
[3]
D. W. Berry and P. C. Costa, Quantum8, 1369 (2024)
work page 2024
- [4]
-
[5]
D. Jennings, M. Lostaglio, R. B. Lowrie, S. Pallister, and A. T. Sornborger, Quantum8, 1553 (2024)
work page 2024
-
[6]
D. An, A. M. Childs, and L. Lin, Commun. Math. Phys. 407, 19 (2026)
work page 2026
-
[7]
Sign Embedding Quantum Algorithms for Matrix Equations and Matrix Functions
Y. Wang and J.-P. Liu, (2026), arXiv:2604.25333 [quant- ph]
work page internal anchor Pith review Pith/arXiv arXiv 2026
-
[8]
B. O. Koopman, PNAS17, 315 (1931)
work page 1931
- [9]
- [10]
- [11]
-
[12]
J. F. Dur´ an-Siguenza, L. I. Minchala, L. E. Garza- Casta˜ n´ on, and H. Zhang, J. Franklin Inst. , 108256 (2025)
work page 2025
- [13]
-
[14]
R. D. Somma, R. King, R. Kothari, T. E. O’Brien, and R. Babbush, Nat. Commun.16, 2690 (2025)
work page 2025
-
[15]
M. Stroeks, D. Lenterman, B. M. Terhal, and Y. Herasy- menko, Phys. Rev. Res.7, 043176 (2025)
work page 2025
-
[16]
Here we treatx∈C d as a real vector inR 2d
-
[17]
For a matrixA, this is an efficient quantum circuitU A encodingA/αin its top-left block, withα >0 being a scale factor [59]
-
[18]
Here a history-state encodes dynamical data over a time interval. It takes the form P k ak|ψ(tk)⟩|tk⟩, where the second register encodes a discrete set of timest k over the interval, the first register encodes the solution at that timet k, anda k are associated norm-weightings due to potentially non-unitary dynamics
- [19]
-
[20]
D. W. Berry, G. Ahokas, R. Cleve, and B. C. Sanders, Commun. Math. Phys.270, 359 (2007)
work page 2007
-
[21]
D. W. Berry, A. M. Childs, R. Cleve, R. Kothari, and R. D. Somma, Phys. Rev. Lett.114, 090502 (2015)
work page 2015
-
[22]
A. Y. Kitaev, (1995), arXiv:quant-ph/9511026
work page internal anchor Pith review Pith/arXiv arXiv 1995
- [23]
-
[24]
A. Gily´ en, Y. Su, G. H. Low, and N. Wiebe, inPro- ceedings of the 51st annual ACM SIGACT symposium on theory of computing(2019) pp. 193–204
work page 2019
-
[25]
G. H. Low and I. L. Chuang, Quantum3, 163 (2019)
work page 2019
-
[26]
C. W. Groth, M. Wimmer, A. R. Akhmerov, and X. Waintal, New J. Phys.16, 063065 (2014)
work page 2014
- [27]
-
[28]
L. G. Valiant, inProceedings of the Thirty-Third Annual ACM Symposium on Theory of Computing(Association for Computing Machinery, 2001) p. 114–123
work page 2001
-
[29]
B. M. Terhal and D. P. DiVincenzo, Phys. Rev. A65, 032325 (2002)
work page 2002
- [30]
-
[31]
M. van Caspel, S. E. Tapias Arze, and I. P´ erez Castillo, SciPost Phys.6, 026 (2019)
work page 2019
- [32]
- [33]
-
[34]
Namely, a finite fraction of fermionic modes remains a constant away from the infinite temperature state throughout the evolution
-
[35]
H.-P. Breuer and F. Petruccione,The theory of open quantum systems(OUP Oxford, 2002)
work page 2002
-
[36]
C. Fleming, N. Cummings, C. Anastopoulos, and B.-L. Hu, J. Phys. A: Math. Theor.43, 405304 (2010)
work page 2010
- [37]
-
[38]
Explicit Error Bounds for Carleman Linearization
M. Forets and A. Pouly, (2017), arXiv:1711.02552 [math- NA]
work page internal anchor Pith review Pith/arXiv arXiv 2017
- [39]
-
[40]
J.-P. Liu, H. Ø. Kolden, H. K. Krovi, N. F. Loureiro, K. Trivisa, and A. M. Childs, PNAS118, e2026805118 (2021)
work page 2021
-
[41]
D. Jennings, K. Korzekwa, M. Lostaglio, A. T. Sornborger, Y. Subasi, and G. Wang, (2025), arXiv:2509.07155 [quant-ph]
-
[42]
D. Jennings, K. Korzekwa, M. Lostaglio, R. Ashworth, E. Marsili, and S. Rolston, (2025), arXiv:2512.03758 [quant-ph]. 6
- [43]
- [44]
-
[45]
Note that any polynomial nonlinearity can be reduced to a quadratic one
- [46]
- [47]
- [48]
- [49]
- [50]
-
[51]
More precisely, ifη 1(x(t)) andη 2(x(t)) are eigenmodes of ˜KA with eigenvaluesλ 1 andλ 2 respectively, then η1(x(t))η2(x(t)) is an eigenmode of ˜KA with eigenvalue λ1 +λ 2
-
[52]
With the usual caveat that theR-number is a sufficient but not necessary criterion of convergence
-
[53]
D. W. Berry, Y. Su, C. Gyurik, R. King, J. Basso, A. D. T. Barba, A. Rajput, N. Wiebe, V. Dunjko, and R. Babbush, PRX Quantum5, 010319 (2024)
work page 2024
-
[54]
S. Greenaway, W. Pol, and S. Sim, (2024), arXiv:2404.01396 [quant-ph]
- [55]
- [56]
-
[57]
M. O. Williams, I. G. Kevrekidis, and C. W. Rowley, J. Nonlinear Sci.25, 1307 (2015)
work page 2015
-
[58]
S. Simon, D. W. Berry, and R. D. Somma, (2026), arXiv:2605.16195 [quant-ph]
work page internal anchor Pith review Pith/arXiv arXiv 2026
-
[59]
Lin, (2022), arXiv:2201.08309 [quant-ph]
L. Lin, (2022), arXiv:2201.08309 [quant-ph]
- [60]
- [61]
-
[62]
A. M. Dalzell, (2024), arXiv:2406.12086 [quant-ph]
work page internal anchor Pith review Pith/arXiv arXiv 2024
- [63]
- [64]
-
[65]
K. Casas-Garc´ ıa, L. Quezada-T´ ellez, S. Carrillo-Moreno, J. Flores-Godoy, and G. Fern´ andez-Anaya, Nova Scientia 8, 41 (2016)
work page 2016
- [66]
-
[67]
L. K. Grover, Phys. Rev. Lett.95, 150501 (2005)
work page 2005
-
[68]
T. J. Yoder, G. H. Low, and I. L. Chuang, Phys. Rev. Lett.113, 210501 (2014)
work page 2014
-
[69]
J. M. Martyn, Z. M. Rossi, A. K. Tan, and I. L. Chuang, PRX Quantum2, 040203 (2021). 7 Supplemental Material S1. RELA TION TO PRIOR WORKS A schematic construction of a QKA is shown in Fig. S1. One first specifies a dynamical system, i.e., a vector fieldF in (1) (A1), and then selects a family of observables{g m}(A2), which induces the Koopman evolution (2...
work page 2021
-
[70]
[5, 60], the complexityQfor outputting a stateϵ-close to Eq
Query complexity cost Using the results in Ref. [5, 60], the complexityQfor outputting a stateϵ-close to Eq. (6) is Q=O αmax τ∈[0,t] ∥eBτ ∥g(t)tlog max{t,∥vec(Y)∥} ϵΓmin(t) log g(t)t ϵ ,(S29) whereαis the block-encoding prefactor forBand g(t) := max τ∈[0,t] ∥Γ(τ)∥ F ∥Γ(t)∥F ,(S30a) Γmin(t) := min τ∈[0,t] ∥Γ(τ)∥ F .(S30b) The complexityQmeasures the number...
-
[71]
Efficient block-encoding ofB In order to efficiently block-encodeB, it is sufficient to block-encodehandX, which can then be combined using an LCU. It is known that a sparsehcan be efficiently block-encoded under the assumptions of Result 1, so we only need to block-encodeX: X= 2 X µ Re(l† µlµ).(S37) 12 To achieve this, we will construct a block-encodingU...
-
[72]
Efficient state preparation ofvec(Y) Our aim now is to construct an efficient state preparation unitaryU |vec(Y)⟩ of the normalized version|vec(Y)⟩of the vector vec(Y) =−4 X µ Im(l∗ µ ⊗l µ).(S49) If there are only Θ(1) nonzero elements of Iml µi, then|vec(Y)⟩can trivially be prepared efficiently, and the norm ∥vec(Y)∥will be Θ(1). Otherwise, when there ar...
-
[73]
Prepare the window state together with two registers initialized to the all-zero state, with one serving as the system register and the other as an ancilla: J−1X j=0 βj|j⟩|0⟩|0⟩. 30
-
[74]
The implementation details of this operation are given in Appendix S7 I
Apply the controlled ODE-solver unitary |j⟩|0⟩|0⟩ 7→ |j⟩W(T 1 +j∆t)|0⟩|0⟩. The implementation details of this operation are given in Appendix S7 I. This prepares a state close to J−1X j=0 βj|j⟩|0⟩|¯g(T1 +j∆t)⟩= J−1X j=0 βj ∥g(T1 +j∆t)∥ |j⟩|0⟩|g(T1 +j∆t)⟩(S197) ≈ 1 ∥aS∥ J−1X j=0 βj|j⟩|0⟩|g(T1 +j∆t)⟩.(S198)
-
[75]
Apply the inverse quantum Fourier transform to the first register
-
[76]
Measure the first register. If the outcome isℓ∈ {0,1, . . . , J−1}, output ˆωℓ := 2πℓ J∆t ,0≤ℓ≤ J−1 2 , 2πℓ J∆t − 2π ∆t , J+1 2 ≤ℓ≤J−1. (S199) The decoding rule above is the frequency version of the signed phase decoding from the previous subsection. The Nyquist-margin assumption becomes −π+c≤ω k∆t≤π−c(S200) for every relevant frequencyω k. Under ...
-
[77]
If this step fails, the algorithm terminates
Use quantum singular value transformation (QSVT) [24] to approximately implement a reflection about the kernel ofG r, and apply this operation to|e D⟩. If this step fails, the algorithm terminates
-
[78]
If the outcome corresponds toI− |eD⟩⟨eD|, output the final state; otherwise, the algorithm fails
Perform the measurement{I− |e D⟩⟨eD|,|e D⟩⟨eD|}on the resulting state. If the outcome corresponds toI− |eD⟩⟨eD|, output the final state; otherwise, the algorithm fails. To achieve accuracyϵ 2 in the output, the QSVT in the second step requires ˜O(κ A log (1/ϵ2)) controlled queries to UA,U b, and their inverses. Ref. [62] shows that the above procedure suc...
-
[79]
In the following, we ignore the small ˜O(l) difference between these costs. Recall that |y⟩ ≈ m−1X s=0 |s⟩|g(sh)⟩+ m+p−1X s=m |s⟩|g(t)⟩.(S257) Hence, Ur|0⟩1|0⟩2|0⟩3 ≈ crqPm−1 s=0 ∥g(sh)∥2 +p∥g(t)∥ 2 |0⟩1 m−1X s=0 |s⟩2|g(sh)⟩3 + m+p−1X s=m |s⟩2|g(t)⟩3 ! + X i̸=0 |i⟩1|eξr,i⟩2,3.(S258) Letd= (p−m)/2 and define Π =|0⟩⟨0| ⊗P [m+p]\[m+d]. Then (Π1,2 ⊗I 3)(Ur)1,...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.