Pith. sign in

REVIEW 3 major objections 6 minor 59 references

A spectral correction built from the Fokker–Planck generator cancels non-adiabatic lag at any driving speed.

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 · grok-4.5

2026-07-31 16:00 UTC pith:6TB6M2X2

load-bearing objection Solid 1D spectral method for density-level counterdiabatic FP escorting; math and numerics check out, but the free-energy payoff that frames the paper is never demonstrated. the 3 major comments →

arxiv 2607.24393 v1 pith:6TB6M2X2 submitted 2026-07-27 physics.comp-ph cond-mat.stat-mechcs.LG

Stochastic Counterdiabatic Driving via Biorthogonal Liouvillian Eigenmodes

Sandeep Suresh Cranganore , Sebastian Lehner , Johannes Brandstetter , Max Welling This is my paper
classification physics.comp-ph cond-mat.stat-mechcs.LG
keywords counterdiabatic drivingFokker-Planck equationbiorthogonal eigenmodesshortcuts to adiabaticityescorted free energynon-adiabatic lagLiouvillianstochastic thermodynamics
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.

Finite-time driving of a diffusing particle makes its probability density lag the moving equilibrium, which creates excess heat and ruins free-energy estimates that rely on rare lucky trajectories. This paper shows how to cancel that lag exactly by reading a counterdiabatic correction straight from the biorthogonal eigenmodes of the time-dependent Fokker–Planck generator, without inventing new coordinates or training a map. Under ordinary detailed balance the correction collapses to a simple rank-one operator fixed by the instantaneous Boltzmann weight and its time derivative. On a driven double well and a driven harmonic trap the corrected density tracks equilibrium to machine precision, cutting total-variation lag by about twelve orders of magnitude and dissipated work to numerical noise at every protocol speed. The same spectral sum can be truncated to the slowest modes, giving a controllable trade-off between cost and residual lag.

Core claim

The biorthogonal spectral formula for the counterdiabatic Liouvillian, which under detailed balance equals the rank-one operator (∂π_eq/∂t)1^T, enforces the exact tracking condition ∂π_eq/∂t = L_CD π_eq. Consequently the deterministically propagated Fokker–Planck density follows the instantaneous equilibrium at arbitrary protocol speed, with total-variation distance ~10^{-12}, KL divergence ~10^{-16}, and dissipated work at the quadrature floor on the double-well and harmonic examples.

What carries the argument

Liouvillian counterdiabatic driving (LCD): the operator assembled from the biorthogonal eigenpairs of the discrete Fokker–Planck generator, L_CD(t) = −Σ_{n≠0} [ℓ_n^T (∂L/∂t) r_0 / λ_n] r_n ℓ_0^T, which under detailed balance collapses to (∂π_eq/∂t)1^T and cancels the adiabatic gauge connection that pumps probability out of the zero mode.

Load-bearing premise

That perfect tracking and vanishing dissipated work for the deterministically evolved discrete density are enough to underwrite the escorted free-energy program the paper sets out to serve; no stochastic trajectories or Jarzynski averages are computed.

What would settle it

Propagate the discrete Fokker–Planck density under L + L_CD on the double-well or harmonic protocol and check whether max_t TVD(ρ(t), π_eq(t)) and |W_diss(t)| stay at the reported machine/quadrature floors (~10^{-12}); any systematic rise with protocol speed would refute the claim.

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

If this is right

  • Escorted free-energy protocols can be built by adding a spectral (or closed-form rank-one) correction to the Fokker–Planck generator instead of learning coordinate maps.
  • Spectral truncation to the slowest M modes yields imperfect but systematically improvable escorting whose error scales with driving speed and inverse gap.
  • When the spectral gap closes the spectral assembly becomes ill-conditioned while the closed-form rank-one operator remains bounded, separating representation failure from physical control failure.
  • The same biorthogonal construction is proposed as a route to systems without detailed balance once the steady state is obtained numerically.

Where Pith is reading between the lines

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

  • Mapping the dense L_CD operator back to a physically realizable drift field u(x,t) is the missing link needed before trajectory-based zero-variance Jarzynski estimators can use this method.
  • Sparse Lanczos or Krylov constructions suggested in the limitations could make the approach practical in two and three dimensions where full diagonalization is impossible.
  • Complementarity with flow-based methods is sharp: spectral LCD fails when gaps vanish, while configuration-space flows are gap-blind but may be harder when the target density is not analytic.

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 / 6 minor

Summary. The manuscript constructs a counterdiabatic correction for the discretized Fokker–Planck generator from its biorthogonal eigendecomposition. Differentiating the stationarity condition L̂π_eq=0 yields a spectral (pseudo-inverse) formula, Eq. (34), for an operator L̂_CD satisfying L̂_CD π_eq = ∂π_eq/∂t; under detailed balance every term shares the left factor ℓ_0^T=1^T, so the sum collapses to the rank-one closed form L̂_CD=(∂π_eq/∂t)1^T, Eq. (35), computable at O(N) from the Boltzmann weight alone. Propagating ρ with L̂+L̂_CD via a fourth-order Magnus integrator on Sasa–Tasaki grids for a driven double well and harmonic trap, the authors verify zero-mode/equilibrium alignment to ~1e-15, TVD ~1e-12, KL ~1e-16, and dissipated work at the quadrature floor across protocol speeds spanning four decades. A truncated-M-mode construction gives systematically improvable "imperfect escorting" (Tables S6–S8), and a quartic-coalescence stress test shows the spectral assembly (not the closed form) failing when the gap drops below machine precision.

Significance. If the claims hold, the paper delivers a parameter-free, optimization-free construction with exact algebraic verification: Eq. (35) is derived, not fitted, and the tracking numbers (TVD ~1e-12, KL at machine floor) are consistent with the stated Magnus+Simpson O(Δt^4) accuracy. I credit in particular (i) the honest identification that under detailed balance the spectral operator is rank one with a closed form, avoiding an inflated novelty claim in the body; (ii) the truncation study quantifying mode count vs. protocol speed; (iii) the coalescence stress test, which correctly distinguishes failure of the spectral representation from failure of the control (parity-protected numerator, ill-conditioned eigenbasis); and (iv) a concrete reproducibility statement. Significance is nonetheless bounded: for equilibrium problems the exact correction coincides with the known continuum result (Guéry-Odelin et al., Ref. [40]), and the motivating escorted-Jarzynski payoff — a realizable flow field u(x,t) and trajectory-level zero-variance estimator — is explicitly not delivered, since L̂_eff is not a stochastic generator. The durable contributions are the spectral-gap diagnostic, the truncation/f

major comments (3)
  1. [§VII C heading and Abstract] The heading "Trajectory-Resolved Dissipated Work" is inaccurate: no trajectories are sampled anywhere in the paper. W(t) in Eq. (39) is an ensemble moment of the deterministically propagated density, and W_diss≡0 follows algebraically from the enforced identity ρ(t)=π_eq(t) via ⟨∂V/∂t⟩_{π_eq}=dF/dt — as the authors themselves state (§VII C, second sentence). Table II therefore re-measures the same fact as Table I's TVD rather than independently confirming it, and the abstract's "we demonstrate vanishing dissipated work" should be framed as a consistency check of the quadrature, not an independent result. Please retitle §VII C (e.g., "Ensemble dissipated work along the protocol") and add one clause to the abstract noting the equivalence. Relatedly, §VII C asserts a "transient energetic cost" of the CD drive that "scales aggressively" with the gap (citing [43]) but never computes it; eithe
  2. [§V A and Abstract / Introduction framing] Because L̂_CD=(∂π_eq/∂t)1^T is dense with sign-indefinite off-diagonals (stated correctly in §V A), L̂_eff is not a rate matrix: no Markov jump process or Langevin ensemble realizes the escorted dynamics, so the work functional W_u of Eq. (3) and the zero-variance EJE estimator that motivate the entire Introduction (Eqs. 1–3) are never touched by the evidence. The Introduction's final parenthetical concedes this, but a reader of the abstract alone would reasonably conclude the method yields a physical escorting field. The abstract should state explicitly that L̂_CD is a numerical gauge operator on the density, not a realizable control field, and that mapping it to a flow field u(x,t) (per Vaikuntanathan–Jarzynski) is open. This is a correctness-risk framing issue, not a flaw in the derivation.
  3. [Abstract and §V B] The abstract leads with "exact spectral decomposition of the time-dependent Fokker–Planck generator" as the method, but §V B proves that under detailed balance — the only regime treated — the spectral sum collapses to a rank-one operator fixed by the Boltzmann weight at O(N) cost, with no eigendecomposition, and that this is the matrix form of a known continuum result [40]. The genuine novelties are the truncation scheme, the gap diagnostic, and the NESS path. The abstract and the bullet list in §I should be rebalanced so the rank-one collapse and the equivalence to [40] are visible up front; as written, the primary contribution is easy to over-read. This matters for the significance assessment more than for correctness.
minor comments (6)
  1. [§VII D] Typos in heading ("T runcatation") and elsewhere: "counterdibatic" (§VII C), garbled sentence "Thus, for despite much finer discretizations..." (§II A), Fig. 1(b) axis label renders as "0(t)" instead of κ_0(t).
  2. [References] Duplicate entries: [11] and [32] are the same Berry 2009 paper; [35] and [42] are the same Sasa–Tasaki work. Please consolidate.
  3. [§VII D / Limitations] The claim that truncation "circumvents the computationally prohibitive requirement of full matrix diagonalization" is only true in conjunction with the sparse Lanczos strategy of §VII F — computing the M slowest biorthogonal modes still requires an eigensolver pass at each time step. Please state the assumed per-step cost of the truncated construction explicitly.
  4. [§VII E / §VII C (Supplement)] In the symmetric coalescence case the slow-mode numerator ℓ_1^T(∂L̂/∂ζ)r_0 vanishes by parity, so the breakdown is a 0/0 numerical artifact. It would strengthen the stress test to show one asymmetric barrier-crossing example where the numerator is nonzero and the spectral coefficients genuinely diverge, to substantiate the claim in Supplement §VII C that the classical control diverges only when probability must be transported through the closure.
  5. [Reproducibility statement] The repository link appears as a placeholder (" LCD-FPE") in the manuscript. Please verify the URL resolves before publication.
  6. [§IV, Eq. (26)] The statement that Σ ˜ρ_n ≠ 1 "in the adiabatic frame" is correct but easy to misread as a violation of normalization; a sentence noting that the individual ˜ρ_n are signed mode amplitudes in a non-probability basis (already implied by Eq. S30–S33) would help readers.

Circularity Check

1 steps flagged

Mostly non-circular construction; only mild tautology is treating W_diss≈0 as an independent diagnostic when the paper itself equates it to ρ=π_eq.

specific steps
  1. self definitional [§VII C; Eqs. (39)–(40); Table II vs Table I]
    "because ⟨∂V/∂t⟩_πeq = dF/dt at every instant [Section VI A], perfect tracking of the instantaneous equilibrium, ρ(t) = π_eq(t), is equivalent to vanishing dissipation, W_diss(t) = 0. The trajectory-resolved W_diss(t) is therefore the thermodynamic corollary of the distributional tracking established in Section VI A"

    W_diss is defined as W−ΔF with W=∫⟨∂V/∂t⟩_ρ dt. Under the LCD dynamics the paper enforces ρ=π_eq, which makes ⟨∂V/∂t⟩_ρ = dF/dt identically, hence W_diss≡0 by algebra. Reporting max|W_diss|~10^{-12} as a separate success criterion does not independently confirm escorting beyond the TVD/KL tracking already shown; it re-evaluates the same enforced identity (plus quadrature error). The paper states the equivalence, so this is mild rhetorical circularity, not a hidden fit.

full rationale

The load-bearing derivation is a standard solve-then-verify chain, not a fit or self-citation loop. The counterdiabatic condition ∂π_eq/∂t = L̂_CD π_eq is imposed as a design requirement; the spectral formula (Eq. 34) and its detailed-balance collapse to the rank-one operator (∂π_eq/∂t)1^T (Eq. 35) are algebraic consequences of biorthogonality and 1^T π_eq = 1, not parameters fitted to the reported TVD/KL/W_diss tables. Numerical Magnus propagation then checks that the discrete dynamics realize that solution to quadrature/machine floor—an independent floating-point test, not a prediction forced by a fit. Citations (Berry, Demirplak–Rice, Jarzynski/Vaikuntanathan, Iram et al., Guéry-Odelin) supply external STA/escorting context; none is a same-author uniqueness theorem that forbids alternatives or smuggles the result. The sole soft spot is rhetorical: once tracking is enforced, W_diss(t)≡0 follows from ⟨∂V/∂t⟩_πeq = dF/dt by the paper’s own identity (§VII C), so Table II largely remeasures Table I. That is a minor self-definitional overclaim of “diagnostic” independence, not a circular derivation of the method. Score 1.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 1 invented entities

The work sits on standard overdamped FP/stochastic-thermodynamics assumptions plus a specific spatial discretization and the choice to validate on deterministic density evolution. No fitted physical constants. The main invented framing is ‘Liouvillian counterdiabatic driving (LCD)’ as a named spectral procedure; the operator itself reduces to prior continuum CD under detailed balance.

free parameters (3)
  • Spatial grid size N and domain bounds = N=80
    N=80 and intervals (e.g. [-2.5,2.5] DW, [-4,4] harmonic) chosen for numerics; affect discretization error though results are reported at machine-level tracking.
  • Integration time step Δt schedule = τ-dependent, 1e-7..1e-4
    Δt scaled from 1e-7 to 1e-4 with τ to control Magnus/Simpson error; hand-chosen for uniform fidelity, not learned from data.
  • Modal truncation rank M = varied 5–35 vs full 79
    When using imperfect escorting, M is a free accuracy/cost knob (tables for M=5..35).
axioms (5)
  • domain assumption Overdamped Langevin / Fokker–Planck dynamics with Itô interpretation and reflecting boundaries on a finite interval.
    Section II; entire construction is for the FP generator, not underdamped or jump processes.
  • domain assumption Detailed balance holds so π_eq ∝ e^{-βV} is the exact right zero mode and the CD operator collapses to rank one.
    Stated throughout; NESS extension deferred to future work (Limitations).
  • domain assumption Sasa–Tasaki local rates on a uniform grid converge to the continuum FP operator and preserve the discrete stationary Boltzmann weights.
    Section II A and Supplemental II; underpins identifying r_0 with π_eq.
  • standard math Perron–Frobenius structure: unique zero eigenvalue, strictly negative real spectrum, left zero mode = 1^T.
    Section IV; used for biorthogonality and probability conservation.
  • ad hoc to paper Perfect density tracking (ρ=π_eq) is the operational definition of successful escorting for reporting W_diss≈0, without requiring sampled EJE variance.
    Introduction and §VII C explicitly distinguish deterministic FP diagnostics from trajectory free-energy estimation.
invented entities (1)
  • Liouvillian counterdiabatic driving (LCD) / spectral ˆL_CD assembly independent evidence
    purpose: Name and construct the counterdiabatic generator from biorthogonal eigenpairs of the discrete FP operator, with optional truncation.
    Framing device; under detailed balance it equals the pre-existing rank-one classical CD operator, so independent physical content is limited to the spectral representation and numerics.

pith-pipeline@v1.2.0-grok45-kimik3 · 46048 in / 3693 out tokens · 80941 ms · 2026-07-31T16:00:48.701187+00:00 · methodology

0 comments
read the original abstract

Finite-time driving of stochastic systems generates excess dissipation, causing the evolving probability distribution to lag behind the instantaneous equilibrium, and consequently degrading the convergence of nonequilibrium free energy estimators based on the Jarzynski equality. Escorted free energy simulations address the non-adiabatic lag by engineering control fields $\mathbf{u}$ that eliminate the lag, enforcing the trajectory-wise equality $\mathcal{W}_\mathbf{u} = \Delta \mathcal{F}$, and yielding zero-variance estimators. However, constructing the escorting field in closed form remains a challenge, approached variously through flow-field methods, targeted free energy perturbation, or learned diffeomorphisms. In this work, we construct a complementary numerical framework based on gauge-type transforms instead of generalized coordinate transforms for perfect escorting based on the exact spectral decomposition of the time-dependent Fokker-Planck generator. The biorthogonal decomposition of the Liouville operator directly yields a counterdiabatic correction whose action on the instantaneous equilibrium distribution exactly cancels the non-adiabatic lag at arbitrary driving speed in formal analogy with shortcuts-to-adiabaticity techniques such as Berry's transitionless driving for quantum systems. Numerical verification for simulations of an overdamped particle in a time-varying double-well potential and harmonic traps confirms that the counterdiabatic condition is satisfied to machine precision, with the non-adiabatic lag suppressed by roughly twelve orders of magnitude in total variation distance and sixteen orders in KL divergence relative to the unescorted dynamics. As a diagnostic, we demonstrate vanishing dissipated work $\mathcal{W}_{\text{diss}}(t) \approx 0$ for the deterministically propagated Fokker-Planck density across all protocol speeds.

Figures

Figures reproduced from arXiv: 2607.24393 by Johannes Brandstetter, Max Welling, Sandeep Suresh Cranganore, Sebastian Lehner.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Convergence metrics of the spectrally truncated counterdiabatic expansion as a function of the active mode [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

59 extracted references · 3 canonical work pages

  1. [1]

    While the LCD isolates the system from non-adiabatic excitations, ensuring that the dissipated work vanishes, this tracking is bought at the expense of a transient ener- getic cost

    The trajectory-resolvedW diss(t) is therefore the ther- modynamic corollary of the distributional tracking estab- lished in Section VI A; reporting it additionally confirms that the escorting equality holds for the work functional. While the LCD isolates the system from non-adiabatic excitations, ensuring that the dissipated work vanishes, this tracking i...

    2021

  2. [2]

    Albash and D

    T. Albash and D. A. Lidar, Adiabatic quantum compu- tation, Rev. Mod. Phys.90, 015002 (2018)

    2018

  3. [3]

    Jarzynski, Equalities and inequalities: irreversibility and the second law of thermodynamics at the nanoscale, Annu

    C. Jarzynski, Equalities and inequalities: irreversibility and the second law of thermodynamics at the nanoscale, Annu. Rev. Condens. Matter Phys.2, 329 (2011)

    2011

  4. [4]

    Sch¨ afer and E

    T. Sch¨ afer and E. V. Shuryak, Instantons in qcd, Rev. Mod. Phys.70, 323 (1998)

    1998

  5. [5]

    Jarzynski, Nonequilibrium equality for free energy dif- ferences, Phys

    C. Jarzynski, Nonequilibrium equality for free energy dif- ferences, Phys. Rev. Lett.78, 2690 (1997)

    1997

  6. [6]

    R. M. Neal, Annealed importance sampling (1998), arXiv:physics/9803008 [physics.comp-ph]

    Pith/arXiv arXiv 1998

  7. [7]

    G. E. Crooks, Entropy production fluctuation theorem and the nonequilibrium work relation for free energy dif- ferences, Phys. Rev. E60, 2721 (1999)

    1999

  8. [8]

    Vaikuntanathan and C

    S. Vaikuntanathan and C. Jarzynski, Escorted free en- ergy simulations: Improving convergence by reducing dis- sipation, Phys. Rev. Lett.100, 190601 (2008)

    2008

  9. [9]

    Lee and C

    S. Lee and C. Jarzynski, Estimating free energy dif- ferences with virtually escorted trajectories (2026), arXiv:2606.30451 [cond-mat.stat-mech]

    Pith/arXiv arXiv 2026

  10. [10]

    Gu´ ery-Odelin, A

    D. Gu´ ery-Odelin, A. Ruschhaupt, A. Kiely, E. Tor- rontegui, S. Mart ´ ınez-Garaot, and J. G. Muga, Short- cuts to adiabaticity: concepts, methods, and applica- tions, Rev. Mod. Phys.91, 045001 (2019)

    2019

  11. [11]

    Kolodrubetz, D

    M. Kolodrubetz, D. Sels, P. Mehta, and A. Polkovnikov, Geometry and non-adiabatic response in quantum and classical systems, Phys. Rep.697, 1 (2017)

    2017

  12. [12]

    M. V. Berry, Transitionless quantum driving, Journal of Physics A: Mathematical and Theoretical42, 365303 (2009)

    2009

  13. [13]

    Demirplak and S

    M. Demirplak and S. A. Rice, Adiabatic population transfer with control fields, J. Phys. Chem. A107, 9937 (2003)

    2003

  14. [14]

    del Campo, Shortcuts to adiabaticity by counterdia- batic driving, Phys

    A. del Campo, Shortcuts to adiabaticity by counterdia- batic driving, Phys. Rev. Lett.111, 100502 (2013)

    2013

  15. [15]

    S. Iram, E. Dolson, J. Chiel, J. Pelesko, N. Krishnan, ¨O. G¨ ung¨ or, B. Kuznets-Speck, S. Deffner, E. Ilker, J. G. Scott, and M. Hinczewski, Controlling the speed and tra- jectory of evolution with counterdiabatic driving, Nature Physics17, 135 (2021)

    2021

  16. [16]

    Vaikuntanathan and C

    S. Vaikuntanathan and C. Jarzynski, Escorted free en- ergy simulations, The Journal of Chemical Physics134, 054107 (2011)

    2011

  17. [17]

    Jarzynski, Generating shortcuts to adiabaticity in quantum and classical dynamics, Phys

    C. Jarzynski, Generating shortcuts to adiabaticity in quantum and classical dynamics, Phys. Rev. A88, 040101(R) (2013)

    2013

  18. [18]

    Patra and C

    A. Patra and C. Jarzynski, Shortcuts to adiabaticity us- ing flow fields, New J. Phys.19, 125009 (2017)

    2017

  19. [19]

    G. Li, H. T. Quan, and Z. C. Tu, Shortcuts to isother- mality and nonequilibrium work relations, Phys. Rev. E 96, 012144 (2017)

    2017

  20. [20]

    Jarzynski, Targeted free energy perturbation, Phys

    C. Jarzynski, Targeted free energy perturbation, Phys. Rev. E65, 046122 (2002)

    2002

  21. [21]

    Wirnsberger, A

    P. Wirnsberger, A. J. Ballard, G. Papamakarios, S. Aber- crombie, S. Racani` ere, A. Pritzel, D. Jimenez Rezende, and C. Blundell, Targeted free energy estimation via learned mappings, The Journal of Chemical Physics153, 144112 (2020)

    2020

  22. [22]

    M. R. Shirts and J. D. Chodera, Statistically optimal analysis of samples from multiple equilibrium states, The Journal of Chemical Physics129, 124105 (2008)

    2008

  23. [23]

    Lipman, R

    Y. Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le, Flow matching for generative modeling (2023), arXiv:2210.02747 [cs.LG]

    Pith/arXiv arXiv 2023

  24. [24]

    D. J. Rezende and S. Mohamed, Variational infer- ence with normalizing flows (2016), arXiv:1505.05770 [stat.ML]

    Pith/arXiv arXiv 2016

  25. [25]

    Nielsen, P

    D. Nielsen, P. Jaini, E. Hoogeboom, O. Winther, and M. Welling, Survae flows: Surjections to bridge the gap between vaes and flows (2020), arXiv:2007.02731 [cs.LG]

    Pith/arXiv arXiv 2020

  26. [26]

    H. Wu, J. K¨ ohler, and F. No´ e, Stochastic normaliz- ing flows, inAdvances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6- 12, 2020, virtual, edited by H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin (2020)

    2020

  27. [27]

    Holdijk, N

    L. Holdijk, N. M. Anand, M. M. Bronstein, and M. Welling, Learning escorted protocols for multistate free-energy estimation, inThe Fourteenth International Conference on Learning Representations(2026)

    2026

  28. [28]

    Vargas, S

    F. Vargas, S. Padhy, D. Blessing, and N. N¨ usken, Trans- port meets variational inference: Controlled monte carlo diffusions, inThe Twelfth International Conference on Learning Representations(2024)

    2024

  29. [29]

    M. S. Albergo and E. Vanden-Eijnden, Nets: A non- equilibrium transport sampler (2025), arXiv:2410.02711 [cs.LG]

    Pith/arXiv arXiv 2025

  30. [30]

    Holderrieth, M

    P. Holderrieth, M. S. Albergo, and T. Jaakkola, Leaps: A discrete neural sampler via locally equivariant networks, inProceedings of the 42nd International Conference on Machine Learning(2025)

    2025

  31. [31]

    Dieudonn´ e, On biorthogonal systems., Michigan Math- ematical Journal2, 7 (1953)

    J. Dieudonn´ e, On biorthogonal systems., Michigan Math- ematical Journal2, 7 (1953)

    1953

  32. [32]

    D. C. Brody, Biorthogonal quantum mechanics, Journal of Physics A: Mathematical and Theoretical47, 035305 (2013)

    2013

  33. [33]

    M. V. Berry, Transitionless quantum driving, J. Phys. A 42, 365303 (2009)

    2009

  34. [34]

    Jarzynski, Equilibrium free-energy differences from nonequilibrium measurements: A master-equation ap- proach, Phys

    C. Jarzynski, Equilibrium free-energy differences from nonequilibrium measurements: A master-equation ap- proach, Phys. Rev. E56, 5018 (1997)

    1997

  35. [35]

    Hummer and A

    G. Hummer and A. Szabo, Free energy recon- struction from nonequilibrium single-molecule pulling experiments, Proceedings of the Na- tional Academy of Sciences98, 3658 (2001), https://www.pnas.org/doi/pdf/10.1073/pnas.071034098

    work page doi:10.1073/pnas.071034098 2001

  36. [36]

    Sasa and H

    S.-I. Sasa and H. Tasaki, Steady state thermodynamics, Journal of Statistical Physics125, 125 (2006)

    2006

  37. [37]

    Ilker, O

    E. Ilker, O. m. c. G¨ ung¨ or, B. Kuznets-Speck, J. Chiel, S. Deffner, and M. Hinczewski, Shortcuts in stochastic systems and control of biophysical processes, Phys. Rev. X12, 021048 (2022). 1

    2022

  38. [38]

    A. C. Santos and M. S. Sarandy, Generalized transition- less quantum driving for open quantum systems, Phys. Rev. A104, 062421 (2021)

    2021

  39. [39]

    Vacanti, R

    G. Vacanti, R. Fazio, S. Montangero, G. M. Palma, M. Paternostro, and V. Vedral, Transitionless quantum driving in open quantum systems, New Journal of Physics 16, 053017 (2014)

    2014

  40. [40]

    Demirplak and S

    M. Demirplak and S. A. Rice, Assisted adiabatic passage revisited, The Journal of Physical Chemistry B109, 6838 (2005)

    2005

  41. [41]

    Gu´ ery-Odelin, C

    D. Gu´ ery-Odelin, C. Jarzynski, C. A. Plata, A. Prados, and E. Trizac, Driving rapidly while remaining in control: classical shortcuts from hamiltonian to stochastic dynam- ics, Reports on Progress in Physics86, 035902 (2023)

    2023

  42. [42]

    Zhong and M

    A. Zhong and M. R. DeWeese, Beyond linear response: Equivalence between thermodynamic geometry and op- timal transport, Phys. Rev. Lett.133, 057102 (2024)

    2024

  43. [43]

    ichi Sasa and H

    S. ichi Sasa and H. Tasaki, Steady state thermodynamics (2005), arXiv:cond-mat/0411052 [cond-mat.stat-mech]

    Pith/arXiv arXiv 2005

  44. [44]

    Campbell and S

    S. Campbell and S. Deffner, Trade-off between speed and cost in shortcuts to adiabaticity, Phys. Rev. Lett.118, 100601 (2017)

    2017

  45. [45]

    M. G. Evans and M. Polanyi, Some applica- tions of the transition state method to the cal- culation of reaction velocities, especially in so- lution, Transactions of the Faraday Society 31, 875 (1935), https://pubs.rsc.org/tf/article- pdf/doi/10.1039/TF9353100875/1608808/tf9353100875.pdf

    work page doi:10.1039/tf9353100875/1608808/tf9353100875.pdf 1935

  46. [46]

    B. Derrida, Non-equilibrium steady states: fluctuations and large deviations of the density and of the current, Journal of Statistical Mechanics: Theory and Experi- ment2007, P07023 (2007)

    2007

  47. [47]

    C. Lanczos, An iteration method for the solution of the eigenvalue problem of linear differential and integral op- erators, Journal of Research of the National Bureau of Standards45, 255 (1950)

    1950

  48. [48]

    Golub and C

    G. Golub and C. Van Loan,Matrix Computations, Johns Hopkins Studies in the Mathematical Sciences (Johns Hopkins University Press, 1996)

    1996

  49. [49]

    Seifert, Entropy production along a stochastic tra- jectory and an integral fluctuation theorem, Phys

    U. Seifert, Entropy production along a stochastic tra- jectory and an integral fluctuation theorem, Phys. Rev. Lett.95, 040602 (2005)

    2005

  50. [50]

    Hatano and S.-i

    T. Hatano and S.-i. Sasa, Steady-state thermodynamics of langevin systems, Phys. Rev. Lett.86, 3463 (2001)

    2001

  51. [51]

    Morawetz and A

    S. Morawetz and A. Polkovnikov, Universal counterdia- batic driving in krylov space, PRX Quantum6, 040320 (2025)

    2025

  52. [52]

    A. W. Shrestha, B. Bhattacharjee, and A. del Campo, Shortcuts to adiabaticity for non-hermitian systems in krylov space (2026), arXiv:2607.07802 [quant-ph]

    Pith/arXiv arXiv 2026

  53. [53]

    Magnus, On the exponential solution of differential equations for a linear operator, Commun

    W. Magnus, On the exponential solution of differential equations for a linear operator, Commun. Pure Appl. Math.7, 649 (1954)

    1954

  54. [54]

    Blanes, F

    S. Blanes, F. Casas, J. A. Oteo, and J. Ros, The Magnus expansion and some of its applications, Phys. Rep.470, 151 (2009)

    2009

  55. [55]

    N. J. Higham, The scaling and squaring method for the matrix exponential revisited, SIAM Journal on Matrix Analysis and Applications26, 1179 (2005), https://doi.org/10.1137/04061101X

    work page doi:10.1137/04061101x 2005

  56. [56]

    Baker and P

    G. Baker and P. Graves-Morris,Pad´ e Approximants, En- cyclopedia of Mathematics and its Applications (Cam- bridge University Press, 1996)

    1996

  57. [57]

    I. A. Mart ´ ınez, A. Petrosyan, D. Gu´ ery-Odelin, E. Trizac, and S. Ciliberto, Engineered swift equilibration of a brownian particle, Nature Physics12, 843 (2016)

    2016

  58. [58]

    K. J. Laidler and M. C. King, Development of transition- state theory, The Journal of Physical Chemistry87, 2657 (1983)

    1983

  59. [59]

    Oberhofer, C

    H. Oberhofer, C. Dellago, and P. L. Geissler, Biased sam- pling of nonequilibrium trajectories: can fast switching simulations outperform conventional free energy calcula- tion methods?, The journal of physical chemistry. B109, 6902 (2005). 2 SUPPLEMENTAL MATERIAL I. ZERO-V ARIANCE FREE ENERGY ESTIMA TION VIA PERFECT BIJECTIVE MAPPINGS For an arbitrary se...

    2005