The reviewed record of science sign in
Pith

arxiv: 2607.06414 · v1 · pith:7I46RGA2 · submitted 2026-07-07 · physics.chem-ph · cond-mat.mtrl-sci

Economised path integrals

Zezhu Zeng , David E. Manolopoulos This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-07-08 06:27 UTCglm-5.2pith:7I46RGA2record.jsonopen to challenge →

classification physics.chem-ph cond-mat.mtrl-sci PACS 02.70.Ss31.15.xk67.10.Ba
keywords pathintegraltrotterfrequenciesmodeomegacalculationeconomised
0
0 comments X

The pith

Optimised ring-polymer frequencies match 4th-order accuracy at 2nd-order cost

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

Path integral molecular dynamics (PIMD) is the standard way to include nuclear quantum effects in materials simulations, but the most widely used discretisation (the primitive Trotter splitting) converges slowly with the number of ring-polymer beads P, often requiring hundreds of beads for properties like heat capacities. Higher-order schemes such as the 4th-order Suzuki-Chin method converge faster but demand expensive Hessian calculations and bespoke estimators. This paper identifies an underused structural freedom in the Trotter ring-polymer spring matrix: although the matrix is a symmetric circulant with a zero centroid eigenvalue and degenerate pairs, there remain floor(P/2) independent non-zero normal-mode frequencies that can be freely chosen without breaking any symmetry of the exact path integral. The authors propose to set these free frequencies not at the standard Trotter values but at values obtained by a least-squares fit to the exact quantum-mechanical radii of gyration of harmonic oscillators spanning the physical frequency range of the target system. The result is the economised (Eco) path integral: a ring-polymer Hamiltonian whose internal mode frequencies are tuned so that, in the harmonic limit, the polymer reproduces the correct quantum fluctuations far more accurately than Trotter for the same P. Crucially, because the Eco spring matrix retains the same symmetric circulant structure as Trotter, every standard Trotter estimator (thermodynamic, centroid virial, radial distribution function) carries over unchanged. The modification to an existing Trotter code amounts to replacing the circulant coefficients, which is a few-line change. The paper validates the approach on hexagonal ice at 100 K and on the metal-organic framework MOF-5 with a machine-learned potential. In ice, the Eco path integral converges as roughly P^{-4} (like a 4th-order method) rather than P^{-2} (Trotter), and achieves comparable or better per-force-evaluation accuracy than projected-Hessian Suzuki-Chin without needing any Hessian evaluation. In MOF-5, Eco converges the negative thermal expansion coefficient and the temperature-dependent heat capacity at P=128-256 where Trotter would need well over a thousand beads. The authors argue that Eco should become the default for path integral calculations because it delivers higher-order convergence at purely second-order Trot

Core claim

The central discovery is that the symmetric circulant structure of the Trotter ring-polymer spring matrix leaves floor(P/2) free normal-mode frequencies, and that tuning these frequencies by a simple least-squares fit to exact harmonic radii of gyration yields a path integral that converges as ~P^{-4} on anharmonic condensed-phase systems while remaining a pure second-order calculation with standard Trotter estimators and no Hessian evaluations.

What carries the argument

The key machinery is the symmetric circulant spring matrix K of the ring polymer, its normal-mode eigenvalues (the free frequencies omega_1 through omega_{floor(P/2)}), and the least-squares objective function s(y) that minimises the root-mean-square fractional error between the finite-P ring-polymer radius of gyration R^2(omega) and the exact quantum result over the frequency range 0 <= omega <= omega_max.

If this is right

  • Any existing Trotter PIMD code can be upgraded to Eco accuracy by replacing the circulant spring coefficients, which lowers the barrier to adoption for large-scale condensed-phase simulations with machine-learned potentials.
  • Properties that are prohibitively expensive with Trotter, such as heat capacities and thermal expansion coefficients of hydrogen-containing materials, become accessible at modest bead counts, broadening the range of systems where nuclear quantum effects can be routinely included.
  • The Eco spring matrix includes both attractive and repulsive inter-bead interactions (unlike Trotter, which has only nearest-neighbour attraction), and its highest normal-mode frequency is lower than Trotter's, which may permit slightly larger integration time steps.
  • The authors suggest extending the frequency-fitting procedure to include imaginary (parabolic barrier) frequencies, which could improve ring-polymer molecular dynamics and instanton rate calculations for chemical reactions involving tunnelling.

Load-bearing premise

The method is derived by fitting to harmonic oscillator radii of gyration, but its demonstrated P^{-4} convergence on anharmonic systems (ice, MOF-5) is shown empirically, not proven from first principles; if a system's anharmonic modes couple strongly to frequencies outside the fitted range, the acceleration could diminish.

What would settle it

A condensed-phase system whose anharmonic vibrational spectrum couples modes outside the fitted 0-to-omega_max range in a way that the harmonic radius-of-gyration fit does not capture, causing Eco to converge no faster than Trotter at the same P.

Figures

Figures reproduced from arXiv: 2607.06414 by David E. Manolopoulos, Zezhu Zeng.

Figure 1
Figure 1. Figure 1: FIG. 1. Root-mean-square fractional errors in [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Comparison of Trotter (black) and Eco (red) normal [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Comparison of Trotter (black) and Eco (red) circulant [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: (Here the error bars are standard errors in the [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Comparison of Trotter and Eco (a) kinetic and (b) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Quantum and classical radial distribution functions [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: compares the Te PIGS VDOS of ice at this temperature obtained from a fully converged (P = 32) 3300 3400 3500 3600 wavenumber (cm−1 ) 0 0.0005 0.001 0.0015 0.002 0.0025 VDOS (cm) P=1 P=2 P=4 P=8 P=16 (a) Eco 3300 3400 3500 3600 wavenumber (cm−1 ) 0 0.0005 0.001 0.0015 0.002 0.0025 VDOS (cm) P=1 P=2 P=4 P=8 P=16 (b) Trotter FIG. 9. Convergence of (a) Eco and (b) Trotter Te PIGS cal￾culations of the ice VDOS … view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. Convergence of (a) Eco and (b) Trotter T [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Two-dimensional projection of the MOF-5 crystal [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11. Temperature dependence of the average lattice con [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
read the original abstract

The Hessian of the ring polymer spring potential in the standard Trotter path integral is a $P\times P$ symmetric circulant matrix with a centroid eigenvalue of zero. All such matrices commute and are diagonalised by the same bead to normal mode transformation matrix, and their eigenvalues contain $\lceil P/2\rceil-1$ degenerate pairs by symmetry. However, this still leaves some freedom to improve on the Trotter approximation: one can optimise the remaining $\lfloor P/2\rfloor$ independent non-zero normal mode frequencies to fit the exact quantum mechanical radii of gyration of harmonic ring polymers with frequencies in the range $0\le\omega\le\omega_{\rm max}$, where $\omega_{\rm max}$ is the maximum physical frequency in the problem of interest. The optimisation involves solving a simple least squares problem for the optimum (economised or "Eco") internal mode frequencies. The remainder of the calculation then proceeds in the same way as a Trotter path integral calculation. An example application to hexagonal ice shows that the convergence of the Eco path integral is comparable to that of the 4th order Suzuki-Chin path integral, but with purely 2nd order Trotter effort. There is no need to calculate the projected Hessians that arise in the Suzuki-Chin method by finite differences, there is no need to develop any new estimators for observables, and once the Eco frequencies have been calculated the implementation of the Eco path integral involves changing just a few lines of a Trotter path integral code. To provide a more impressive example we have implemented the Eco method in GPUMD and used it to converge the (negative) thermal expansion coefficient and the constant pressure heat capacity of MOF-5 with a machine-learned neuroevolution potential.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 6 minor

Summary. The manuscript introduces an 'economised' (Eco) path integral in which the ⌊P/2⌋ independent non-zero normal-mode frequencies of the ring polymer spring matrix are optimized to reproduce the exact harmonic radius of gyration R²(ω) over a physically relevant frequency range. The construction exploits the fact that any real symmetric circulant P×P matrix with a zero centroid eigenvalue is diagonalized by the same bead-to-normal-mode transformation, leaving free parameters that can be tuned without breaking the symmetries of the path integral. The resulting Eco path integral uses standard Trotter estimators, requires no projected Hessians, and involves only a few lines of code change. Empirical validation on hexagonal ice (q-TIP4P/F, 100 K) and MOF-5 (machine-learned NEP, 60–100 K) shows P⁻⁴-like convergence in anharmonic observables, comparable to fourth-order Suzuki-Chin at second-order Trotter computational cost.

Significance. The central contribution is a genuinely practical acceleration of PIMD convergence that is essentially drop-in for existing Trotter codes. Specific strengths include: (1) a clean and correct mathematical derivation of the circulant algebra and estimator equivalence (Eqs. 5–13, 25); (2) a well-posed least-squares optimization (Eq. 17) with a provided reference Fortran implementation in the SI; (3) use of standard Trotter estimators with no new estimator derivation required; (4) falsifiable empirical predictions on two condensed-phase systems including a 3328-atom MOF-5 supercell with a machine-learned potential. The GPUMD patch, to be made public, enhances reproducibility. The method is likely to be adopted broadly if its correctness is established.

major comments (1)
  1. Sec. II.B, discussion following Fig. 2: The manuscript acknowledges that the Eco frequencies 'might all eventually tend to the Trotter frequencies in the large P limit' but does not prove that Z_P^{Eco} → Z as P→∞. For the standard Trotter path integral, convergence follows from the Trotter product formula; for the Eco path integral, the spring matrix is not derived from any operator splitting of e^{-βĤ}. The concern is that if the optimized Eco frequencies do not approach the Matsubara frequencies 2πk/(βℏ) as P→∞, the discretized path integral could converge to an incorrect limit. The empirical evidence (Figs. 4–5, 11–12) is supportive but limited to two systems at low temperatures. A more rigorous argument—even a heuristic one connecting the least-squares fit (Eq. 17) to the Matsubara limit, or a demonstration that the Eco circulant coefficients approach the Trotter coefficients κ_0=2,
minor comments (6)
  1. Sec. II.B: 'One the Eco frequencies have been found' should read 'Once the Eco frequencies have been found'.
  2. Appendix A title 'Pigs might fly' is informal for a journal article; a more descriptive title such as 'On-the-fly Te PIGS implementation' would be appropriate.
  3. Sec. IV: The value of ω_max used for the MOF-5 calculations is not stated explicitly (it is given as 4000 cm⁻¹ for ice in Sec. III). This should be specified.
  4. Fig. 5: The y-axis label 'T − T_P' could be misread; clarifying that this is |T − T_P| or the absolute error would help.
  5. Sec. III.A: The statement that Eco P=48 is 'better converged' than Trotter P=128 is supported by Fig. 4 but the quantitative basis (error bars) should be noted in the text, not only in the figure caption.
  6. Sec. V, Conclusion: The claim that 'Eco should become the default method for path integral calculations' is strong given the current evidence base of two systems; softening to 'a promising default' or similar would be more proportionate.

Simulated Author's Rebuttal

4 responses · 1 unresolved

The referee raises a single major comment: whether the Eco path integral can be proven to converge to the exact quantum partition function Z as P→∞, given that the Eco spring matrix is not derived from an operator splitting of e^{-βĤ}. We agree this is a legitimate and important theoretical question. We will add a new subsection to the revised manuscript providing a heuristic argument for convergence, supported by numerical evidence that the Eco circulant coefficients approach the Trotter coefficients as P increases. We also note a standing limitation: a fully rigorous proof of convergence in the P→∞ limit for general anharmonic potentials is not currently available, and we will state this honestly in the revised text.

read point-by-point responses

  1. Referee: The manuscript acknowledges that the Eco frequencies 'might all eventually tend to the Trotter frequencies in the large P limit' but does not prove that Z_P^{Eco} → Z as P→∞. The concern is that if the optimized Eco frequencies do not approach the Matsubara frequencies 2πk/(βℏ) as P→∞, the discretized path integral could converge to an incorrect limit. A more rigorous argument—even a heuristic one connecting the least-squares fit (Eq. 17) to the Matsubara limit, or a demonstration that the Eco circulant coefficients approach the Trotter coefficients κ_0=2, κ_1=κ_{P-1}=-1, κ_n=0 (n≥2) as P→∞—is requested.

    Authors: We thank the referee for raising this important point, which we agree deserves a more thorough treatment in the manuscript. We will address it through both a heuristic argument and additional numerical evidence in the revised text. revision: partial

  2. Referee: Heuristic argument connecting the least-squares fit to the Matsubara limit

    Authors: The key observation is as follows. The least-squares objective in Eq. (17) minimizes the rms fractional error in R²(ω) over the frequency range 0 ≤ ω ≤ ω_max. As P increases, the Trotter frequencies ω_k = 2ω_P sin(kπ/P) already converge to the Matsubara frequencies 2πk/(βℏ) in the sense that the Trotter R²(ω) converges to the exact R²(ω) as P→∞ (with O(P⁻²) error). The Eco optimization can only reduce the error below the Trotter value, so the Eco rms fractional error is bounded above by the Trotter rms fractional error, which itself tends to zero as P→∞. Therefore the Eco fit also becomes exact in the large-P limit, and the Eco frequencies must approach a set of frequencies that reproduces the exact R²(ω) — which is uniquely achieved by the Matsubara frequencies. More concretely, since the objective function s(y) is continuous in the frequency parameters {y_k} and has a unique global minimum (as can be verified from the structure of the problem), and since the Trotter frequencies provide an increasingly accurate starting point as P grows, the optimized Eco frequencies must converge to the same large-P limit as the Trotter frequencies, i.e., to the Matsubara frequencies. We will include this argument in a new paragraph in Section II.B of the revised manuscript. revision: yes

  3. Referee: Demonstration that the Eco circulant coefficients approach the Trotter coefficients as P→∞

    Authors: We agree that a numerical demonstration would strengthen the argument. In the current manuscript, Fig. 3 shows the Eco and Trotter circulant coefficients for P = 25, 50, and 100, and the text notes that 'an oscillatory pattern develops in the attractive and repulsive coefficients that brings them closer to their Trotter counterparts.' We will extend this analysis to larger values of P (up to P = 400 or 512) in the revised manuscript and include a quantitative measure of the difference between the Eco and Trotter circulant coefficients (e.g., ||κ^{Eco} - κ^{Trotter}||₂) as a function of P. Our preliminary calculations show that this difference decreases roughly as P⁻¹, consistent with the heuristic argument above. We will add a figure or table showing this convergence. revision: yes

  4. Referee: Concern that the Eco path integral could converge to an incorrect limit because the spring matrix is not derived from an operator splitting of e^{-βĤ}

    Authors: This is a fair concern. We note that the Eco path integral preserves all the symmetries of the exact path integral (cyclic permutation invariance, bead reversal symmetry, and centroid decoupling), as shown in Eqs. (5)–(7). These symmetries are sufficient to ensure that the standard Trotter estimators remain valid (Section II.C) and that the path integral has the correct classical limit as β→0. The additional requirement for convergence to the correct quantum limit is that the Eco frequencies approach the Matsubara frequencies as P→∞, for which we provide the heuristic argument above. However, we acknowledge that a fully rigorous proof of convergence for general anharmonic potentials — analogous to the Trotter product formula — is not currently available. We will state this limitation explicitly in the revised manuscript, noting that the empirical evidence (Figs. 4–5, 11–12) is consistent with convergence to the correct limit but that a formal proof remains an open problem. We believe this honest accounting is appropriate and does not undermine the practical utility of the method, which is validated empirically on two condensed-phase systems with consistent results. revision: yes

standing simulated objections not resolved
  • A fully rigorous proof that Z_P^{Eco} → Z as P→∞ for general anharmonic potentials is not currently available. The heuristic argument and numerical evidence we provide are supportive but not a substitute for a formal proof. We will state this honestly in the revised manuscript.

Circularity Check

0 steps flagged

No circularity found: the Eco frequencies are fitted to an external analytical benchmark (exact harmonic R²(ω)), and the anharmonic application results are genuine predictions not used in the fitting.

full rationale

The paper's derivation chain is self-contained and non-circular. The Eco frequencies are obtained by minimizing the rms fractional error in the exact quantum mechanical radius of gyration R²(ω) (Eq. 14, left side — an independent analytical formula from quantum mechanics) against the discrete path integral approximation (Eq. 14, right side). The optimization target (Eq. 17) depends only on the exact R²(ω) and the free parameters {y_k}, not on any system being simulated. The anharmonic convergence results for hexagonal ice (Figs. 4-5) and MOF-5 (Figs. 11-12) are genuine predictions: the frequencies were fitted to harmonic oscillator physics, not to ice or MOF-5 data, and the paper explicitly acknowledges that the exponential harmonic convergence does not necessarily carry over to anharmonic problems ('Of course this does not imply that the convergence of the Eco path integral will also be exponential when it is applied to anharmonic problems'). The estimator derivations (Sec. II.C, Eqs. 18-27) follow straightforwardly from the symmetric circulant structure of K and do not depend on the specific frequency values. Self-citations (e.g., to PIGLET Ref. 11, to GPUMD Ref. 30) are not load-bearing for the central theoretical claim. The Hunt-Althorpe citation (Ref. 14) provides inspiration but the actual optimization formulation is self-contained in this paper. The skeptic's concern about whether the Eco path integral converges to the correct limit as P→∞ is a correctness risk, not a circularity issue — the paper itself flags this as an open question rather than hiding it.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 0 invented entities

The method introduces no new physical entities, particles, or forces. The Eco frequencies are optimized parameters within the standard path integral framework, not new postulated objects. The axiom ledger is minimal: three standard mathematical results and one domain assumption that is the load-bearing premise of the method.

free parameters (2)
  • ω_max (maximum physical frequency) = 4000 cm⁻¹ (ice), system-dependent (MOF-5)
    The upper limit of the frequency range used in the least-squares fit (Eq. 17). It is a physical input from the problem, not fitted to the simulation output, but it is a user-specified parameter that affects the Eco frequencies.
  • Eco normal-mode frequencies {ω_k} = Determined by least-squares minimization of Eq. 17
    These are the ⌊P/2⌋ independent frequencies optimized to fit the exact harmonic R²(ω). They are fitted to an external analytical benchmark, not to the simulation results themselves.
axioms (4)
  • standard math The ring polymer spring potential Hessian K is a real symmetric circulant matrix with centroid eigenvalue zero (Eqs. 5-7).
    This follows from the Trotter product structure and the cyclic/reversal symmetries of the path integral (Sec. II.A). Standard result from circulant matrix theory (Ref. 15).
  • standard math All real symmetric P×P circulant matrices commute and are diagonalized by the same orthogonal transformation (Eq. 9).
    Standard result in circulant matrix theory, cited to Davis (Ref. 15).
  • standard math The exact quantum radius of gyration R²(ω) for a harmonic oscillator is given by the left side of Eq. (14).
    Standard quantum mechanics result for the harmonic oscillator partition function.
  • domain assumption Improving the harmonic radius of gyration approximation will improve convergence for anharmonic systems.
    This is the key domain assumption underlying the method. It is not proven but is empirically validated on ice and MOF-5. The authors acknowledge it is not guaranteed (Sec. II.B).

pith-pipeline@v1.1.0-glm · 20681 in / 2874 out tokens · 314265 ms · 2026-07-08T06:27:21.617634+00:00 · methodology

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

46 extracted references · 46 canonical work pages

  1. [1]

    Physical Review Letters , year =

    J. Physical Review Letters , year =. doi:10.1103/PhysRevLett.98.146401 , url =

    work page doi:10.1103/physrevlett.98.146401

  2. [2]

    Albert P. Bart. Physical Review Letters , year =. doi:10.1103/PhysRevLett.104.136403 , url =

    work page doi:10.1103/physrevlett.104.136403

  3. [3]

    Wolynes , title =

    David Chandler and Peter G. Wolynes , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.441588 , url =

    work page doi:10.1063/1.441588

  4. [4]

    Markland and Michele Ceriotti , title =

    Thomas E. Markland and Michele Ceriotti , title =. Nature Reviews Chemistry , year =. doi:10.1038/s41570-018-0109 , url =

    work page doi:10.1038/s41570-018-0109

  5. [5]

    and Althorpe, Stuart C

    Hunt, Andrew C. and Althorpe, Stuart C. , title =. The Journal of Chemical Physics , year =. doi:10.1063/5.0314385 , url =

    work page doi:10.1063/5.0314385

  6. [6]

    Feynman and Albert R

    Richard P. Feynman and Albert R. Hibbs , title =

    work page

  7. [7]

    Manolopoulos , title =

    Mariana Rossi and Michele Ceriotti and David E. Manolopoulos , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.4883861 , url =

    work page doi:10.1063/1.4883861

  8. [8]

    Davis , title =

    Philip J. Davis , title =. 1994 , isbn =

    work page 1994

  9. [9]

    Markland and David E

    Michele Ceriotti and Michele Parrinello and Thomas E. Markland and David E. Manolopoulos , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.3489925 , url =

    work page doi:10.1063/1.3489925

  10. [10]

    Manolopoulos and Michele Parrinello , title =

    Michele Ceriotti and David E. Manolopoulos and Michele Parrinello , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.3556661 , url =

    work page doi:10.1063/1.3556661

  11. [11]

    Manolopoulos , title =

    Michele Ceriotti and David E. Manolopoulos , title =. Physical Review Letters , year =. doi:10.1103/PhysRevLett.109.100604 , url =

    work page doi:10.1103/physrevlett.109.100604

  12. [12]

    Chemical Science , year =

    Igor Poltavsky and Alexandre Tkatchenko , title =. Chemical Science , year =. doi:10.1039/C5SC03443D , url =

    work page doi:10.1039/c5sc03443d

  13. [13]

    Journal of the Physical Society of Japan , year =

    Minoru Takahashi and Masatoshi Imada , title =. Journal of the Physical Society of Japan , year =. doi:10.1143/JPSJ.53.3765 , url =

    work page doi:10.1143/jpsj.53.3765

  14. [14]

    Physics Letters A , year =

    Masuo Suzuki , title =. Physics Letters A , year =. doi:10.1016/0375-9601(95)00266-6 , url =

    work page doi:10.1016/0375-9601(95)00266-6

  15. [15]

    Chin , title =

    Siu A. Chin , title =. Physics Letters A , year =. doi:10.1016/S0375-9601(97)00003-0 , url =

    work page doi:10.1016/s0375-9601(97)00003-0

  16. [16]

    The Journal of Chemical Physics , year =

    Venkat Kapil and J. The Journal of Chemical Physics , year =. doi:10.1063/1.4971438 , url =

    work page doi:10.1063/1.4971438

  17. [17]

    M. F. Herman and E. J. Bruskin and B. J. Berne , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.442815 , url =

    work page doi:10.1063/1.442815

  18. [18]

    The Journal of Chemical Physics , year =

    Michele Parrinello and Aneesur Rahman , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.446900 , url =

    work page doi:10.1063/1.446900

  19. [19]

    Markland and David E

    Scott Habershon and Thomas E. Markland and David E. Manolopoulos , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.3167790 , url =

    work page doi:10.1063/1.3167790

  20. [20]

    Hayward and Jeffrey R

    Jennifer A. Hayward and Jeffrey R. Reimers , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.473300 , url =

    work page doi:10.1063/1.473300

  21. [21]

    The Journal of Chemical Physics , year =

    Giovanni Bussi and Davide Donadio and Michele Parrinello , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.2408420 , url =

    work page doi:10.1063/1.2408420

  22. [22]

    Willatt and Stuart C

    George Trenins and Michael J. Willatt and Stuart C. Althorpe , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.5100587 , url =

    work page doi:10.1063/1.5100587

  23. [23]

    Benson and George Trenins and Stuart C

    Benjamin E. Benson and George Trenins and Stuart C. Althorpe , title =. The Journal of Chemical Physics , year =. doi:10.1063/5.0065266 , url =

    work page doi:10.1063/5.0065266

  24. [24]

    Lawrence and David E

    Theo Fletcher and Andrew Zhu and Joseph E. Lawrence and David E. Manolopoulos , title =. The Journal of Chemical Physics , year =. doi:10.1063/5.0076704 , url =

    work page doi:10.1063/5.0076704

  25. [25]

    Lawrence and Annina Z

    Joseph E. Lawrence and Annina Z. Lieberherr and Theo Fletcher and David E. Manolopoulos , title =. The Journal of Physical Chemistry B , year =. doi:10.1021/acs.jpcb.3c05028 , url =

    work page doi:10.1021/acs.jpcb.3c05028

  26. [26]

    The Journal of Chemical Physics , year =

    F. The Journal of Chemical Physics , year =. doi:10.1063/5.0120386 , url =

    work page doi:10.1063/5.0120386

  27. [27]

    Cao and G

    J. Cao and G. J. Martyna , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.470959 , url =

    work page doi:10.1063/1.470959

  28. [28]

    Ivanov and Ian M

    Sergei D. Ivanov and Ian M. Grant and Dominik Marx , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.3290958 , url =

    work page doi:10.1063/1.3290958

  29. [29]

    Magnus Rahm and Alexander J

    Zheyong Fan and Yanzhou Wang and Penghua Ying and Keke Song and Junjie Wang and Yong Wang and Zezhu Zeng and Ke Xu and Eric Lindgren and J. Magnus Rahm and Alexander J. Gabourie and Jiahui Liu and Haikuan Dong and Jianyang Wu and Yue Chen and Zheng Zhong and Jian Sun and Paul Erhart and Yanjing Su and Tapio Ala-Nissila , title =. The Journal of Chemical P...

    work page doi:10.1063/5.0106617

  30. [30]

    Physical Review B , volume =

    Zheyong Fan and Zezhu Zeng and Cunzhi Zhang and Yanzhou Wang and Keke Song and Haikuan Dong and Yue Chen and Tapio Ala-Nissila , title =. Physical Review B , volume =. 2021 , publisher =. doi:10.1103/PhysRevB.104.104309 , url =

    work page doi:10.1103/physrevb.104.104309 2021

  31. [31]

    Manolopoulos , title =

    Zezhu Zeng and David E. Manolopoulos , title =. 2026 , publisher =

    work page 2026

  32. [32]

    Cameron and Vanessa K

    Nina Lock and Yue Wu and Mogens Christensen and Lisa J. Cameron and Vanessa K. Peterson and Adam J. Bridgeman and Cameron J. Kepert and Bo B. Iversen , title =. The Journal of Physical Chemistry C , volume =. 2010 , publisher =. doi:10.1021/jp103212z , url =

    work page doi:10.1021/jp103212z 2010

  33. [33]

    F. A. Kloutse and R. Zacharia and D. Cossement and R. Chahine , title =. Microporous and Mesoporous Materials , volume =. 2015 , publisher =. doi:10.1016/j.micromeso.2015.05.047 , url =

    work page doi:10.1016/j.micromeso.2015.05.047 2015

  34. [34]

    npj Computational Materials , volume =

    Sandro Wieser and Egbert Zojer , title =. npj Computational Materials , volume =. 2024 , doi =

    work page 2024

  35. [35]

    The Journal of Chemical Physics , volume =

    Penghua Ying and Wenjiang Zhou and Lucas Svensson and Esm. The Journal of Chemical Physics , volume =. 2025 , doi =

    work page 2025

  36. [36]

    Proceedings of the National Academy of Sciences of the United States of America , volume =

    Zezhu Zeng and Zheyong Fan and Michele Simoncelli and Chen Chen and Ting Liang and Yue Chen and Geoff Thornton and Bingqing Cheng , title =. Proceedings of the National Academy of Sciences of the United States of America , volume =. 2025 , doi =

    work page 2025

  37. [37]

    Goodfellow and Bao N

    Alister S. Goodfellow and Bao N. Nguyen , title =. Journal of Chemical Theory and Computation , year =. doi:10.1021/acs.jctc.5c02073 , url =

    work page doi:10.1021/acs.jctc.5c02073

  38. [38]

    The Journal of Chemical Physics , year =

    Toshiyuki Yamamoto , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.2013257 , url =

    work page doi:10.1063/1.2013257

  39. [39]

    Lawrence and David E

    Joseph E. Lawrence and David E. Manolopoulos , title =. Faraday Discussions , year =. doi:10.1039/C9FD00084D , url =

    work page doi:10.1039/c9fd00084d

  40. [40]

    Craig and David E

    Ian R. Craig and David E. Manolopoulos , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.1850093 , url =

    work page doi:10.1063/1.1850093

  41. [41]

    Craig and David E

    Ian R. Craig and David E. Manolopoulos , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.2772629 , url =

    work page doi:10.1063/1.2772629

  42. [42]

    Richardson and Stuart C

    Jeremy O. Richardson and Stuart C. Althorpe , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.3267318 , url =

    work page doi:10.1063/1.3267318

  43. [43]

    Suleimanov and F

    Yury V. Suleimanov and F. Javier Aoiz and Hua Guo , title =. The Journal of Physical Chemistry A , year =. doi:10.1021/acs.jpca.6b07861 , url =

    work page doi:10.1021/acs.jpca.6b07861

  44. [44]

    Wodtke and Hua Guo , title =

    Liang Zhang and Florian Nitz and Dmitriy Borodin and Alec M. Wodtke and Hua Guo , title =. Precision Chemistry , year =. doi:10.1021/prechem.4c00099 , url =

    work page doi:10.1021/prechem.4c00099

  45. [45]

    Richardson , title =

    Jeremy O. Richardson , title =. The Journal of Chemical Physics , year =. doi:10.1063/1.5028352 , url =

    work page doi:10.1063/1.5028352

  46. [47]

    Ansari and Eric R

    Imaad M. Ansari and Eric R. Heller and George Trenins and Jeremy O. Richardson , title =. Nature Communications , year =. doi:10.1038/s41467-024-48463-2 , url =

    work page doi:10.1038/s41467-024-48463-2