Pith. sign in

REVIEW 25 references

Alchemical thermodynamic integration for ab initio free-energy calculations in solutions

T0 review · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read By mixing forces from two density-functional calculators during a single simulation, this paper turns ab initio mixing free energies of liquid alloys into a routine calculation.

desk verdict Solid alchemical-TI implementation with a real but fixable gap in the Li-Na free-energy validation. read the letter →

arxiv 2607.18750 v1 pith:FZDF5G6P submitted 2026-07-21 cond-mat.mtrl-sci

Liangrui Wei , Feng Zhang , Renata M. Wentzcovitch , Kai-Ming Ho , Yang Sun This is my paper · ORCID
classification cond-mat.mtrl-sci
keywords alchemicalthermodynamicintegrationfree-energycalculationabinitiomoleculardynamicsliquidalloysmiscibilitygapLi-Nasolutionequationofstatephasediagram
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper aims to make Gibbs free energies of liquid solutions calculable from first principles at molecular-dynamics cost. It does so by alchemical thermodynamic integration: during one Monte Carlo or molecular dynamics run, two density-functional calculators evaluate the forces of the pure and doped systems on the same atomic configuration, and the equations of motion use their linearly mixed forces. The free-energy difference between pure A and solution A1-xBx is then an integral of the energy difference along the mixing parameter λ, plus an analytic ideal-mixing term and an equation-of-state volume correction. Validations on a classical Fe-Ni potential reproduce an existing mixed-potential implementation, ab initio Fe90Ni10 matches a prior regular-solution result to 0.001 eV/atom, and Li-Na reproduces the observed miscibility gap. If correct, this gives a practical route to ab initio phase diagrams of solutions.

What carries the argument

The alchemical thermodynamic integration (ATI) path: an auxiliary end state in which solute atoms interact as solvent atoms (same mass, A-like forces), with a linear coupled Hamiltonian H_λ=(1-λ)H_reference+λH_target and coupled forces f_λ=(1-λ)f_reference+λf_target. The free-energy difference is obtained as ∫⟨U_target-U_reference⟩_λ dλ from two concurrently running DFT calculators that share the same ionic trajectory; an analytic ideal free-energy term F_mass accounts for the mass/identity change, and a volume term F_PV accounts for the equation-of-state difference between solution and pure endmember. This machinery allows thermodynamic integration without explicit analytical Hamiltonians,

What would settle it

Recompute the Li-Na F_PV using the raw ab initio P(V) data points (or a more flexible EOS) over the full compression range and re-fit the Redlich-Kister curve; if the phase boundaries move by more than about 1-2 meV/atom, the reported miscibility gap is an artifact of the EOS extrapolation. Alternatively, run the ATI path at x_Li=0.85 or 0.9 with more λ-points and longer sampling to see whether the free-energy curve crosses the convex hull at the same points.

Watch

Extended reading notes

Core claim

The central claim is that free energies of mixing for liquid solutions can be obtained ab initio by coupling two DFT force calculators on the fly, without writing an explicit Hamiltonian. For a system A1-xBx, the code simulates an auxiliary system where B' has B's mass but A's interactions, and integrates ⟨U_A1-xBx - U_A1-xB'x⟩ over λ using forces f_λ=(1-λ)f_A+λf_B. Combined with the ideal entropy of mixing and a volume/EOS correction, this yields Gibbs free energies of mixing at fixed pressure. In the Fe-Ni test it reproduces the regular-solution value of 0.143 eV/atom, and in Li-Na it predicts liquid-liquid phase separation between roughly x_Li=0.1 and 0.95, matching experiment. The paper

Load-bearing premise

In the Li-Na application, the large F_PV term is evaluated by extrapolating a third-order Birch-Murnaghan fit to pure liquid sodium's P(V) relation to volumes nearly half of its equilibrium volume, and the paper does not quantify how sensitive the predicted phase boundaries are to that extrapolation.

Editorial extensions

If this is right

  • Mixing free energies of liquid alloys and solutions become accessible at roughly AIMD wall time, replacing entropy-model approximations in phase-diagram construction.
  • The same integrator can use any force/energy calculator, including machine-learned potentials, since it only requires forces and energies, not an explicit Hamiltonian.
  • The Fe-Ni benchmark gives a quantitative check: the alchemical result agrees with the regular-solution model to within about 0.001 eV/atom at 6000 K and 323 GPa, supporting use of inexpensive entropy models in similar systems.
  • The Li-Na application shows that the method can capture liquid-liquid immiscibility from first principles, with a predicted gap that matches experiment on the Li-rich side.
  • Combining ATI with an equation-of-state term extends the method to high pressures, where pure and solution volumes differ substantially.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the two DFT runs share the same trajectory, their wavefunction extrapolation makes the overhead small; a direct test would be to run the same protocol with only one calculator restarted at every step, but the paper's wall-time measurement suggests most of the cost is inside the electronic-structure solve.
  • The method's dependence on the pure-sodium equation of state for the Li-Na F_PV term is the softest point: the third-order Birch-Murnaghan fit is extrapolated to volumes far below the pure-Na equilibrium volume, and a different EOS ansatz could shift the phase boundary by more than the quoted 1-2 meV/atom uncertainty.
  • The swap-MC variant appears to converge closer to equilibrium in phase-separating systems (0.004 eV/atom lower at λ=1), suggesting ATI combined with Monte Carlo swaps could map homogeneous free energies inside miscibility gaps, where plain MD is biased.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; Fe–Ni uses same-group prior work but the central derivation and Li–Na/classical benchmarks are independent.

full rationale

The derivation chain is standard and self-contained. Equations (1)–(7) are textbook statistical mechanics: F_mass is the exact ideal-mixing free energy, F_TI is the Hamiltonian thermodynamic-integration integral over coupled Hamiltonians, and F_PV follows directly from G = F + PV with an EOS for the reference endmember. None of these quantities is defined in terms of the target free-energy curve or phase boundaries, so there is no self-definitional or fitted-input-called-prediction reduction. The classical benchmark against the native LAMMPS pair/hybrid implementation is an external, code-level test that does not depend on any fitted parameter of this paper. The Li–Na application is compared with the independent AIMD-based results of Huang et al. and with the experimental miscibility gap, neither of which enters the ATI derivation or the EOS fit. The only self-citation worth noting is the Fe–Ni comparison against Ref. [18], which shares authors with the present paper (Sec. 3.2). That comparison is used as a numerical point of agreement rather than as a foundational premise, and the central claim that the scheme is practical and accurate does not reduce to it; the independent classical and Li–Na comparisons carry that claim. The unquantified sensitivity of F_PV to the Birch–Murnaghan fit to liquid-Na P(V) is a correctness/robustness concern, not a circularity, because the EOS is an input fitted to AIMD P–V data, not a renamed version of the predicted free energy. Overall, the paper is not circular; at most it has one minor self-citation that does not bear the weight of the argument.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No new physical entities are introduced. The method rests on standard statistical mechanics plus several domain assumptions about DFT accuracy, EOS extrapolation, and ergodicity of short AIMD/MCMD runs.

free parameters (4)
  • Na Birch-Murnaghan EOS parameters (B0, B0', V0) = B0=5.143 GPa, B0'=3.943, V0=41.552 Å3/atom
    Fitted to AIMD P(V) of liquid Na (Fig. 4c); used in Eq. (7) to compute F_PV over a wide compression range.
  • Volume-composition quadratic fit = Red line in Fig. 4(b)
    Fitted to AIMD equilibrium volumes of LixNa1-x to set V_sol for each ATI run.
  • Redlich-Kister coefficients = 2.9776 eV/atom, 0.6524 eV/atom
    Fitted to the computed ΔG(x) curve (Fig. 5d) to define the phase-separation boundary; the boundaries are extracted from this fit.
  • Quadratic fits to alchemical integrand = Per-composition polynomial coefficients
    Used to integrate ⟨U_B−U_A⟩ over λ; a modeling choice, not a physical parameter, but affects F_TI if the integrand is not truly quadratic.
assumptions (5)
  • domain assumption PBE-GGA + PAW DFT accurately describes Fe-Ni and Li-Na liquid energetics and volumes (DFT volume underestimated by ~2%).
    Used throughout; known DFT limitation acknowledged in Section 3.3.
  • domain assumption Nuclei are classical; Fmass in Eq. (1) uses the ideal mixing/mass expression.
    Standard in AIMD; the mass term in Eq. (1) is exact only for classical nuclei with no quantum nuclear effects.
  • domain assumption A third-order Birch-Murnaghan EOS form is valid for liquid Na P(V) over the volume range from pure Na down to Li-rich solution volumes.
    Required for F_PV in Eq. (7); no stability check or alternative EOS form is tested.
  • domain assumption The λ-linear coupled Hamiltonian H_λ=(1−λ)H_A+λH_B has a smooth reversible path with no phase transition along λ.
    Needed for TI integrand to be well-behaved; supported by the smooth integrands in Figs. 3(b) and 5(a).
  • domain assumption MCMD at λ=1 for Li0.6Na0.4 samples the homogeneous metastable liquid, not a phase-separated state.
    The system is inside the miscibility gap; the paper asserts homogeneity based on 10 ps runs and PCFs (inset of Fig. 5b).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Alchemical thermodynamic integration for ab initio free-energy calculations in solutions." pith.science (2026). https://pith.science/paper/FZDF5G6P

@misc{pith2026260718750,
  author       = {Pith},
  title        = {Pith review of: Alchemical thermodynamic integration for ab initio free-energy calculations in solutions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FZDF5G6P}},
  note         = {Machine review of arXiv:2607.18750}
}
read the original abstract

We develop an alchemical thermodynamic integration scheme that couples ab initio force calculators on the fly during Monte Carlo and molecular dynamics simulations. The implementation is validated against existing hybrid-Hamiltonian approaches. The scheme yields ab initio free energies of high-pressure Fe-Ni and ambient Li-Na liquid solutions that agree with previous calculations and reproduce the experimentally observed Li-Na miscibility gap. The code has an efficiency comparable to standard ab initio molecular dynamics. These results establish this scheme as a practical alchemical-integration framework for first-principles free-energy calculations in solutions.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references

  1. [1]

    P.Y. Chew, A. Reinhardt, The Journal of Chemical Physics 158 (2023) 030902

    2023

  2. [2]

    A. Sher, M. Van Schilfgaarde, A.-B. Chen, W. Chen, Phys. Rev. B 36 (1987) 4279–4295

    1987

  3. [3]

    Baranyai, D.J

    A. Baranyai, D.J. Evans, Phys. Rev. A 40 (1989) 3817–3822

    1989

  4. [4]

    Paula Leite, R

    R. Paula Leite, R. Freitas, R. Azevedo, M. De Koning, The Journal of Chemical Physics 145 (2016) 194101

    2016

  5. [5]

    Conclusions We have developed an ATI code for evaluating free energies in solution systems. The implementation couples two resident force and energy calculators to an external MC/MD runner, allowing the forces and energies from two alchemical end states to be combined on the fly during MC/MD simulations. We validated the implementation using both classica...

  6. [6]

    Van Der Hoef, The Journal of Chemical Physics 113 (2000) 8142–8148

    M.A. Van Der Hoef, The Journal of Chemical Physics 113 (2000) 8142–8148

    2000

  7. [7]

    S u n , M

    Y. S u n , M . I . M e n d e l e v, F. Z h a n g , X . L i u , B . D a , C . Wang, R.M. Wentzcovitch, K. Ho, Geophysical Research Letters 50 (2023) e2022GL102447

    2023

  8. [8]

    Alfè, G.D

    D. Alfè, G.D. Price, M.J. Gillan, Phys. Rev. B 65 (2002) 165118

    2002

Show all 25 references
  1. [9]

    Z. Wu, C. Gao, F. Zhang, S. Wu, K.-M. Ho, R.M. Wentzcovitch, Y. Sun, Phys. Rev. Lett. 136 (2026) 016103

    2026

  2. [10]

    F i d a l g o C â n d i d o , F

    V. F i d a l g o C â n d i d o , F. M a t u s a l e m , M . D e K o n i n g , T h e Journal of Chemical Physics 158 (2023) 064502

    2023

  3. [11]

    L. Yang, Y . Sun, Z. Ye, F. Zhang, M.I. Mendelev, C.Z. Wang, K.M. Ho, Computational Materials Science 150 (2018) 353–357

    2018

  4. [12]

    Kresse, J

    G. Kresse, J. Furthmüller, Phys. Rev. B 54 (1996) 11169–11186

    1996

  5. [13]

    Wentzcovitch, J.L

    R.M. Wentzcovitch, J.L. Martins, P.B. Allen, Phys. Rev. B 45 (1992) 11372–11374

    1992

  6. [14]

    Mermin, Phys

    N.D. Mermin, Phys. Rev. 137 (1965) A1441–A1443

    1965

  7. [15]

    Nosé, The Journal of Chemical Physics 81 (1984) 511–519

    S. Nosé, The Journal of Chemical Physics 81 (1984) 511–519

    1984

  8. [16]

    Thompson, H.M

    A.P. Thompson, H.M. Aktulga, R. Berger, D.S. Bolintineanu, W.M. Brown, P.S. Crozier, P.J. In ’T Veld, A. Kohlmeyer, S.G. Moore, T.D. Nguyen, R. Shan, M.J. Stevens, J. Tranchida, C. Trott, S.J. Plimpton, Computer Physics Communications 271 (2022) 108171

    2022

  9. [17]

    Sadigh, P

    B. Sadigh, P. Erhart, A. Stukowski, A. Caro, E. Martinez, L. Zepeda-Ruiz, Phys. Rev. B 85 (2012) 184203

    2012

  10. [18]

    Kresse, J

    G. Kresse, J. Hafner, Phys. Rev. B 47 (1993) 558–561

    1993

  11. [19]

    L. Wei, Z. Wu, K.-M. Ho, R.M. Wentzcovitch, Y . Sun, Science Advances 11 (2025) eadu1998

    2025

  12. [20]

    J. Jost, D. Heydt, J. Spehr, H. Ruppersberg, J. Phys.: Condens. Matter 6 (1994) 321–326

    1994

  13. [21]

    Khairulin, R.N

    R.A. Khairulin, R.N. Abdullaev, S.V . Stankus, J. Engin. Thermophys. 28 (2019) 472–483

    2019

  14. [22]

    H u a n g , M

    Y. H u a n g , M . Wi d o m , M . C . G a o , P h y s . R e v. M a t e r i a l s 6 (2022) 013802

    2022

  15. [23]

    P. Haas, F. Tran, P. Blaha, Phys. Rev. B 79 (2009) 085104

    2009

  16. [24]

    Lejaeghere, G

    K. Lejaeghere, G. Bihlmayer, T. Björkman, et al., Science 351 (2016) aad3000

    2016

  17. [25]

    Redlich, A.T

    O. Redlich, A.T. Kister, Ind. Eng. Chem. 40 (1948) 345–348

    1948

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.