ChemFit: A framework for automated high-dimensional model parameter optimization

Alberto Ocana; Amrita Goswami; Elvar \"O. J\'onsson; Hannes J\'onsson; Jorge R. Espinosa; Moritz Sallermann; Rosana Collepardo-Guevara

arxiv: 2603.11769 · v2 · submitted 2026-03-12 · ⚛️ physics.chem-ph · cond-mat.mes-hall

ChemFit: A framework for automated high-dimensional model parameter optimization

Moritz Sallermann , Amrita Goswami , Rosana Collepardo-Guevara , Alberto Ocana , Hannes J\'onsson , Elvar \"O. J\'onsson , Jorge R. Espinosa This is my paper

Pith reviewed 2026-05-15 12:03 UTC · model grok-4.3

classification ⚛️ physics.chem-ph cond-mat.mes-hall

keywords ChemFitparameter optimizationblack-box optimizationforce field parameterizationmolecular simulationcoarse-grained modelscomputational chemistrymultiscale modeling

0 comments

The pith

ChemFit provides a Python framework to automate optimization of simulation model parameters using black-box algorithms on composite objective functions.

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

The paper introduces ChemFit as a flexible Python framework for defining, composing, and evaluating objective functions from simulation outputs to fit model parameters automatically. This targets the common problem of laborious manual iteration in computational chemistry by supporting concurrent runs with gradient-free optimizers suited to noisy, non-differentiable, or heterogeneous data. Demonstrations include fitting Lennard-Jones parameters to argon density, a polarizable water potential to DFT cluster structures, and a coarse-grained protein force field to experimental critical temperatures for hnRNPA1 variants. A sympathetic reader would care because the method promises scalable and reproducible parameterization for multiscale models without requiring analytic derivatives or model-specific code.

Core claim

ChemFit enables the definition, composition, and massively concurrent evaluation of simulation-based objective functions for use with black-box optimization algorithms, as demonstrated by successful parameterization in three cases of increasing complexity: Lennard-Jones for liquid argon from density, polarizable flexible potential for H2O clusters from DFT, and residue-level coarse-grained force field for hnRNPA1 protein to reproduce critical solution temperatures.

What carries the argument

The ChemFit framework for definition, composition, and concurrent evaluation of objective functions from heterogeneous simulation outputs, allowing optimizer-agnostic fitting.

If this is right

Lennard-Jones parameters for simple liquids can be fitted directly from experimental density measurements.
Polarizable and flexible potentials can be optimized to match quantum-derived structures of small molecular clusters.
Residue-level coarse-grained force fields can be tuned to experimental thermodynamic properties such as critical solution temperatures.
Parameter fitting becomes scalable and reproducible across multiscale models by composing objective functions from separate simulations.
The same framework works with any gradient-free optimizer without changes to the model or objective definition.

Where Pith is reading between the lines

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

The approach could be adapted to parameter optimization tasks in adjacent fields such as materials science or soft matter simulations.
Automated composition of objectives might reduce variability across independent research groups developing force fields for the same systems.
Coupling ChemFit to emerging global optimization techniques could extend its reach to even higher-dimensional parameter spaces.

Load-bearing premise

That suitable objective functions can be defined from heterogeneous simulation outputs and that black-box optimizers will locate useful parameter sets without excessive cost or trapping in poor local minima.

What would settle it

Running the hnRNPA1 example through ChemFit with multiple optimizers and finding no parameter sets that reproduce the experimental critical solution temperatures for both sequences within error bars after feasible computation time.

read the original abstract

The parameterization of simulation-based models is a central yet laborious task in computational chemistry and physics, often driven by human intuition and manual iteration. Automating this task necessitates the definition of suitable objective functions, which tend to be expensive to evaluate, noisy, non-differentiable, or composed of heterogeneous contributions originating from separate sets of simulations. Gradient-free and black-box optimization algorithms are powerful tools which are particularly well-suited to minimizing such objective functions. Here, we introduce ChemFit, a flexible Python framework for the definition, composition, and massively concurrent evaluation of simulation-based objective functions, which is designed to operate in conjunction with these algorithms. We demonstrate the broad applicability of this approach by using ChemFit for three representative examples of increasing complexity and real-world relevance. First, we obtain the parameters of the Lennard-Jones potential for liquid argon from experimental measurements of the density. Second, we parameterize a polarizable and flexible potential energy function to reproduce the structure of small H$_2$O clusters obtained from density functional theory calculations. Finally, we tune a small subset of the parameters of a residue-level coarse-grained protein force-field, with the goal to reproduce the experimental critical solution temperature of the low complexity domain of the wild-type hnRNPA1 sequence and an arginine-enriched mutant of this protein. hnRNPA1 is an RNA-binding protein linked to amyotrophic lateral sclerosis. Together, these examples illustrate how ChemFit enables scalable, reproducible, and optimizer-agnostic parameter fitting for broadly applicable multiscale models.

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.

Desk Editor's Note private letter to a colleague

ChemFit is a useful new Python framework for composing and concurrently evaluating heterogeneous objective functions from simulations, but the high-dimensional claims rest on low-dimensional demonstrations only.

read the letter

ChemFit gives users a way to define objective functions that pull from different simulation runs, combine them, and run the evaluations in parallel before feeding them to black-box optimizers. The three examples show this in practice: fitting Lennard-Jones parameters to argon density data, tuning a polarizable water model against DFT cluster structures, and adjusting a handful of parameters in a coarse-grained protein force field to match experimental critical solution temperatures for hnRNPA1 and a mutant. These are real tasks that people currently do by hand, so the framework addresses a genuine workflow pain point and makes the process more reproducible and optimizer-agnostic. The code and composition features appear to be the actual novelty here. The main limitation is that none of the demonstrations use more than a small number of free parameters. The title and abstract emphasize high-dimensional optimization and scalability, yet the reported cases stay in the low-dimensional regime where black-box methods already work reasonably well. There is no evidence yet that the concurrency or objective-composition tools actually reduce the evaluation burden or help escape poor minima once the parameter count grows to dozens or more. The paper also does not report quantitative timings or success rates across repeated runs, which would help judge whether the claimed practicality holds up. This work is aimed at computational chemists who routinely fit multiscale models and want a structured way to automate the objective-function side. A reader already doing similar fitting will get immediate practical value from the framework description and examples. It is worth sending to peer review because the engineering contribution is concrete and the applications are relevant, even though the scalability argument needs more direct testing in higher dimensions.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces ChemFit, a flexible Python framework for defining, composing, and concurrently evaluating simulation-based objective functions to enable automated parameter optimization with gradient-free black-box algorithms. It demonstrates the approach on three examples of increasing complexity: fitting Lennard-Jones parameters for liquid argon to experimental density data, parameterizing a polarizable flexible potential for small H2O clusters against DFT structures, and tuning a small subset of parameters in a residue-level coarse-grained protein force field to reproduce experimental critical solution temperatures for wild-type and mutant hnRNPA1 sequences.

Significance. If the framework's concurrency and objective-composition features prove effective beyond the presented cases, it would address a practical bottleneck in computational chemistry by enabling reproducible, optimizer-agnostic fitting of multiscale models. The demonstrations illustrate applicability across scales from simple liquids to biomolecular systems, and the open provision of such a tool could promote standardization in parameterization workflows.

major comments (2)

[Abstract] Abstract: The central claim that ChemFit enables 'scalable' fitting for 'high-dimensional' multiscale models is not supported by the demonstrations. The three examples optimize only low-dimensional subsets (~2 parameters for the Lennard-Jones case, a limited set for water clusters, and a 'small subset' for the protein force field), providing no evidence that the framework mitigates the exponential growth in evaluations or local-minima issues typical of black-box optimization in high dimensions.
[Demonstrations] Demonstrations section: The scalability assertion requires explicit testing with larger numbers of free parameters. Without such cases, the paper extrapolates the utility of its concurrency and composition features from untested regimes, weakening the support for the broad applicability claim.

minor comments (2)

The manuscript would benefit from reporting the exact number of free parameters, objective-function evaluations, and convergence metrics for each demonstration to allow readers to assess computational cost and optimizer performance.
Clarify how heterogeneous simulation outputs are normalized or weighted when composing the objective functions, as this is central to the framework's claimed flexibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We agree that the original claims regarding scalability for high-dimensional models were not fully supported by the demonstrations and have revised the manuscript to align the language more closely with the presented evidence while preserving the framework's intended utility.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim that ChemFit enables 'scalable' fitting for 'high-dimensional' multiscale models is not supported by the demonstrations. The three examples optimize only low-dimensional subsets (~2 parameters for the Lennard-Jones case, a limited set for water clusters, and a 'small subset' for the protein force field), providing no evidence that the framework mitigates the exponential growth in evaluations or local-minima issues typical of black-box optimization in high dimensions.

Authors: We agree that the demonstrations involve only modest numbers of parameters and do not test high-dimensional regimes or directly address the curse of dimensionality. The framework itself does not mitigate inherent challenges of black-box optimization such as exponential evaluation growth or local minima; those depend on the chosen optimizer. The concurrency and composition features primarily target the practical bottleneck of evaluating expensive, heterogeneous simulation-based objectives in parallel. In the revised manuscript we have removed the phrase 'high-dimensional' from the abstract and replaced 'scalable' with 'efficient' to better reflect the evidence. We have also added a limitations paragraph noting that explicit high-dimensional benchmarks remain future work. revision: yes
Referee: [Demonstrations] Demonstrations section: The scalability assertion requires explicit testing with larger numbers of free parameters. Without such cases, the paper extrapolates the utility of its concurrency and composition features from untested regimes, weakening the support for the broad applicability claim.

Authors: We concur that the demonstrations do not include explicit tests with large numbers of free parameters, so claims of scalability in dimensionality are not empirically supported. The examples were selected to illustrate applicability across chemical scales and the handling of composed objectives rather than to benchmark high-dimensional performance. We have revised the demonstrations and discussion sections to remove broad scalability assertions and instead emphasize that the concurrent evaluation and objective-composition capabilities facilitate reproducible fitting workflows for multiscale models. This revision ensures all claims are directly supported by the reported results. revision: yes

Circularity Check

0 steps flagged

No circularity: framework definition and example applications remain independent.

full rationale

The paper introduces ChemFit as a Python framework for defining, composing, and concurrently evaluating simulation-based objective functions to pair with black-box optimizers. It then applies the framework to three separate fitting tasks that draw on external data sources (experimental densities for LJ argon, DFT cluster structures for water, and experimental critical temperatures for the protein). No equation, definition, or claim reduces the framework's stated capabilities to a redefinition of its own inputs or to a self-citation chain; the demonstrations function as external validation rather than tautological outputs. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework claim rests on the domain assumption that black-box optimizers can be applied to composed simulation objectives without additional theoretical guarantees; no new physical entities or fitted constants are introduced by the paper itself.

axioms (1)

domain assumption Black-box optimization algorithms can locate useful minima for the composed objective functions arising from molecular simulations
Invoked implicitly in the design and in the choice of the three demonstration cases

pith-pipeline@v0.9.0 · 5607 in / 1156 out tokens · 61066 ms · 2026-05-15T12:03:42.319158+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

ChemFit splits the computation of the loss value into two steps: (1) The computation of intermediate quantities via explicit simulations... (2) The loss is computed by applying a function that maps from the quantities... Parameters → Quantities → Loss Value.
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We demonstrate the versatility of ChemFit for different applications such as: (i) determination of Lennard-Jones parameters for liquid Argon... (ii) the parameterization of a polarizable force-field for H2O...

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.