REVIEW
PTagent: An LLM-assisted tool for finite-temperature model construction in cosmological phase transitions
High Energy Physics - Phenomenology · Submitted 15 Oct 2026 · 47 references
Pith's verdict: A genuinely useful tool paper: the contract-plus-backend separation is a real idea, but the interpretation layer's error rate is unmeasured, and that is exactly where the reproducibility claim needs support.
T0 · AI review
Why Pith says this
Highlight any text to comment on it.
Referee report
The paper
The paper claims
PTagent claims finite-temperature model construction becomes inspectable and reproducible when an LLM agent is confined to building a reviewable physical model contract and deterministic backends generate numerical potentials only from…
The reading
What carries the argument
The load-bearing object is the model contract: a human-readable physical model specification that records the field basis and parameters, particle spectra and multiplicities, thermal treatment, renormalization and resummation prescriptions, and phase symmetries, together with the provenance of every expression. The mechanism is the enforced separation of duties -- the LLM agent interprets sources, organizes the physical ingredients, and flags unresolved choices, while the backends implement only the approved specification and refuse to run while essential information or prescriptions are unresolved. The contract is what makes the pipeline inspectable: each choice carries a source, the provenance graph distinguishes imported, derived, and user-confirmed information, and recorded hashes tie each generated implementation back to the approved specification.
Extended reading notes
Core claim
PTagent's claim is that the fragile part of finite-temperature model construction -- reading scattered sources, assembling field-dependent masses and thermal corrections, and fixing gauge, renormalization, resummation, and particle-content conventions -- can be delegated to an LLM agent without giving up control, provided the agent produces a physical model specification rather than code. The contract records model expressions, their sources, numerical inputs, and adopted prescriptions; physical choices left unresolved by the sources are returned to the user with their implications explained, and backend generation is blocked until they are resolved. Only after approval do the dedicated backends compile the specification into CosmoTransitions-style and PhaseTracer-style numerical potentials, following the contract without adding physical assumptions, with hashes tying each implementation to the approved specification. The paper asserts that for fixed model inputs, physical prescriptions, and backend version, the resulting numerical potential is reproducible, and reports consistent effective potentials across three different LLM models. Validation operates at several levels: dimensional and spectrum consistency checks, an independent check of derived scalar thermal self-energies against the Hessian of the leading $T^2$ thermal contribution, and pointwise comparisons between formats and against independently supplied EFT expressions.
Load-bearing premise
Everything downstream assumes that the LLM agent reads sources correctly, correctly spots which physical choices are genuinely unresolved, and that human review catches whatever the agent misses -- if a wrong convention slips through unflagged, the approved contract and the faithfully compiled potential are wrong together, and the paper's validation checks would not detect it because they compare PTagent output against PTagent output.
Editorial extensions
If this is right
- A researcher can hand PTagent a SARAH model directory, an arXiv paper, or a DRalgo file and receive an approved, audited finite-temperature potential ready for phase-structure, tunnelling, and gravitational-wave calculations.
- Every physical convention -- gauge, resummation scheme, particle content, renormalization scale -- is recorded in the contract, so downstream results can be traced to explicit choices rather than implicit ones.
- Because backends generate code only from the approved contract and block while essential physical information or prescriptions remain unresolved, the pipeline refuses to silently default on ambiguous physics.
- Recorded hashes link generated implementations to the approved specification, so editing the contract flags outdated implementations.
- Both output formats are built from the same approved prescription, and their numerical consistency is verified pointwise at twenty field-temperature points with normalized differences below $3\times10^{-8}$.
Reading between the lines
- The paper does not quantify an error rate for the interpretation layer; a planted-error benchmark would supply it. Because the reported checks compare PTagent outputs against PTagent outputs, a wrong convention that the agent silently adopts would propagate through the whole chain undetected -- and the xSM example shows the agent does not always volunteer checks, since the Hessian inconsistency wa
- A standardized contract could become a provenance record for comparing phenomenology claims: two groups running the same model could first compare contracts and decide whether a difference in predictions is physics or convention.
- The same contract-mediated split -- fallible interpreter, deterministic compiler, human approval gate -- could plausibly be applied to other convention-heavy stages of high-energy physics pipelines wherever the danger is silent adoption of an implicit convention.
Original abstract
Cosmological phase-transition calculations require well-defined finite-temperature numerical models, whose construction involves assembling model information under consistent theoretical prescriptions. In this work we present \texttt{PTagent}, a large language model (LLM) agent-based framework for constructing such models from SARAH model files, research papers, and DRalgo sources. The LLM agent interprets and organizes model information and identifies unresolved physical choices, while dedicated backends perform the model-dependent calculations. After review and approval of the resulting physical model specification, the backends generate the corresponding CosmoTransitions-style or PhaseTracer-style numerical potentials. By separating model interpretation from numerical implementation, PTagent provides an inspectable and reproducible approach to finite-temperature model construction for cosmological phase-transition studies.
The compiled record
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
The abstract, sentence by sentence
Cosmological phase-transition calculations require well-defined finite-temperature numerical models, whose construction involves assembling model information under consistent theoretical prescriptions. In this work we present \texttt{PTagent}, a large language model (LLM) agent-based framework for constructing such models from SARAH model files, research papers, and DRalgo sources. The LLM agent interprets and organizes model information and identifies unresolved physical choices, while dedicated backends perform the model-dependent calculations. After review and approval of the resulting physical model specification, the backends generate the corresponding CosmoTransitions-style or PhaseTracer-style numerical potentials. By separating model interpretation from numerical implementation, PTagent provides an inspectable and reproducible approach to finite-temperature model construction for cosmological phase-transition studies.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Cosmological phase-transition calculations require well-defined finite-temperature numerical models, whose construction involves assembling model information under consistent theoretical prescriptions. | background 1.00 | 0.05 | 0.04 | asserted 1.00 |
| s1 | In this work we present \texttt{PTagent}, a large language model (LLM) agent-based framework for constructing such models from SARAH model files, research papers, and DRalgo sources. | contribution 0.99 | 0.91 | 0.13 | asserted 1.00 |
| s2 | The LLM agent interprets and organizes model information and identifies unresolved physical choices, while dedicated backends perform the model-dependent calculations. | method 0.99 | 0.57 | 0.09 | asserted 1.00 |
| s3 | After review and approval of the resulting physical model specification, the backends generate the corresponding CosmoTransitions-style or PhaseTracer-style numerical potentials. | method 0.99 | 0.56 | 0.12 | asserted 1.00 |
| s4 | By separating model interpretation from numerical implementation, PTagent provides an inspectable and reproducible approach to finite-temperature model construction for cosmological phase-transition studies. | interpretation 0.70 | 0.72 | 0.72 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
Figures
Figures reproduced from arXiv:2610.21118 by the authors.
References
Discussion
Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.