C-MORAL: Controllable Multi-Objective Molecular Optimization with Reinforcement Alignment for LLMs

Morteza Ziyadi; Rui Gao; Swastik Roy; Xiang 'Anthony' Chen; Youngseung Jeon

arxiv: 2604.23061 · v2 · pith:U5ZII53Pnew · submitted 2026-04-24 · 💻 cs.LG · cs.AI

C-MORAL: Controllable Multi-Objective Molecular Optimization with Reinforcement Alignment for LLMs

Rui Gao , Youngseung Jeon , Swastik Roy , Morteza Ziyadi , Xiang 'Anthony' Chen This is my paper

Pith reviewed 2026-07-04 16:04 UTC · model glm-5.2

classification 💻 cs.LG cs.AI

keywords c-moralmolecularoptimizationachievesllmstasksacrossalignment

0 comments

The pith

Dry-transferred Bragg mirrors achieve strong coupling in monolayer WS2

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

The paper introduces a deterministic dry-transfer fabrication method for complete SiO2/TiO2 dielectric microcavities built around monolayer WS2. By fabricating both top and bottom Bragg mirrors on PVA-coated substrates, detaching them with PDMS, and stacking them with the 2D material sandwiched between, the authors avoid the energetic particle bombardment and material incompatibility problems of conventional top-mirror deposition. The resulting cavities reach a quality factor of roughly 4,000 and exhibit strong exciton-photon coupling with a Rabi splitting of approximately 36 meV at room temperature, persisting down to cryogenic temperatures, with the sample remaining stable over eight months of repeated thermal cycling.

Core claim

The central discovery is that transferable SiO2/TiO2 distributed Bragg reflectors, assembled via the same viscoelastic dry-transfer process used for 2D materials themselves, can form complete, robust optical microcavities around atomically thin semiconductors without degrading the emitter. The method yields cavities that preserve the optical quality of monolayer WS2 and demonstrate strong light-matter coupling (Rabi splitting ~36 meV) at both room and cryogenic temperatures.

What carries the argument

The key mechanism is the PVA-mediated release and PDMS-mediated pickup of pre-fabricated SiO2/TiO2 Bragg mirror stacks, enabling deterministic placement of micron-scale DBR fragments above and below a monolayer semiconductor. The strong-coupling signature is extracted by fitting the angle-resolved photoluminescence and reflection spectra to a two-level coupled oscillator Hamiltonian, whose eigenvalues yield the lower and upper polariton branch energies and the coupling strength g.

Load-bearing premise

The extraction of coupling strength relies on fitting a coupled oscillator model to spectra in which the upper polariton branch is not distinctly resolved and a bare cavity resonance persists alongside the polariton branches, meaning the strong-coupling evidence rests partly on the behavior of the lower polariton branch alone.

What would settle it

If the observed lower polariton branch shift can be fully explained by a weak-coupling cavity-exciton interaction (e.g., a dispersive shift or absorption-induced mode pulling) rather than genuine anti-crossing, then the claim of strong coupling would not hold. A definitive observation of both polariton branches anti-crossing as detuning is tuned, or a coupling strength g exceeding the combined linewidth threshold, would be needed to rule this out.

Figures

Figures reproduced from arXiv: 2604.23061 by Morteza Ziyadi, Rui Gao, Swastik Roy, Xiang 'Anthony' Chen, Youngseung Jeon.

**Figure 1.** Figure 1: Overview of C-MORAL generation, training pipeline view at source ↗

**Figure 2.** Figure 2: Ablation study of reward aggregation on the HLMPQ task using view at source ↗

**Figure 4.** Figure 4: Contour plots of aggregation functions on 2- view at source ↗

**Figure 5.** Figure 5: An example of the highly structured prompt template used in view at source ↗

**Figure 6.** Figure 6: Optimization of different MISTRAL-based models on the BPQ task. The group-relative advantage is defined as A GRPO i,j = r GRPO i,j − µi σi + ϵgrp , where ϵgrp is a small constant for numerical stability. For token t in response yi,j , the importance ratio is ρi,j,t(Θ) = πΘ(yi,j,t | xi , yi,j,<t) πΘold(yi,j,t | xi , yi,j,<t) . The GRPO objective is LGRPO(Θ) = 1 B X B i=1 1 G X G j=1 1 |yi,j | | X yi,j | t=… view at source ↗

read the original abstract

Large language models (LLMs) show promise for molecular optimization, but aligning them with selective and competing drug-design constraints remains challenging. We propose C-Moral, a reinforcement learning post-training framework for controllable multi-objective molecular optimization. C-Moral combines group-based relative optimization, property score alignment for heterogeneous objectives, and bottleneck-sensitive non-linear reward aggregation to improve stability across competing molecular properties. Experiments on C-MuMOInstruct and S$^2$-Bench MolOpt show that C-Moral achieves the best performance among compared methods on both benchmarks. On C-MuMOInstruct, C-Moral achieves the best Success Optimized Rate (SOR) of 48.9\% on in-domain tasks and 39.5\% on out-of-domain tasks while preserving scaffold similarity. On S$^2$-Bench MolOpt, it also achieves the strongest results across LogP, MR, and QED optimization tasks. These results suggest that C-Moral is an effective way to align molecular LLMs with continuous and constrained molecular design objectives. Our code and models are publicly available at https://github.com/Rwigie/C-MORAL.

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

The reader's review is about the wrong paper — they reviewed the optics paper (arXiv:2604.23062), not the C-MORAL molecular optimization paper (arXiv:2604.23061). The full text provided is also from the optics paper, not C-MORAL.

read the letter

There's a fundamental mismatch here. The assigned paper is C-MORAL (RL-based molecular optimization with LLMs), but both the reader's review and the provided full text are from a completely different paper — a physics.optics paper about transferable dielectric mirrors for strong light-matter coupling (arXiv:2604.23062). The reader's verdict, claims, and rationale all pertain to SiO2/TiO2 Bragg mirrors, WS2 monolayers, and Rabi splitting. None of it touches C-MORAL's actual claims about group-based relative optimization, bottleneck-sensitive reward aggregation, or SOR benchmarks on C-MuMOInstruct. The stress-test flag is correct on this point. I cannot evaluate the C-MORAL paper because I don't have its text. The reader's analysis cannot be applied to the paper it was supposed to review — it's simply about a different manuscript. The optics paper itself (what was actually provided) is a reasonable incremental experimental contribution: a PVA-mediated dry transfer for SiO2/TiO2 DBRs around monolayer WS2, with room-temperature strong coupling (~36 meV Rabi splitting) and Q factors around 10^3. The reader's assessment of that paper is fair — the indistinct UPB and persistent bare cavity mode are legitimate concerns, though the LPB anti-crossing and coupled-oscillator fit satisfying the strong coupling criterion are adequate evidence. But that assessment is irrelevant to the assigned paper. Recommendation: do not send this to a referee until the correct paper text is provided. The reader needs to re-review arXiv:2604.23061, or the assignment needs to be corrected.

Referee Report

1 major / 1 minor

Summary. The full text provided for review is the manuscript 'Deterministic Transferable Planar Dielectric Mirrors for Investigating Strong Light–Matter Coupling' (arXiv:2604.23062v1, physics.optics) by Patra et al. This paper presents a dry-transfer method for fabricating SiO2/TiO2 distributed Bragg reflector (DBR) microcavities around monolayer WS2, achieving a Q factor of ~946 (experimental) and demonstrating strong exciton–photon coupling with a Rabi splitting of ~36 meV at room temperature, maintained down to 175 K. The central claim is that this deterministic transfer approach preserves emitter integrity while enabling compact, material-efficient polaritonic devices. NOTE: The abstract and title supplied in the assignment correspond to a different paper ('C-MORAL: Controllable Multi-Objective Molecular Optimization with Reinforcement Alignment for LLMs', arXiv:2604.23061, cs.LG). The reader's report also reviews the optics paper. This referee report addresses the optics manuscript (arXiv:2604.23062v1), which is the full text actually provided.

Significance. The fabrication approach is practical and addresses a real materials-integration problem. The use of transferable DBRs matched to the flake footprint, combined with standard 2D-material viscoelastic pick-up, is a sensible contribution to the polaritonics toolkit. The authors provide TMM simulations using independently measured refractive indices, and the temperature-dependent coupling-strength data (Fig. 4b) constitute a falsifiable experimental observation. The reported Rabi splitting (~36 meV) is within the expected range for monolayer WS2 in dielectric cavities, and the robustness over eight months and multiple thermal cycles is a useful practical result. The claim is modest and proportionate to the evidence presented.

major comments (1)

The full text provided is for arXiv:2604.23062v1 (Patra et al., physics.optics), not arXiv:2604.23061 (C-MORAL, cs.LG). The abstract, title, and assignment metadata correspond to the C-MORAL paper, but the manuscript body is the optics paper. This referee report assesses the optics manuscript as provided. The editor should confirm which paper is intended for review and ensure the correct manuscript is assigned before a substantive review can proceed.

minor comments (1)

The reader's report and stress-test note both correctly identify the paper mismatch. The reader's detailed analysis of the optics paper (strong coupling, UPB visibility, bare cavity mode) is substantive and aligns with the manuscript content. The editor should route the reader's report to the correct paper assignment.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and the positive assessment of the fabrication approach and experimental results. We address the sole major comment below.

read point-by-point responses

Referee: The full text provided is for arXiv:2604.23062v1 (Patra et al., physics.optics), not arXiv:2604.23061 (C-MORAL, cs.LG). The abstract, title, and assignment metadata correspond to the C-MORAL paper, but the manuscript body is the optics paper. This referee report assesses the optics manuscript as provided. The editor should confirm which paper is intended for review and ensure the correct manuscript is assigned before a substantive review can proceed.

Authors: The referee is correct that there is a mismatch between the assignment metadata (which references arXiv:2604.23061, the C-MORAL cs.LG paper) and the full manuscript text provided for review (arXiv:2604.23062v1, our optics paper on transferable dielectric mirrors). This is an administrative error in the review assignment, not a problem with the manuscript itself. The manuscript we submitted and the full text the referee reviewed—'Deterministic Transferable Planar Dielectric Mirrors for Investigating Strong Light–Matter Coupling' (arXiv:2604.23062v1)—are one and the same, and the referee has correctly identified and reviewed our paper. We are grateful that the referee proceeded to assess the optics manuscript despite the metadata discrepancy and provided a substantive and fair evaluation. We concur with the recommendation that the editor confirm the correct manuscript assignment. No revision to our manuscript is needed on this point; the issue lies entirely with the review system's metadata. revision: no

Circularity Check

0 steps flagged

No circularity found; the paper is a self-contained experimental demonstration with standard fitting to a coupled oscillator model.

full rationale

The paper is primarily experimental, demonstrating a dry-transfer fabrication method for SiO2/TiO2 Bragg mirror microcavities around monolayer WS2. The central quantitative claim—Rabi splitting of ~36 meV and strong exciton-photon coupling—is extracted by fitting angle-resolved PL and reflection spectra to a standard coupled oscillator model (Eq. 1). This is a conventional data-fitting procedure, not a circular derivation: the fit parameters (coupling strength g, detuning δ) are determined from experimental spectral positions of the polariton branches, and the strong-coupling criterion (Eq. 2) is then checked against independently measured linewidths (γ_c, γ_x). No prediction reduces to a fitted input by construction. The TMM simulations use independently measured refractive indices (ellipsometry for SiO2/TiO2, literature for WS2). Self-citation to ref 13 (Federolf et al.) is for prior sputtered-cavity work by overlapping authors, but it is not load-bearing for the present paper's central claim—the strong-coupling observation stands on its own experimental data and standard analysis. The reader's take references a different arXiv number (2604.23061, a CS/ML paper) versus the actual full text provided (2604.23062, an optics paper), but this metadata mismatch does not introduce circularity into the physics derivation itself.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities, particles, or forces. It uses standard optical physics (Bragg mirrors, Fabry-Perot cavities, exciton-polaritons) and standard materials (SiO2, TiO2, WS2, PVA, PDMS). The free parameters are fitted from experimental data using standard models.

free parameters (3)

Coupling strength g = 17.9 meV (RT), 22.9 meV (175K)
Fitted from experimental polariton branch dispersion using the coupled oscillator model (Eq. 1).
Effective refractive index n_eff = 1.65
Determined from the cavity dispersion fitting, stated as consistent with SiO2/TiO2 indices.
Number of DBR pairs = 8.5 (visible), 10.5 (near-IR)
Design choice to achieve target reflectivity; not fitted but selected for the cavity mode.

axioms (4)

standard math Coupled oscillator model (2x2 Hamiltonian) adequately describes the exciton-photon interaction
Eq. 1 is a standard model in polaritonics (citing Deng et al. 2010, ref 32). Used to extract coupling strength.
domain assumption Strong coupling criterion g >= (γ_c + γ_x)/4
Eq. 2, standard criterion (citing Keeling & Kena-Cohen 2020, ref 33). Applied to justify strong coupling regime.
domain assumption PVA residue does not significantly affect optical measurements in visible/IR
Stated in fabrication section citing ref 27. Justifies the use of PVA as a release layer despite potential residue.
domain assumption Refractive index of WS2 from literature is accurate for the simulation
TMM simulation uses n(WS2) from ref 30 (Ermolaev et al. 2020). Affects theoretical Q factor and field distribution calculations.

pith-pipeline@v1.1.0-glm · 12597 in / 2408 out tokens · 218176 ms · 2026-07-04T16:04:39.672654+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages · 1 internal anchor

[1]

Using a WS2 monolayer as the active medium, clear signatures of strong exciton-photon coupling are observed at both room temperature and cryogenic temperatures. These results demonstrate an efficient cavity fabrication approach that preserves the integrity of the emitter of layered materials, enabling next generation integrated photonic devices. a)These a...

work page internal anchor Pith review Pith/arXiv arXiv 2026
[2]

The cavity Q factor can be further 5 enhanced by increasing the number of mirror pairs

The theoretical Q factor of the cavity is 2076, as obtained from the transfer matrix (TMM) calculation. The cavity Q factor can be further 5 enhanced by increasing the number of mirror pairs. FIG

work page 2076
[3]

Exciton binding energy and nonhydrogenic rydberg series in mono- layer WS2,

is satisfied at room temperature. At 175 K, coupling strength is found to be 22.9 meV . As the temperature decreases, the coupling strength increases, as shown in Fig. 4(b). This observed temperature dependence can be attributed to the enhancement of the oscillator strength34 along with a reduction in exciton dephasing and diminished non-radiative contrib...

work page 2014

[1] [1]

Using a WS2 monolayer as the active medium, clear signatures of strong exciton-photon coupling are observed at both room temperature and cryogenic temperatures. These results demonstrate an efficient cavity fabrication approach that preserves the integrity of the emitter of layered materials, enabling next generation integrated photonic devices. a)These a...

work page internal anchor Pith review Pith/arXiv arXiv 2026

[2] [2]

The cavity Q factor can be further 5 enhanced by increasing the number of mirror pairs

The theoretical Q factor of the cavity is 2076, as obtained from the transfer matrix (TMM) calculation. The cavity Q factor can be further 5 enhanced by increasing the number of mirror pairs. FIG

work page 2076

[3] [3]

Exciton binding energy and nonhydrogenic rydberg series in mono- layer WS2,

is satisfied at room temperature. At 175 K, coupling strength is found to be 22.9 meV . As the temperature decreases, the coupling strength increases, as shown in Fig. 4(b). This observed temperature dependence can be attributed to the enhancement of the oscillator strength34 along with a reduction in exciton dephasing and diminished non-radiative contrib...

work page 2014