Some conjugated polymers, the plastic semiconductors used in bioelectronics and electrochromic windows, absorb ions and water when they are electrochemically charged. Most swell gradually. One, PB2T-TEG, does something stranger. Over a 0.1 V window its layered crystal jumps from 14.3 Å to 18.1 Å layer spacing, the swollen and unswollen phases coexist at one voltage, and the boundary of the ion-filled region stays sharp to within a micrometre for months (Bischak et al.). The proposed explanation is that the polymer's interlocked glycol side chains "unzip" once enough ions and water force their way in. Nobody has simulated this. This note proposes the first atomistic test of the unzipping hypothesis. We would build molecular dynamics simulations that treat every water molecule and ion explicitly, driven by a machine-learned force field. We would then ask whether the jump appears, what sets its sharpness, and whether it depends on how well ordered the starting crystal is. That last property is controlled experimentally by thermal annealing through liquid-crystalline phases.
The two research directions play different roles here. Explicit solvent modeling supplies the mechanism and the tools. It has shown that discrete water molecules, rather than a smooth dielectric, decide barriers and stability at charged interfaces. Liquid-crystalline templating supplies the control variable: annealing a polymer through fluid but ordered intermediate phases (phases with orientational or layered order but liquid-like mobility) sets how well its side chains interdigitate. The claim is that the first-order character of the swelling is a cooperative property of ordered, interdigitated lamellae plus explicit water and ion uptake. It should therefore weaken or disappear when the crystal is disordered or when water is replaced by a continuum.
Core Claim
Three predictions follow from the claim, and each can be checked against the experimental record. First, the free energy as a function of layer spacing should be double-welled at intermediate charging for PB2T-TEG and single-welled for the control polymer P3MEEMT. Bischak et al. measured continuous, Fickian swelling for that control, with its ion front blurring beyond 700 µm within 38 hours. Second, the barrier between the two wells should be dominated by the cost of reorganizing water and ion hydration shells, not by side-chain torsions alone. This is the same kind of hidden term that raised a CO2 activation barrier on gold from 0.36 eV in a static calculation to 0.61 eV with dynamic water (Qin et al.). It also echoes the finding that ion desolvation, rather than electron transfer, limits multivalent metal plating (Drews et al.). Third, starting from poorly ordered lamellae, the kind found in as-cast films, should smear the transition into a continuous one. Annealed lamellae should keep it sharp.
Research Program
Systems. We would simulate a periodic PB2T-TEG crystallite of roughly eight chains of six repeat units in contact with an explicit aqueous electrolyte reservoir. The control is P3MEEMT built the same way. Electrochemical oxidation is represented by fixing the number of holes per repeat unit and adding compensating anions. This follows the strategy Qin et al. used to dial surface charge on Au(110) by changing the explicit K+ count. We therefore compare doping-level windows to the experiment rather than claiming an absolute potential scale.
Force field. The simulations need a machine-learned interatomic potential, a neural network fitted to quantum-mechanical energies and forces so it runs far faster than density functional theory (DFT). We would use MACE, an equivariant message-passing network that reaches DFT-level accuracy with two layers and low latency (Batatia et al.; ACEsuit/mace). It has already run nanosecond explicit-water electrode simulations at a reported 100,000-fold speedup over DFT (Tian et al.).
Training. We would start from the MACE-OFF organic foundation model and fine-tune it by active learning on DFT data for charged oligomer–water–ion fragments. Du et al. showed that out-of-the-box foundation potentials fail on unseen interfaces and that this kind of fine-tuning brings MACE below 4 meV/atom (Du et al.).
Known pitfalls. Because the oxidized lattice is charged, finite-cutoff models are a known weakness (Mikkelsen et al.), so the MACE-Polar long-range electrostatics variant is part of the design. Generalized-gradient functionals tend to over-structure water, a problem that forced Mikkelsen et al. to simulate at 400 K. We would therefore validate bulk water density and diffusion before any production run and report results at both 300 K and a corrected temperature.
Sampling. We would use OPES, an enhanced-sampling method that learns a bias on the fly to push the system across rare transitions (Invernizzi). It would bias two collective variables, the lamellar spacing and the number of water molecules inside the side-chain region. From these runs we would reconstruct the free energy surface at each doping level.
Starting structures. Each system gets two kinds of starting structure. "As-cast" lamellae come from fast quenching from a melt. "Templated" lamellae come from slow cooling through the mesophase temperature range, mimicking the annealing that raised lamellar coherence lengths by 280–400% in PAcDOT films (Rinehart et al.). We quantify interdigitation and coherence directly.
Secondary question. The same trajectories also test something raised by Sabury et al. and Makala et al.. Both attribute backbone planarity to non-covalent sulfur–oxygen "locks" evaluated only in vacuum. PB2T-TEG has the same alkoxy-phenylene–thiophene motif, so we measure backbone dihedral distributions in the swollen phase. The question is whether invading water hydrogen-bonds the ether oxygens and loosens these locks.
Baselines
Mean-field reference. The naive reference point is a Flory–Huggins-type swelling model, a classic lattice theory of polymer–solvent mixing, extended with ion uptake and fitted to the experimental charge–swelling data. Mean-field gel theories can produce discontinuous volume transitions, so this is not a strawman. It will be fit on PB2T-TEG and then asked, without refitting, to predict the 14.3 and 18.1 Å end states and the P3MEEMT control. The simulation beats it if it predicts both end-state spacings and the first-order versus continuous distinction from molecular inputs alone. A fitted mean-field model can match an observed jump but cannot say why one side chain gives a jump and another does not.
Strongest competitor. The strongest existing method is classical all-atom molecular dynamics with standard fixed-charge force fields and explicit water. It is the workhorse for swollen polymers and far cheaper than a learned potential. It gets the identical protocol: same cells, starting structures, OPES biases, simulated time, and temperature. Its parameters receive a matched effort, consisting of refitting backbone torsions to the same DFT dihedral data. No reported result found on Lacuna for classical or learned-potential simulation of PB2T-TEG swelling, so the targets are experimental. These are the neutral spacing of 14.3 Å, the swollen spacing of 18.1 Å (a 27% expansion), two-phase coexistence, and continuous swelling for P3MEEMT, all from Bischak et al.
Off-the-shelf learned potential. Untuned MACE-OFF tests whether fine-tuning matters. We expect it to fail, given the out-of-distribution degradation documented by Du et al.
Ablations. Three ablations isolate the key ingredients. The first replaces explicit water with a continuum dielectric while keeping explicit ions and polymer, testing the solvent claim. The second starts from as-cast instead of templated lamellae, testing the ordering claim. The third removes long-range electrostatics, testing the charged-lattice treatment.
What counts as success. The proposal counts as successful if the full model reproduces both spacings within 0.5 Å, shows a double-well free energy for PB2T-TEG and a single well for P3MEEMT, and loses that bimodality in the continuum-water ablation.
Why not just…?
Why not just use classical force fields with explicit water? This is the real threat. Our design treats it as a hypothesis to test rather than dismiss. Makala et al. stress that planar versus helical outcomes in these backbones hinge on subtle sulfur–oxygen energetics that empirical force fields describe poorly. Roncoroni et al. report that empirical mixing rules miss interface polarization (Roncoroni et al.). Neither result concerns this polymer. If the classical baseline reproduces the jump, the control, and the spacings, the conclusion is that unzipping is a packing-and-hydration effect that cheap models capture, and that conclusion is itself publishable. The learned potential is justified only if the classical model misplaces the transition or erases bimodality.
Why not just use an implicit solvent? In the systems studied so far, continuum solvation misses the very effects at stake. It missed potential-driven water reorientation and CO2 chemisorption at silver (Tian et al.). It missed water acting as a proton shuttle (Polino). It also cannot count discrete coordination shells, which Drews et al. found set a 92 versus 74 kJ/mol desolvation difference between two glyme solvents. Here, water counted inside the side-chain region is a collective variable, which a continuum cannot represent. The continuum ablation measures this directly.
Why not just optimize annealing experimentally? Experiments can find good protocols but did not anticipate failures such as the helical trapping of TPT-2T (Makala et al.). Experiments also cannot see the sub-nanometre unzipping. Our simulations would instead yield a falsifiable prediction for experimentalists: as-cast films should show broader coexistence and blurrier ion fronts than annealed ones. If in operando X-ray scattering shows no dependence on annealing, the ordering half of the claim fails.
Decisive Checks
The first check is quantitative agreement with the end-state spacings and with the PB2T-TEG versus P3MEEMT contrast, achieved without fitting to swelling data. The second check is that bimodality survives in the full model but vanishes under continuum water. The third check is a free-energy decomposition showing whether the barrier is mainly ion and water reorganization or side-chain disentanglement. Finally, the model must predict that as-cast versus templated starting order changes the transition's width. That prediction can then be tested by scattering measurements on films annealed differently. If the classical baseline passes all four checks, we report that and drop the learned potential.
References
- A Reversible Structural Phase Transition by Electrochemical Ion Injection into a Conjugated Polymer
- CO2 activation at Au(110)–water interfaces: An ab initio molecular dynamics study
- Modeling of Electron‐Transfer Kinetics in Magnesium Electrolytes: Influence of the Solvent on the Battery Performance
- MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields
- Electrochemical potential-driven water dynamics control CO2 electroreduction at the Ag/H2O interface
- Accelerating and Enhancing Thermodynamic Simulations of Electrochemical Interfaces
- Structure and energetics of liquid water–hydroxyl layers on Pt(111)
- OPES: On-the-fly Probability Enhanced Sampling Method
- Impact of Side-Chain Polarity and Symmetry on the Structure and Properties of Acyclic Dioxythiophene Polymers
- Non-covalent planarizing interactions yield highly ordered and thermotropic liquid crystalline conjugated polymers
- Non-Covalent Interactions and Helical Packing in Thiophene-Phenylene Copolymers: Tuning Solid-State Ordering and Charge Transport for Organic Field-Effect Transistors
- Deriving effective electrode–ion interactions from free-energy profiles at electrochemical interfaces
- Kinetics of Aqueous Media Reactions via Ab Initio Enhanced Molecular Dynamics: The Case of Urea Decomposition