At gaps below a few tens of nanometres, two surfaces separated by a film feel a force caused by fluctuating electromagnetic fields. This is the Casimir–Lifshitz or van der Waals force. Whether it pulls the surfaces together or pushes them apart depends on how polarizable the film is compared with the two surfaces, and that comparison runs over a wide range of frequencies. Conjugated polymers that pass through liquid-crystalline phases, meaning mobile phases in which the long molecules share a common direction, are extremely anisotropic. Along the backbone they are far more polarizable than across it. Annealing through these phases controls which way the backbones point relative to the substrate. Our claim is that, for a suitably chosen substrate, the backbone orientation alone decides the sign of the van der Waals interaction across the film. If so, the thermal history that templates the film also writes in whether the film is held on stably or dewets spontaneously.
The proposal has two parts. On the computational side, we build a chain from density functional theory (DFT, the standard quantum-mechanical method for computing electron behaviour from the electron density) to the full anisotropic dielectric tensor at imaginary frequencies, $\epsilon_{ij}(i\xi)$, averaged over molecular conformations sampled in each liquid-crystalline phase. On the experimental side, we test the predicted sign flip with a classical, quantitative readout: the wavelength of spontaneous dewetting in ultrathin films. Nothing here requires exotic boundary conditions. The tunability comes from a real material property that annealing already controls.
Core Claim
For a uniaxial film between a substrate and air, the strength and sign of the interaction are set by the Hamaker constant $A_{s/f/a}$. Its leading term has the form
$$A_{s/f/a} \propto \sum_{n} \Big[\bar\epsilon_s(i\xi_n)-\bar\epsilon_f(i\xi_n)\Big]\Big[\epsilon_a(i\xi_n)-\bar\epsilon_f(i\xi_n)\Big],$$
where $\xi_n$ are the Matsubara frequencies (a discrete set of imaginary frequencies fixed by temperature), $\epsilon_s$ and $\epsilon_a=1$ are the substrate and air permittivities, and $\bar\epsilon_f$ is an orientation-dependent effective film permittivity. For a uniaxial film, $\bar\epsilon_f$ is built from the in-plane component $\epsilon_\parallel$ and out-of-plane component $\epsilon_\perp$ of the film tensor. Its value differs between face-on stacking (backbones and π-planes lying in the substrate plane) and edge-on stacking (π-planes standing up). The exact multilayer Lifshitz formula mixes these components differently for the two polarization channels, so we compute it exactly rather than use this sketch. The sketch is only meant to show the mechanism. Because $\epsilon_a<\bar\epsilon_f$ always, the film is held on (negative $A$, stable film) when $\bar\epsilon_f<\bar\epsilon_s$ and pushed toward rupture when $\bar\epsilon_f>\bar\epsilon_s$, at the frequencies that dominate. Those frequencies lie in the ultraviolet for gaps of order 10 nm.
The new claim is this: there exist accessible substrates whose $\bar\epsilon_s(i\xi)$ falls between the two orientation-dependent values of $\bar\epsilon_f$ for a polymer such as TPT-TT. For such substrates, templating to one orientation stabilizes a sub-20 nm film and templating to the other destabilizes it. A cheap orientation-blind model would predict no such effect, and the experiment can tell the two apart.
Where this sits relative to prior work
Three bodies of work motivate the idea, and none of them makes the connection.
On the force side, the review Casimir Effect in MEMS: Materials, Geometries, and Metrologies—A Review surveys geometries and materials for controlling Casimir forces. It notes that an intervening fluid can turn attraction into repulsion when the fluid's permittivity lies between those of the two walls, and it names accurate dielectric data and ab initio input as unfinished business. It never treats liquid crystals as a medium whose effective permittivity can be switched, and that is the gap we fill. Casimir stress inside planar materials gives a renormalized theory of Casimir stress inside media whose permittivity varies along one axis. It does so only for synthetic scalar profiles such as $n=\sec z$, with no real material and no tensor anisotropy. We supply a real anisotropic tensor, although our main observable is the simpler interfacial force rather than internal stress.
Analytic models of tunable sign give theoretical context but do not shape our design. Fractional-Neumann boundary conditions (Repulsive Casimir force from fractional Neumann boundary conditions) and temperature-driven sign changes in closed cavities (Finite-temperature Casimir effect in piston geometry and its classical limit) are both stated for idealized massless scalar fields. The fractional paper itself flags experimental realization as open. We deliberately avoid mapping a liquid crystal onto a "fractional boundary", because no physical microstructure is known to correspond to a given fractional order. We work instead with dispersive dielectric tensors, which are measurable.
On the materials side, Non-covalent planarizing interactions yield highly ordered and thermotropic liquid crystalline conjugated polymers supplies the model system. TPT-TT has a nematic window from 218 °C to 107 °C, a smectic A window from 107 °C to 48 °C, and a crystalline solid below 48 °C. DFT-verified S···O and S···H–C conformational locks keep it planar, and grazing-incidence X-ray diffraction shows that annealing raises structural order. That work and its relatives (Non-Covalent Interactions and Helical Packing in Thiophene-Phenylene Copolymers; Impact of Side-Chain Polarity and Symmetry on the Structure and Properties of Acyclic Dioxythiophene Polymers, which reports annealing-driven lamellar coherence gains of up to 400%) stop at structure and charge transport. None computes $\epsilon_{ij}(i\xi)$ or considers interfacial forces.
For the method, we reuse two tools. DFT is the source of electronic response. MACE (MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields, code at ACEsuit/mace) is a fast machine-learned interatomic potential that lets us sample phase-specific conformational ensembles at near-DFT accuracy.
Research Program
Conformational ensembles. We train a MACE potential on DFT energies and forces for TPT-TT oligomers and periodic stacks. Training data are dihedral scans across the conformational locks plus snapshots of disordered melts. MACE runs molecular dynamics at temperatures in the isotropic, nematic, smectic and solid windows, in slabs with face-on and edge-on starting orientations. We use MACE for configurations only, not for dielectric response.
Dielectric tensors. From sampled snapshots we compute frequency-dependent polarizability or dielectric tensors with linear-response DFT, using a hybrid functional or a scissor correction calibrated to the measured optical gap. We then continue these to imaginary frequency and average them into phase- and orientation-resolved $\epsilon_\parallel(i\xi)$ and $\epsilon_\perp(i\xi)$ up to roughly 30 eV. Substrate tensors come from tabulated optical data where reliable, and from DFT otherwise.
Screening. We compute exact anisotropic Lifshitz free energies for substrate/film/air stacks over film thicknesses of 3 to 30 nm. Candidate substrates are silica, sapphire, silicon with native oxide, fluoropolymer coatings and gold. The output is a map of $A$ versus orientation and phase. From it we pick one "sign-flip" substrate, where the two orientations give opposite signs, and one "control" substrate, where both orientations give the same sign but different magnitudes.
Experiment. Orientation is set by cooling protocol and by molecular monolayers about 1 nm thick that control surface anchoring. The monolayers are included explicitly in the Lifshitz stack, and orientation is confirmed by diffraction. Ultrathin films are then held in the nematic window, and we observe whether they stay stable or dewet.
Baselines
Do nothing. The naive reference is an orientation-blind Lifshitz calculation using the isotropic average of the film tensor, $(2\epsilon_\parallel+\epsilon_\perp)/3$. It predicts a single Hamaker constant per substrate regardless of templating. Any measured orientation dependence in the sign or magnitude of $A$ beyond its error bars beats it.
Strongest existing method. The strongest competitor is standard empirical Lifshitz practice. Spectroscopic ellipsometry measures the ordinary and extraordinary optical constants of oriented films over the instrument's range, a damped-oscillator fit with Kramers–Kronig continuation extends them to imaginary frequency, and the result goes into the same Lifshitz code. This is what an experienced surface-forces group would actually do, and it needs no DFT. We give it every advantage: the same films, the same thickness range, the same Lifshitz solver, and ellipsometry on films prepared in each orientation. We found no reported result on Lacuna for ellipsometry-based Hamaker constants of thermotropic conjugated polymers, so the comparison point must be produced here. The DFT pipeline beats this baseline if it predicts the measured $A$, from the dewetting wavelength, with smaller error in both orientations, or if it gets the sign right where the ellipsometric extrapolation gets it wrong. If the two agree within error, the DFT route is redundant for prediction, and we will say so.
Ablations. Two ablations isolate the key ingredients. The first replaces the MD-averaged tensor with one computed for a single static, perfectly ordered crystal cell. This tests whether phase-specific conformational disorder, which is what templating changes, matters. The second generates ensembles with a classical empirical force field instead of MACE, at matched simulation time, to test whether accurate treatment of the conformational locks matters for $\epsilon(i\xi)$. For the potential itself, the reference point is MACE's reported thermal extrapolation on the 3BPA benchmark, where it "reduced errors by roughly 30% compared to the best existing models" at 1200 K, trained in about 30 minutes on one A100 GPU, according to the MACE paper. We check that our TPT-TT potential reaches comparable force errors on held-out high-temperature dihedral scans before trusting its ensembles.
Why not just…?
Why not just measure $n_o$ and $n_e$ by ellipsometry and plug them into Lifshitz theory? This is the real threat. For gaps near 10 nm, the dominant imaginary frequencies lie well into the ultraviolet, beyond most ellipsometers, so the answer depends on how the spectrum is extrapolated. For a film whose backbone polarizability is strongly anisotropic and depends on conformation, that extrapolation is exactly where the sign can flip. Whether DFT helps is an empirical question. The baseline comparison above settles it directly, and the ellipsometry data also serve as a check on the DFT tensors in the overlapping range.
Why not just choose an isotropic fluid or coating with the right permittivity, the standard route to repulsion noted in the MEMS review? That gives a fixed sign for a fixed material pair. What we add is a single material whose sign is set after deposition by thermal history and anchoring. If the screening step finds no accessible substrate with a sign flip, and only magnitude changes, the claim shrinks to "orientation tunes van der Waals strength". We will report that outcome as such rather than stretch it.
Why not just attribute dewetting to residual stress, molecular weight, electrostatic patches or capillarity? The MEMS review itself identifies patch potentials and contamination as the main sources of disagreement between theory and experiment. Our control is to keep chemistry, molecular weight, thickness and substrate fixed and vary only orientation. We then require the scaling that is the signature of van der Waals-driven spinodal dewetting: the characteristic wavelength grows as $h^2/\sqrt{|A|}$ with film thickness $h$, which follows from balancing the van der Waals pressure against surface tension. Stress-driven or nucleated rupture does not produce this scaling.
Why not just treat orientation with classical anchoring and elasticity theory, with no fluctuation forces at all? We agree, and we do. Anchoring theory sets the orientation. The van der Waals interaction is the consequence we predict, not a cause of the ordering. Whether the van der Waals contribution feeds back on the ordering, the speculative third gap in the file, is outside the core claim.
Proposed Experiments and Decisive Checks
The decisive experiment uses the sign-flip substrate chosen by screening. Films of TPT-TT, with oligomer analogues used if melt viscosity makes the experiment too slow, are spin-coated at thicknesses from 4 to 20 nm. Their orientation is templated face-on or edge-on, checked by grazing-incidence diffraction, and then held at about 150 °C, inside the nematic window. Atomic force microscopy topography and its Fourier analysis give the dewetting wavelength versus thickness, which yields the magnitude of $A$ and, through whether spinodal patterns appear at all, its sign.
The central prediction is that one orientation stays stable while the other shows spinodal dewetting with $h^2$ scaling. On the control substrate, both orientations should behave the same way qualitatively, with different wavelengths in the ratio the tensors predict. Three outcomes would falsify the idea: both orientations behave identically on the sign-flip substrate, the dewetting does not show $h^2$ scaling, or the orientation-blind baseline fits as well as the anisotropic model.
A secondary check reuses the same tensors to predict colloidal-probe force curves in air against room-temperature crystalline films templated by the two protocols. The point is to confirm that one dielectric dataset predicts two independent observables.
A final, exploratory check uses ferric tosylate doping, which raises TPT-TT's conductivity from 18 to 55 S/cm in the templating study. Doping should sharply increase the low-frequency permittivity and override the orientation effect at larger thicknesses. This marks the boundary of the mechanism rather than extending the claim.
References
- Casimir Effect in MEMS: Materials, Geometries, and Metrologies—A Review
- Casimir stress inside planar materials
- Repulsive Casimir force from fractional Neumann boundary conditions
- Finite-temperature Casimir effect in piston geometry and its classical limit
- 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
- Impact of Side-Chain Polarity and Symmetry on the Structure and Properties of Acyclic Dioxythiophene Polymers
- MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields