ImpAgingSim

Simulation code and study data for sequence-controlled A/B heteropolymer melts. The model separates the range of sequence correlations from their amplitude and measures post-quench composition fluctuations at fixed chemistry, chain length, density and mean composition. Follow-up studies are maintained in HetPoly.

Committed run data · Campaigns and datasets · Figure sources · Data archive

Model and sequence construction

The model is a periodic melt of A/B bead–spring chains, simulated with Langevin dynamics in OpenMM. Harmonic bonds connect adjacent monomers. Every pair shares a WCA repulsive core; a separate attractive branch gives like pairs a deeper attraction than A–B pairs. Changing the A–B attraction therefore preserves excluded volume.

Bead–spring geometry. Teal and orange identify A and B beads. The periodic boundary translates a chain segment by one box length. This is a schematic; the baseline production system contains 144 chains of 40 beads.

Baseline parameter	Value
Chains × beads per chain	144 × 40 = 5,760 beads
Box length; bead density	$22\sigma$; $0.54095\sigma^{-3}$
Mean composition	$f_A=0.5$
Attraction depths	$\epsilon_{AA}=\epsilon_{BB}=1$, $\epsilon_{AB}=0.1$
Core depth; attractive cutoff	$\epsilon_{\mathrm{core}}=1$; $2.5\sigma$
Bond stiffness; rest length	$200$; $\sigma$
Equilibration → quench temperature	$T^*=5\rightarrow0.7$
Langevin friction; time step	$1/\tau$; $0.005\tau$
Equilibration; production	30,000; 250,000 steps
Stored spectra; run summary	Every 2,000 steps; mean of final five spectra

A Markov backbone $z_i$ sets persistence through $\lambda=2\pi-1$. An independent mask retains each backbone label with probability $\kappa$ and otherwise redraws it at the same mean composition: $s_i=b_i z_i+(1-b_i)u_i$. A redraw can produce the same species.

Independent sequence controls. For the unconditioned generator, $\mathrm{Var}(s_i)=4f_A(1-f_A)$, while at positive contour lag $\mathrm{Cov}(s_i,s_{i+\ell})=4f_A(1-f_A)\kappa^2\lambda^\ell$. Thus $\pi$ sets the exponential range and $\kappa$ scales its nonzero-lag amplitude. The plotted covariance curves are analytical illustrations.

Stochastic chains are drawn independently within their own boundaries; they are neither copies of one sequence nor required to be unique. Their composition is fixed in expectation. The fixed-density size comparison conditions full chain sets on an exact global 50:50 composition. Deterministic alternating and $A_4B_4$ controls repeat the specified sequence on every chain.

Committed study data

Fixed-density structure. Left: 144-chain spectra at $\pi=0.99$ and $T^*=0.7$. Right: the condition-selected peaks across three sizes. The channel is the total-bead-normalized $S_{\psi\psi}^{(N)}/2$, measured directly at periodic reciprocal vectors. Bands and bars show mean ± SEM across five seeds.

Chains $M$	Box length $L/\sigma$	$\kappa=0$ peak	$\kappa=1$ peak	Ratio
144	22.000	6.23 ± 1.22	540.9 ± 13.4	86.9×
288	27.718	6.60 ± 0.75	773.3 ± 58.6	117.2×
576	34.923	6.83 ± 1.25	575.4 ± 39.6	84.3×

Here $L=22(M/144)^{1/3}\sigma$ keeps bead density fixed. For each condition, the shell maximizing the across-seed mean final-window spectrum is selected, then all five seed amplitudes are evaluated at that same shell. The correlated peak lies at $q_{\min}=2\pi/L$ in the two smaller boxes and $\sqrt2q_{\min}$ in the largest. Its nonmonotonic amplitude and box-dependent position leave the bulk wavelength and scaling unresolved.

The committed campaign includes its design manifest, metadata, completion checksums, saved spectra, snapshots and derived condition summaries. The matrix-RPA implementation is in melt/rpa_matrix.py; it returns structure factors only for stable homogeneous kernels. The baseline bare closure is unstable and does not determine a physical spinodal.

Run and reproduce

Install Python dependencies and the test runner:

python3 -m pip install -r requirements.txt pytest

Run one baseline production condition (CUDA-capable OpenMM installation required for --platform CUDA; use CPU or Reference when appropriate):

python3 -m melt.run \
  --out output/melt/demo --run_id k1 \
  --sequence correlated --kappa 1 --pi 0.99 --f_A 0.5 \
  --n_chains 144 --chain_length 40 --box_size 22 \
  --T_equilibrate 5 --T_quench 0.7 --lj_eps_AB 0.1 \
  --equilibration 30000 --n_steps 250000 --snapshot_interval 2000 \
  --compute_density --grid_size 56 --save_trajectory \
  --seed 1 --platform CUDA

melt.scan runs the original sequence, persistence–amplitude, composition and cross-attraction grids; use python3 -m melt.scan --help for the grid arguments and the campaign manifest for the production designs.

Plan and launch a fresh 30-run fixed-density campaign from a clean committed checkout. Store its output outside the checkout so committed study outputs and their manifests remain intact:

python3 -m melt.fixed_density_size_scan --dry-run
python3 -m melt.fixed_density_size_scan \
  --manifest-only --out ../study-runs --campaign-id reproduction --platform CUDA
python3 -m melt.fixed_density_size_scan \
  --run --out ../study-runs --campaign-id reproduction --platform CUDA
python3 -m melt.fixed_density_size_scan \
  --analyze --out ../study-runs --campaign-id reproduction

The launcher pins code provenance, verifies completed-run checksums and skips verified runs. --run-index selects a scheduler-array task. The committed completed campaign is fixed_density_pi099_v2.

Run the code checks and regenerate the README's model and structure figures:

python3 -m pytest tests/ -q
python3 scripts/render_schematic_figures.py --export
python3 scripts/render_potential_figure.py
python3 scripts/render_structure_figure.py

Figure documentation describes the data inputs and rendering dependencies. AWS instructions cover the static parameter suite and fixed-density campaign.

Repository layout

Location	Contents
melt/	Simulation engine, sequence generation, observables and RPA implementation
output/melt/fixed_density_size/fixed_density_pi099_v2/	Completed 30-run campaign and measured summaries
scripts/	README figure renderers and archive download/upload tools
aws/	Campaign launchers and dataset manifest
tests/	Simulation and numerical regression checks
docs/figures/	README model and measured-structure figures

Data availability

Large study datasets are archived at 10.5281/zenodo.20499120. The campaign manifest identifies the relevant runs.

Archive file	Study data
heteropolymer_microphase_data.tar	Production metadata, stored spectra, snapshots and available trajectories
fixed_density_campaign.tar	Completed 30-run size comparison and earlier execution provenance
Supplementary_Data_1_source_tables.zip	Numerical source and sensitivity tables, selected validation trajectories and historical analysis scripts

Download and checksum-verify the simulation archives:

python3 scripts/fetch_zenodo.py --dest output/zenodo --only \
  heteropolymer_microphase_data.tar fixed_density_campaign.tar