Development Status & Performance

This page tracks the development history of RheoJAX across 11 completed phases and provides performance benchmarks.

Development Phases

Phase 1 (Complete): Core Infrastructure

  • Base abstractions (BaseModel, BaseTransform, RheoData, Parameters)

  • Test mode detection system

  • Mittag-Leffler functions for fractional calculus

  • Optimization integration with JAX gradients (NLSQ 5-270x speedup)

  • Model and Transform registry system

  • File I/O (TRIOS, CSV, Excel, Anton Paar readers; HDF5, Excel writers)

  • Visualization with matplotlib

Phase 2 (Complete): Models and Transforms

  • 53 rheological models across 22 families

  • 7 data transforms (FFT, Mastercurve/TTS with auto-shift, Mutation Number, OWChirp/LAOS, Smooth Derivative, SRFS, SPP Decomposer)

  • Pipeline API for fluent workflows

  • 246 tutorial notebooks across 21 categories (basic, transforms, bayesian, advanced, dmta, and 16 model family suites)

Phase 3 (Complete): Bayesian Inference

  • NumPyro NUTS sampling with NLSQ warm-start (2-5x faster convergence)

  • Uncertainty quantification via credible intervals and posterior distributions

  • ArviZ integration (6 diagnostic plot types: pair, forest, energy, autocorr, rank, ESS)

  • Model comparison (WAIC/LOO)

  • BayesianMixin: All 53 models support Bayesian inference

  • BayesianPipeline with fluent API for NLSQ -> NUTS workflows

  • 9 Bayesian inference tutorial notebooks (including SPP LAOS)

Phase 4 (Complete): Performance & Correctness (v0.3.0-v0.4.0)

  • v0.3.0: Generalized Maxwell Model with transparent element minimization (R-based)

  • v0.3.0: Automatic shift factors for time-temperature superposition (TTS)

  • v0.3.0: Bayesian prior safety mechanism (tiered hard_failure/suspicious/good)

  • v0.3.1: JAX-native foundation (30-45% improvement) - 5 foundational optimizations

  • v0.3.2: Vectorization optimizations (50-75% cumulative) - intelligent convergence & batching

  • v0.4.0: Mode-aware Bayesian inference (CRITICAL correctness fix for creep/oscillation)

  • v0.4.0: GMM element search optimization (2-5x speedup with warm-start)

  • v0.4.0: TRIOS auto-chunking for large files (50-70% memory reduction)

Phase 5 (Complete): Soft Glassy Rheology & SPP (v0.5.0)

  • SGRConventional model (Sollich 1998) for soft glassy materials

  • SGRGeneric model (GENERIC thermodynamic framework, Fuereder & Ilg 2013)

  • SGR kernel functions (Fourier transforms, yield stress, aging dynamics)

  • SRFS transform (Strain-Rate Frequency Superposition for flow curves)

  • Shear banding detection and coexistence analysis utilities

  • SPPDecomposer transform (Sequence of Physical Processes for LAOS)

  • SPPYieldStress model with Bayesian inference support

  • SPPAmplitudeSweepPipeline for amplitude sweep workflows

Phase 6 (Complete): Shear Transformation Zone (v0.6.0)

  • STZConventional model (Langer 2008 effective temperature formulation)

  • Three complexity variants: minimal, standard, full

  • Multi-protocol support: steady shear, transient, SAOS, LAOS

  • State evolution: effective temperature, STZ density, orientation

Phase 7 (Complete): Elasto-Plastic Models (v0.6.0)

  • LatticeEPM: Mesoscopic lattice model with FFT-based stress redistribution

  • TensorialEPM: Scaffolding for full tensor implementation

  • EPM Kernels: JAX-accelerated Eshelby propagator and plastic event logic

Phase 8 (Complete): Advanced Constitutive Models (v0.6.0)

  • Fluidity-Saramito EVP: Tensorial viscoelasticity with thixotropic fluidity

  • IKH/FIKH: Isotropic-kinematic hardening with fractional variants (4 models)

  • Hébraud-Lequeux: Mean-field model for concentrated emulsions

  • Giesekus: Single-mode and multi-mode nonlinear viscoelastic models

  • DMT: de Souza Mendes-Thompson thixotropic models (local + nonlocal)

  • ITT-MCT: Mode-Coupling Theory for dense colloids (schematic + isotropic)

Phase 9 (Complete): Transient Network Models (v0.6.0)

  • TNT: 5 transient network variants (SingleMode, LoopBridge, StickyRouse, Cates, MultiSpecies)

  • VLB: 4 Vernerey-Long-Brighenti models (Local, MultiNetwork, Variant with Bell/FENE, Nonlocal PDE)

  • Distribution-tensor formulation with analytical SAOS + diffrax ODE for transients

Phase 10 (Complete): Vitrimer and Nanocomposite Models (v0.6.0)

  • HVM: Hybrid Vitrimer Model with 3-subnetwork architecture (P + E + D)

  • HVNM: Hybrid Vitrimer Nanocomposite Model extending HVM with interphase network

  • TST kinetics with Arrhenius temperature dependence

  • Factor-of-2 relaxation, Guth-Gold strain amplification for HVNM

  • 5 factory methods each for limiting cases

Phase 11 (Complete): DMTA / DMA Support (v0.6.0)

  • DeformationMode enum (SHEAR, TENSION, BENDING, COMPRESSION) in core.test_modes

  • convert_modulus() utility for array-level E* ↔ G* conversion via Poisson’s ratio

  • BaseModel fit()/predict()/fit_bayesian() accept deformation_mode and poisson_ratio

  • Automatic E* → G* conversion at model boundary; G* → E* reconversion in predict()

  • CSV reader auto-detects E’/E’’ columns and sets deformation metadata

  • ModelRegistry find(deformation_mode=) filtering — 49 models support TENSION, 10 shear-only

  • Poisson ratio presets: rubber (0.5), glassy polymer (0.35), semicrystalline (0.40)

  • 8 DMTA handbook pages (theory, workflows, models, protocols, knowledge, numerical, extensions)

  • 8 DMTA tutorial notebooks covering basics through cross-domain validation

  • 33 new tests (modulus conversion, deformation mode enum, registry integration)

Performance Benchmarks

RheoJAX delivers exceptional performance through JAX acceleration and systematic optimizations (v0.3.1-v0.3.2):

Performance Benchmarks (v0.6.0)

Operation

Baseline (NumPy/SciPy)

v0.4.0 (JAX + Optimizations)

Speedup

Mittag-Leffler (1000 points)

45 ms

0.8 ms

56x

NLSQ Parameter Optimization

2.5 s (scipy)

0.08 s (NLSQ)

30x

Mastercurve Transform (10 datasets)

8.5 s

1.7 s (v0.3.2 vectorization)

5x

GMM Element Minimization (N=10)

50 s (cold-start)

10 s (v0.4.0 warm-start)

5x

TRIOS Large File Load (50k points)

35 MB peak memory

11 MB (v0.4.0 auto-chunk)

68% reduction

Cumulative Improvements v0.3.1-v0.3.2: 50-75% end-to-end latency reduction through JAX-native foundation, JIT compilation, vectorization, and intelligent convergence.

Technology Stack

Core Dependencies

  • Python 3.12+

  • JAX >=0.8.3 for acceleration and automatic differentiation

  • NLSQ >=0.6.10 for GPU-accelerated optimization (workflow system, prediction intervals)

  • NumPyro for Bayesian inference (MCMC NUTS sampling, multi-chain parallelization)

  • ArviZ >=0.23.4 for Bayesian visualization and diagnostics

  • NumPy, SciPy for numerical operations

  • Matplotlib for visualization

  • h5py, pandas, openpyxl for I/O

Optional Dependencies

  • CUDA 12+ or 13+ for GPU acceleration (Linux only)