Nominal Drift

Where failed assumptions become working systems.

Nominal Drift is a local-first scientific AI workstation for materials science, corrosion, diffusion physics, and crystal intelligence.

It is being built as a hybrid system where:

• LLM reasoning supports scientific dialogue, orchestration, and reporting
• deterministic numerical engines produce inspectable physical outputs
• mechanism-aware visualisation makes diffusion behaviour visible
• experiment persistence and retrieval preserve validation history
• crystal dataset pipelines support structure search, parsing, and evaluation workflows

This is not a generic chatbot.

The model reasons.
The science lives in the tools.

---

What it does

Nominal Drift is designed to answer scientific questions with traceable computational pathways.

Its current focus is to keep numerical outputs tied to explicit models and named data sources rather than hidden heuristics or vague “AI intuition”.

Core workflow areas include:

• diffusion solvers
• sensitisation screening
• heat-treatment studies
• thermodynamic layer integration
• crystal dataset ingestion and search
• animation and engineering visualisation
• experiment storage and retrieval
• reporting and scientific narration

The objective is simple:

Ask scientific questions in natural language
and receive physically grounded, inspectable outputs.

From chromium depletion in stainless steels to crystal structure browsing and future thermodynamic extensions, Nominal Drift is being built as a scientific copilot that stays local.

---

Current capabilities

Materials and corrosion workflows

• 1D Fickian diffusion engine (Crank–Nicolson)
• chromium / carbon / nitrogen diffusion workflows
• austenitic and ferritic material preset paths
• sensitisation and depletion studies
• heat-treatment mechanism studies
• engineering-profile diffusion animation from real solver output
• schematic microstructure-style visualisation derived from computed depletion fields
• runtime provenance display for constants and active material settings
• experiment memory and comparison database
• CLI and GUI orchestration

Thermodynamic layer (in active integration)

Nominal Drift is being extended with a distinct Layer 2 thermodynamic path using pycalphad.

This layer is intended to sit alongside diffusion, not replace it:

• Layer 1 = baseline diffusion physics
• Layer 2 = thermodynamic equilibrium / phase plausibility
• Layer 3 = crystal structure and dataset workflows

Current design goals for Layer 2:

• explicit TDB-backed thermodynamic queries
• composition + temperature phase plausibility
• separate provenance from diffusion outputs
• support for both:
  • 316L / austenitic Fe-Cr-Ni-C-N
  • 430 stainless steel / ferritic Fe-Cr-C-N

Important:

• diffusion outputs remain authoritative only for diffusion
• thermodynamic outputs must remain clearly separated
• precipitation kinetics, segregation energetics, and CALPHAD extensions must never be implied unless actually implemented

Crystal intelligence lane

The crystal lane is operational as a separate scientific lane.

Supported datasets include:

• MP-20
• MPTS-52
• Perov-5
• Carbon-24

Capabilities include:

• dataset import and on-disk status reporting
• normalized structures.jsonl pipelines
• pymatgen-driven structure parsing
• crystal search and filtering utilities
• CIF-style structure inspection workflows
• generative crystal evaluation bridge
• future DiffCSP-style structure workflows

Important:
these datasets are not currently used to drive diffusion coefficients or sensitisation calculations.
They belong to the structure-aware crystal lane unless explicitly connected by a validated scientific model.

---

Scientific architecture

Nominal Drift is being structured as a three-layer scientific system.

Layer 1 — Baseline physics

This is the authoritative numerical layer for diffusion outputs.

It currently includes:

• Arrhenius diffusion constants
• Fick’s Second Law
• Crank–Nicolson finite-difference solving
• explicit material presets
• strict element/system gating
• no silent fallbacks

This layer is responsible for outputs such as:

• concentration profiles
• depletion depth
• minimum chromium concentration
• time-evolving diffusion fields

Layer 2 — Thermodynamic / kinetics layer

This layer is being introduced to answer questions that diffusion alone cannot answer, including:

• which phases are thermodynamically plausible at a given temperature
• whether carbide/nitride formation is thermodynamically accessible
• how alloy chemistry interacts with phase stability

This layer is where pycalphad belongs.

It is intended to provide:

• TDB-backed phase plausibility
• explicit composition–temperature equilibrium queries
• provenance on active database, phases, and assumptions

This layer is not yet equivalent to full precipitation kinetics, grain-boundary segregation energetics, or DFT-backed migration modelling.

Layer 3 — Structure-aware layer

This layer contains:

• MPTS-52
• MP-20
• Perov-5
• Carbon-24
• pymatgen structure tools
• normalized crystal records
• retrieval / search / evaluation pipelines

Its current honest roles are:

• structure search
• reference retrieval
• crystal data inspection
• generative model evaluation
• future structure-informed workflows

It is not currently authoritative for diffusion outputs.

---

Grounded in

Nominal Drift is built on a combination of:

• literature-grounded diffusion constants
• classical transport theory
• corrosion science
• experimentally interpretable engineering workflows
• thermodynamic modelling pathways
• crystal dataset ecosystems

Chromium diffusion (substitutional)

Primary reference currently used:

Perkins, Padgett & Tunali (1973)
Tracer diffusion of Fe and Cr in Fe-17 wt% Cr-12 wt% Ni austenitic alloy
Metallurgical Transactions, 4(12), 2535–2540
DOI: 10.1007/BF02644258

Canonical values currently used:

• D₀ = 3.6 × 10⁻⁴ m²/s
• Q = 272 kJ/mol

This remains the primary reference for Cr transport in the 18-8 / 316L composition space.

Carbon diffusion (interstitial)

Grounded in:

• Tibbetts (1980)
• Goldschmidt (1967)
• Smith (1990)

Nitrogen diffusion

Grounded in:

• Jack (1951)
• Grabke (1996)
• Frisk (1991)

Numerical foundation

The transport layer is built on:

• Fick’s Second Law
• Arrhenius diffusion formalism
• Crank–Nicolson finite-difference solving

Validated against:

• Crank, The Mathematics of Diffusion (1975)
• Borg & Dienes (1988)

---

What the loaded crystal datasets do — and do not do

The normalized crystal datasets currently support:

• structure parsing
• crystal search
• dataset status inspection
• evaluation workflows
• future structure-aware scientific extensions

They do not currently:

• provide diffusion coefficients
• override D₀ or Q
• drive the diffusion solver
• supply segregation energies
• supply precipitation kinetics
• replace CALPHAD or DFT

In the current architecture:

• diffusion calculations use literature-backed constants and material presets
• crystal datasets support the crystal intelligence lane
• future bridges between these lanes must remain explicit and scientifically justified

---

Visualisation philosophy

Nominal Drift distinguishes carefully between:

• real numerical outputs
• derived engineering visualisations
• future, unimplemented scientific claims

Current animation modes are designed to avoid misleading presentation:

• the engineering profile is driven directly by solver output
• the schematic view is derived from the computed depletion field
• static or decorative visuals should never be presented as if they were true atomistic motion

Nominal Drift does not currently claim:

• literal atomic trajectories
• true segregation energetics
• true precipitation kinetics
• DFT-backed clustering behaviour
• CALPHAD-backed phase evolution unless the thermodynamic layer is explicitly active

---

Dataset and experiment workflows

Dataset import

The dataset import system reports real on-disk status.

Supported dataset lanes include:

• Perov-5
• MP-20
• MPTS-52
• Carbon-24

The import page is intended to show what is actually present locally, rather than pretending that all datasets are always available.

Experiment database

Nominal Drift includes an experiment database for saving and reviewing results.

This supports:

• saving diffusion runs
• preserving notes and labels
• comparing previous runs
• keeping local engineering history inspectable

The experiment database is for result persistence, not for driving diffusion coefficients or thermodynamic predictions.

---

Architecture

Llama / Ollama
       ↓
Scientific Router
       ↓
Core Toolchain
       ├── Layer 1: diffusion_engine
       ├── Layer 2: thermodynamic_layer
       ├── mechanism_animator
       ├── experiment_store
       ├── crystal_datasets
       ├── pymatgen_structure_lane
       ├── reporting
       └── retrieval