Physics World Model

Generate a spec.md for your imaging system — then click Run to reconstruct, design, or run a scientific simulation.

Generating spec.md…

spec.md

What is spec.md?

A structured file that fully describes your imaging system — forward model, noise, algorithm, and objective — so it can be run, simulated, or redesigned by PWM.

spec.md

## Objective

Reconstruct CASSI using RDLUF-MixS². PSNR ≥ 34 dB.

## Configuration

```yaml

variant_key: cassi

usage_type: reconstruct

algorithm: RDLUF-MixS²

```

↑ Run button reads this

## Forward Model

SD-CASSI · prism · 28 spectral channels · 256×256

## Noise Model

Mixed Poisson-Gaussian η₄

1 · Reconstruct

Algorithm on benchmark data

2 · Mismatch

Correct model error + reconstruct

3 · Design

Design system for application

4 · Sci-Sim

Scientific simulation

Physics World Model — SpecLab Guide

Four ways to use spec.md · examples · downloadable datasets

What is Reconstruct?

Runs an inversion algorithm to recover an image x from its measurement y = H·x + noise. Choose a modality (CASSI, CT, MRI, SPC…) and a solver (GAP-TV, FBP, CS-MRI, RDLUF…). PWM runs the algorithm on benchmark data and reports PSNR / SSIM.

When to use:

Benchmarking a solver on standard PWM data
Reconstructing your own captured measurement (.npy / .mat)
Comparing classical vs. deep-learning performance

Workflow

① Describe Tell SpecLab: modality + algorithm + any constraints

② Generate AI produces a spec.md with H, noise model, solver config

③ Run Click Run — PWM reconstructs & shows PSNR/SSIM

④ Refine Ask to switch algorithm or tune parameters in chat

Example prompts — click to use

Example datasets (.npy)

CASSI

Measurement y Ground Truth x

Sinogram y Phantom x

MRI

k-Space y MR Image x

What is Mismatch?

Handles a gap between the calibrated forward model H̃ and the true one H caused by manufacturing tolerances, environment, or drift. PWM quantifies mismatch parameters, corrects H̃ → H, then reconstructs.

Common mismatch sources:

CASSI dispersion shift (lens misalignment)
CT source–detector distance error
MRI B₀ field inhomogeneity
Lensless PSF phase error from vibration

Workflow

① Describe Specify modality + mismatch type + parameter range

② Generate AI produces spec.md with mismatch_params + correction method

③ Run PWM corrects H̃ → H, then reconstructs

④ Quantify Compare PSNR before/after correction in chat

Example prompts — click to use

Example datasets (.npy)

CASSI

Snapshot y Cube x

Sinogram y Phantom x

MRI

k-Space y MR Image x

What is Design?

Generates a full imaging system architecture from a high-level specification. You provide the application, constraints (budget, size, SNR, resolution), and target. PWM outputs a system DAG: light source → optics → aperture → sensor → digitizer → reconstruction algorithm, with component specs and trade-off analysis.

When to use:

Starting a new instrument from scratch
Hardware–algorithm co-design optimization
Procurement spec writing
Comparing system architectures for a given target

Workflow

① Describe State application + constraints (budget, size, resolution, SNR)

② Generate AI outputs a spec.md with full system DAG + component specs

③ Review Read the right-panel spec.md — component list + trade-offs

④ Refine Chat to optimize a component or relax a constraint

No dataset needed — Design works entirely from text.

Example prompts — click to use

What is Scientific Simulation?

Runs a physics-based forward model from first principles — governing equations, material parameters, boundary conditions — to generate synthetic observables. Unlike imaging reconstruction, the "measurement" is a physical field (pressure wave, temperature, gravitational strain, concentration) rather than a detector readout. PWM encodes the PDE / ODE / statistical model in spec.md and runs the simulation, then optionally inverts it.

Scientific domains covered:

Seismic / geophysics (FWI)

Combustion chemistry (GRI-Mech)

Gravitational wave astronomy

Plasma / MHD physics

Computational fluid dynamics

Molecular dynamics / DFT

Atmospheric / climate physics

Quantum chemistry / TDDFT

Workflow

① Describe State the physical system, governing equations, and parameters of interest

② Generate AI produces spec.md with forward model, solver settings, and output observables

③ Simulate PWM runs the forward model and visualises the physical field

④ Invert Optionally: run inversion / parameter estimation to recover unknowns

Example prompts — click to use

Bring your own data. Upload initial conditions, velocity models, or measured spectra via the dataset panel below the chat. PWM will incorporate them into the simulation spec.

Press Esc to close