MAI Physics / Research & proposed pilots
How can an immersive scientific environment help people inspect models, test assumptions and reproduce results?
At MAI Physics, we are developing a research series on scientific virtual worlds: how physics-informed models, open code and careful validation could support collaborative virtual laboratories.
Concept artwork and proposed experiments. The artwork is not simulation evidence.
01 / Research
A scientific workspace should expose its assumptions, governing equations and evidence. We begin with small problems that can be defined and checked before making claims about a complete digital twin.
Scientific Virtual Worlds / 01
A shared 3D environment can help us inspect a model together. Trust grows when we can also examine its assumptions, compare predictions with evidence and reproduce the work.
What evidence would your community require before trusting a simulation?
Scientific Virtual Worlds / 02
Start with a simple heat-diffusion problem. The interactive educational experiment below states the initial and boundary conditions, calculates a finite-difference baseline and compares it with a known analytical solution. It is not a trained-model result.
A spacecraft thermal model also needs its actual geometry, materials, heat loads and radiative environment.
Which boundary condition would you change first?
02 / Proposed services
A clear starting point for a research conversation—not a claim of a finished product, a validated spacecraft model or a live commercial service.
Agree the question, intended use, assumptions, available inputs and what would count as a meaningful baseline.
Choose a small, reproducible demonstration and state what it can—and cannot—establish.
Plan the checks, versioned record and limitations needed for an informed next-step decision.
Proposed pilot structure only. Deliverables, feasibility, timing and commercial terms are not yet agreed. This preview has no payment flow, booking system or data-collection form.
Discuss your question03 / Method
A persuasive scene is not a validation result. Our proposed evidence record keeps the scientific question and the checks behind it visible.
Record the research question, intended use, governing model and known limits.
Keep the data source, code version and execution environment alongside the work.
For learned models: seed and training configuration. For aerospace simulation: geometry, mesh, solver settings and boundary conditions.
State the evaluation method, baseline comparison, failure cases and what the evidence does not support.
04 / Synthetic lab
This small educational experiment solves a one-dimensional heat equation in your browser. It is separate from MAI Field Lab and from the proposed physics-informed neural network research series.
Finite-difference calculation · analytical comparisonThe solver and reference share the same initial and boundary conditions. This is a synthetic calculation—not an experimental measurement, a material-specific prediction or a spacecraft thermal model.
Illustrative, non-dimensional problem: ∂u/∂t = 0.4 ∂²u/∂x² for 0 ≤ x ≤ 1, with u(0,t) = u(1,t) = 0 and u(x,0) = sin(πx). The smooth curve uses the analytical expression directly. The dots are computed independently with a stable finite-difference update. No neural network is trained.
A future PINN experiment could compare its prediction against this reference. No such MAI training result is claimed here. The separate Field Lab heat demonstration uses finite differences, not a PINN.
A small digital world with a known answer. An explicit finite-difference solver evolves heat along an idealised one-dimensional rod. Compare every step with the analytical solution.
Drag the time slider or use its arrow keys. Home resets; End reaches the final state.
The dots nearly overlap the reference. The errors below reveal the difference; they are not rounded to zero.
40 grid intervals · Δx = 0.025 · Δt = 0.000625 · α = 0.4 · r = αΔt/Δx² = 0.4 ≤ 0.5
The equation: ∂u/∂t = α ∂²u/∂x², on 0 ≤ x ≤ 1, with α = 0.4.
Initial condition: u(x, 0) = sin(πx). The middle starts warmest.
Boundary conditions: u(0, t) = u(1, t) = 0. Both ends are held at zero temperature, so heat can leave the rod.
Reference: u(x, t) = exp(−απ²t) sin(πx).
Numerical update: uᵢⁿ⁺¹ = uᵢⁿ + r(uᵢ₋₁ⁿ − 2uᵢⁿ + uᵢ₊₁ⁿ), where r = αΔt/Δx². This forward-time, centred-space (FTCS) method uses a fixed stable timestep.
Error definitions: eᵢ = uᵢ − u(xᵢ,t); L∞ = maxᵢ |eᵢ|; L2 = √(Δx Σᵢ eᵢ²). Errors are evaluated at the grid points, not continuously between them.
All quantities are nondimensional. This is a synthetic educational example, not a prediction for a particular material or a spacecraft thermal model. No neural network is trained: this is not a PINN, and no AI or GPU performance claim is made.
| Position x | Numerical u | Reference u |
|---|
Educational finite-difference demonstration · No model training · No external data or services · Not spacecraft-validated
Interactive results require JavaScript. The equation and assumptions remain available above.
05 / Research notes
The Scientific Virtual Worlds series brings together literature, proposed experiments and a reproducibility-first way of thinking. The next useful step is a clearly defined question—not a larger claim than the evidence can support.
A first inquiry
Which measurement, baseline or failure case would matter most to your community? A useful research note should make those choices explicit, so another person can inspect and challenge them.
IN DEVELOPMENT · No publication or conference claim.
This custom-code integration was prepared with AI assistance. The existing numerical solver is preserved and checked with automated tests; scientific validation remains a separate task.
The original navigation is retained. A verified, owner-approved channel destination is still required.
No video or channel launch is implied. No fabricated social links are used.
Contact / Pilot inquiry
Discuss a research question, a baseline or a possible pilot. Start with the problem and the decision it needs to inform; scope comes next.
Opens your email app. Nothing is sent automatically. This preview collects no information.