Chapter 3
Computational Workflow: From Physics Problem to Simulation
These practical aids help you plan a computational project, keep track of simulation runs, and explain how to reproduce a result. The templates are plain text and can be edited with any text editor. Keep only the fields that are useful for your work and adapt them to the scale of the project.
Downloadable templates
On a narrow screen, scroll horizontally to view the table.
| Template | When to use it | What it records |
|---|---|---|
| Project plan (TXT) | Before implementation, with revisions as the project develops. | The physical question, assumptions, equations, units, numerical strategy, and planned verification and validation. |
| Run record (TXT) | For each important run, or alongside a parameter-sweep table. | Run ID and date, code version, environment, inputs, numerical controls, random generator and seed when relevant, outputs, checks, and conclusions. |
| Project README (TXT) | When the project first runs, then keep it current. | The project purpose, required software, file organization, execution instructions, example results, and limitations. |
Using the templates
Save a copy of the plan with the project. Give important runs distinct identifiers and retain the inputs, source version, and analysis steps needed to reproduce their outputs. Record a change and its purpose before comparing a new run with an earlier one.
A small project needs only a short record. Mark fields that do not
apply, and link to existing input files or parameter tables instead
of copying the same information repeatedly. Test the README by
following its instructions from a clean working folder. In your
project, you may save the README as README.txt or
README.md.
For version control, debugging, profiling, and testing tools, see the Chapter 2 resources.
Verification and validation
Verification asks whether the numerical implementation solves the mathematical model correctly. Checks may include an exact or manufactured solution, numerical refinement, a conservation law, or comparison with a trusted code. Record the check, its acceptance criterion, and the observed result.
Validation asks whether the model adequately describes the physical system for its intended use. Record the physical comparison, experimental or observational uncertainty, and range of conditions tested. Agreement with an exact solution of the same equations supports verification; it does not by itself validate the physical assumptions. State when a physical validation comparison is unavailable.
Before accepting a result
- Does it answer the original physical question?
- Are units, conventions, initial conditions, and boundary conditions consistent?
- Have the relevant tests and diagnostic checks passed?
- Are the conclusions stable under appropriate numerical refinement?
- How sensitive are the conclusions to uncertain parameters and modeling assumptions?
- What evidence supports the physical model, and what remains untested?
- Can the result and its figures be reproduced from the saved source, inputs, environment information, and analysis steps?
Reproduction may mean agreement within a numerical tolerance or statistical uncertainty rather than identical bits on every machine. State the level of agreement relevant to the calculation.