Chapter 40 — Key Takeaways (The Fortran Career)
A one-page reference for where Fortran work is, what it is called, and how to present what you built.
Who employs Fortran programmers (§40.1)
| Sector | Example employers | What they build in Fortran |
|---|---|---|
| National labs | Oak Ridge, Lawrence Livermore, Los Alamos, Argonne, Sandia (US DOE) | Climate, fusion, materials, nuclear, astrophysics on the largest machines |
| Weather & climate | NOAA, ECMWF, UK Met Office | Operational forecast and climate models |
| Aerospace | NASA, Boeing, Airbus, Lockheed | CFD (flow over a wing), structural analysis, propulsion |
| Energy | Oil & gas, nuclear, fusion, wind | Seismic imaging, reservoir & reactor-physics simulation |
| Academia | Climate / astro / chemistry / physics groups | Community codes (CESM, Quantum ESPRESSO, VASP) + in-house sims |
| Finance | Banks, insurers | Legacy pricing, risk, and actuarial systems |
Honest framing: these sectors genuinely use Fortran (a publicly known fact). "Fortran runs everything" is an overstatement to avoid — the precise claim is that Fortran is dominant, often fastest, for the dense numerical kernels at the heart of scientific and engineering computing.
The roles — a spectrum (§40.2)
| Role | End of spectrum | Day filled with | Core skills |
|---|---|---|---|
| Domain scientist | Science | The question; running existing codes | Domain depth; light coding |
| Computational scientist | Science, building | Methods; writing & validating simulations | Numerical methods + Fortran/Python |
| HPC engineer | Software, for research | Performance, parallelism, architecture | Fortran/C; OpenMP/MPI/GPU; profiling |
| Research software engineer (RSE) | Software | Building & maintaining research software | Software engineering first |
You can move along the spectrum over a career. Bold = the two terms this chapter first-defined.
The shortage (§40.3) — and what it means
| Fact | Status |
|---|---|
| More Fortran jobs than people in several sectors | Widely reported — NOT a hard statistic. Treat as a real pattern, not a number. |
| Cause 1 | Universities largely stopped teaching Fortran → supply shrank |
| Cause 2 | Large validated codes still run and grow → demand steady/rising |
| Cause 3 | Original expert authors are retiring → skill + memory loss |
| What it means for you | A genuine differentiator, not a guarantee — pair it with the rest |
| Salaries | Vary enormously; get real numbers from primary sources (labs' career sites, USAJOBS, ECMWF/Met Office). Do not trust a quoted figure. |
Rule to remember: scarcity is leverage. A rare, needed skill (navigate + modernize a big validated code) beats a longer list of abundant ones.
Presenting Fortran on a résumé (§40.4)
| Do | Don't |
|---|---|
| "Modern Fortran (2018)" framed as HPC | "Fortran" alone (reads as legacy) |
| Pair with Python (f2py), C, OpenMP/MPI | List it in isolation |
Name real methods & libraries (LAPACK dgesv, finite differences) |
Vague verbs ("did numerical work") |
| State measured speedups, labeled as measurements | Claim a benchmark you did not run |
| "Modernized… added modules, intent, tests" | "Some old Fortran" (apologetic) |
The winning profile is polyglot: hot kernel in Fortran + orchestration in Python + scale via the HPC toolchain. That pairing is the strongest single line on the résumé.
Where to go next (§40.5) — the portfolio checklist
| Artifact | What it is | Chapter |
|---|---|---|
| Repository | src/ + app/ + test/ layout, fpm-built |
36 |
| README | What it solves, physics, build/run, validation | 38 |
| LICENSE | MIT/BSD — so others may reuse it | this chapter |
| Figure | One VTK/ParaView steady-state image | 26 |
| Abstract | One paragraph, every clause true of your code | this chapter |
| First contribution | Small, correct, tested change to an open code | 36, 37 |
The six themes of the book (the spine you carry out)
| # | Theme | One line |
|---|---|---|
| 1 | Fortran is not dead | It is infrastructure — weather, climate, reactors, the supercomputers' science |
| 2 | Modern Fortran is a modern language | Modules, OOP, allocatables, coarrays, C interop |
| 3 | Arrays are Fortran's superpower | Whole-array ops + column-major layout = readable and fast |
| 4 | Performance is not accidental | It comes from memory layout, the compiler, and parallelism |
| 5 | Legacy code is not a burden | It is validated infrastructure — modernize, don't rewrite |
| 6 | Fortran and Python are better together | Kernel in Fortran, orchestration in Python, bridged by f2py |
The project, finished
- Piece added this chapter: the solver packaged as a portfolio piece — README + LICENSE + figure +
abstract. No physics changed;
code/project-checkpoint.f90is the complete solver in one file, carrying the abstract in its header, with the exact validated $5\times5$ result from Chapter 24 (interior rows28/32/28and4/4/4, max100.00). - The arc: Ch 1 (choose the domain) → Ch 24 (the finite-difference core) → Ch 38 (the capstone paper) → Ch 40 (the artifact you show the world). You built it.
Numbers & facts worth remembering
- The shortage is reported, not measured — never quote it as a statistic.
- A
real(dp)field of $N\times N$ costs $8N^2$ bytes; a solver needs ≥ 2 such fields at once. - Code with no LICENSE is legally unsafe for others to reuse.
- Integer division still truncates ($7/2 = 3$); kinds still matter; column-major still wins — the whole book, in one line.
Go do it
Package the solver. Open one pull request. Break the seal on your Exercise 1.28 sentence and read how far you have come. Then apply for the thing that looked out of reach forty chapters ago.