Why physicists still use Fortran

“I don’t know what the programming language of the year 2000 will look like, but I know it will be called FORTRAN.” –  Charles Anthony Richard Hoare, circa 1982

Fortran is rarely used today in industry — one ranking ranks it behind 29 other languages. However, Fortran is still a dominant language for the large scale simulation of physical systems, ie. for things like the astrophysical modeling of stars and galaxies, hydrodynamics codes (cf. Flash), large scale molecular dynamics, electronic structure calculation codes (cf. SIESTA), large scale climate models, etc.  In the field of high performance computing (HPC), of which large scale numerical simulation is a subset, there are only two languages in use today — C++ and “modern Fortran” (Fortran 90/95/03/08). The popular Open MPI libraries for parallelizing code were developed for these two languages. So basically, if you want fast code that an run on many processors, you are limited to these two options. Modern Fortran also has a feature called ‘coarrays‘ which puts parallelization features directly into the language. Coarrays started as an extension of Fortran 95 and were incorporated into Fortran 2008 standard.

The heavy use of Fortran by physicists often confounds computer scientists and other outsiders who tend to view Fortran as a historical anachronism.

What I would like to do in this article is explain why Fortran is still a useful language. I am not advocating that physics majors learn Fortran — since most physics majors will end up in research, their time may be better invested in learning C++ (or just sticking with Matlab/Octave/Python). What I would like to explain is why Fortran is still used, and show that it is not merely because physicists are ‘behind the time’ (although this is sometimes true – about a year ago I saw a physics student working on a Fortran 77 code, and both the student and adviser were unaware of Fortran 90). Computer scientists should (and do) consider the continued dominance of Fortran in numerical computing as a challenge.

Before digging in I feel obligated to discuss a bit of history, since when many hear “Fortran” they immediately think of punch cards and code with line numbers. The original specification for Fortran was written in 1954.  Early Fortran (designated in allcaps as FORTRAN) was, by modern standards hellish, but it was incredible leap forward from previous programming, which was done in assembly. FORTRAN was often programed with punch cards, as is not-so-fondly recalled by Prof. Miriam Forman here at Stony Brook. Fortran has through many revisions, the most well known are the 66, 77, 90, 95, 03, and 08 standards.

It is often said that the reason Fortran is still used is that it is fast. But is it the fastest? The website http://benchmarksgame.alioth.debian.org allows for a comparison of several benchmarks between C and Fortran. On most of the benchmarks, Fortran and C++ are the fastest.  Note that Python, which is the darling of computer scientists, is usually about 100 times slower, but that is the nature of interpreted code.  Python is unsuited for heavy numerical computation but highly suited for many other things. Interestingly, C++ beats Fortran on all but two of the benchmarks, although they are fairly close on most. However, the two benchmarks where Fortran wins (n-body simulation and calculation of spectra) are the most physics-y. The results vary somewhat depending on whether one compares a single core or quad core machine with Fortran lagging a bit more behind C++ on the quad core. The benchmarks where Fortran is much slower than C++ involve processes where most of the time is spent reading and writing data, for which Fortran is known to be slow.

So, altogether, C++ is just as fast as Fortran and often a bit faster.  The question we really should be asking is “why do physics professors continue to advise their students to use Fortran rather than C++?”

Fortran has legacy code

Given the long history of Fortran, it is no surprise that a large amount of legacy code in physics is written in Fortran. Physicists usually try to minimize the amount of coding they do, therefore, if legacy code exists they will use it. Even if old code is hard to read, poorly documented, and not the most efficient, it is often faster to use old validated code than to write new code. Physicists are not in the business of writing code, after all, they are trying to understand the nature of reality. Researchers usually want to take the path of least resistance. Professors usually have legacy code on hand (often code they wrote themselves decades ago) and pass this code on to their students. This saves their students time, and also takes uncertainty out of the debugging process.

Fortran is easier for physics students to learn than C++

Overall, I think Fortran is easier to learn than C++. Fortran 90 and C are very similar, but Fortran is easier to code in for reasons I will discuss. C...