Full transcript of “Which Font Makes the Best ⌶-Beam?”. Click a timestamp to open the video at that moment. Need another video? Use the YouTube transcript generator.
0:00 - "Which typeface's capital I makes the best I-beam?" It's a question I was asked three years ago, and I've since taken far too seriously. This is what a regular I-beam should look like, and this is what happens if you take it from the front of the Disney logo. Yeah, things only get weirder from here. I built bridges, printed pamphlets, ran simulations, and 1,000 fonts later, well, it turns out that I might not even be the best letter when it comes to typography-based civil engineering. This was a ridiculous idea. (light suspenseful music) If you were to break down the time I spent picking the right font for an essay versus actually writing the thing,
0:46 then you'll have a good idea as to why I failed English. Personally, I blame this thing, the printing press. In 1440, Johannes Gutenberg worked out a way of casting letters into lead, rearranging those letters into words held together in a wooden frame, covering it with ink, stamping it onto paper, and then binding the paper into books. Or rather, one book. He basically just printed the Bible. Traditionally, Bibles were copied out by hand by scribes using an ink and quill. Since that's what he had to base it on, Gutenberg created a font that had big, thick lines in what is now known as a black letter. It looks fantastic, but unfortunately, it's not particularly legible. I can't read it. Although to be fair, that could be 'cause it's written in Latin.
1:32 Eventually, in 1470, Nicholas Jenson came along with the revolutionary new idea of making a typeface that people could actually read. He based it on the Roman carvings in his adoptive home of Italy and called it Roman type. Even today, look at literally any book, and chances are you'll see a descendant of one of Jenson's typefaces. Soon, printing presses were being set up all over Europe, with foundries coming up with new fonts all the time, some to further improve readability, others based on pure geometric form, and still more as artistic expression. Of course, nowadays, we know that the true purpose of a font is to build a bridge out of it. (tense thoughtful music) You don't need an engineering degree to build a bridge.
2:20 All you need is a big piece of driftwood, or a stone slab, or a tree trunk, or whatever other massive monolith you can get to span the two banks of a river. Simple beam bridges like this one can last for thousands of years. Although, getting the material and putting it in place can be a massive pain. Often, it's easier to get a smaller profile, get it to span those same two banks, and it does the job just as well. You can keep doing that, getting smaller and smaller beams right up into the point where... (wood clanks) Yeah, that one might've been too small. Even today, where we have far grander bridges with far more complex structures, at the heart are the same beams we've been using since the Stone Age. Although, nowadays, rather than making them smaller,
3:07 instead, they tend to look like the letter I. If you take a regular profile and apply a load to it, it experiences bending, although it isn't even throughout the structure. At the top, we're in compression, and the material is pushed together, while at the bottom, tension, and it's pulled apart. The reason a material fails is due to its atomic structure. In a metal such as steel, we have a lattice of ions that are attracted to a sea of delocalized electrons. With tension, the ions are being pulled apart, weakening the bonds until they snap. Meanwhile, in compression, the ions are being pushed too close together, and the structure buckles. For the sake of argument, let's say that this distance corresponds to one millimeter over the entire length of the beam.
3:52 Since the entire cross-section is rigidly connected, in order to be pulled apart by one millimeter here around the middle, then the bottom edge must have already been stretched by two millimeters, which is double its theoretical limit. So what actually must happen is that the material failed ages ago at this bottom edge when it first hit that one-millimeter point. This same can be said for compression in the upper portion of the beam. More material means greater resistance to deflection, so our optimal solution is to put literally all of the material in the upper and lower flanges, protecting the stuff held in the middle. In the real world, the way we produce I-beams is by hot rolling structural steel, literally squeezing out the desired shape. Once you get it going, you can produce up to 100 kilometers of beam
4:38 every single hour. Although the reason it works so well is because the eye is intrinsically a very simple shape, that is right up and to the point, we start making silly fonts out of it. Initially, I thought that 3D printing would work, but it turns out that really gives you indication of layer-by-layer bonding strength rather than any sort of properties of the shape itself. Using a laser to blast through a meter of steel would've been amazing, but getting access to a Death Star-level laser beam isn't something that I could conceivably do. So, instead, I used a four-axis CNC router with blocks of HDPE plastic, eventually producing six beams and two massive packs of shavings. Since fonts come in wildly different sizes and I only had blocks of 12-by-12-centimeter plastic,
5:26 I scaled everything up or down to fit inside that area. The most important result will be the load carried versus mass of material used, which is why you can't cheat the system with an I that's actually just a solid block of material. To perform the testing, I used an Instron machine with a custom heavy-duty mount. First up was the traditional I-beam like you'd see in a bridge. It bends, buckles, and finally breaks. (plastic snapping) Looking at the stress-strain data, we can see that load increases in kind of an arc before maxing out at 16,000 newtons, which is what we call the ultimate strength. That's about the same amount of load as dumping in a medium-sized car. While the actual failure occurred beyond here,
6:13 it wasn't able to hold any extra load, and this failure was just as likely to be from the actual material breaking as it was from the beam deflecting so far that it messed up the mounting bracket. So, to keep things fair, I'm using ultimate strength as beam strength. Under the premise of more flanges equals more better, I used the Google font Kablammo for the next test. And it worked, surviving to almost 26,000 newtons. In recognition of the fact that most I's are written as straight lines, I tested out one of these in the form of Arial Black. It gets to 20,000 newtons. Although the I-beam is only really optimized for bending, it's so versatile that it ends up being used in a whole bunch of different load scenarios as well. In buckling, the Apple standard font San Francisco
7:01 gets to 4,000 newtons, while Windows's darling Times New Roman gets to 24,000 before breaking in the most terrifying way possible. A standard Windows license comes with 192 fonts, macOS has 353, and Google Fonts provides a nice neat download of 1,000. That was an insane amount of I-beams to build, so rather than doing them more for real, instead, I worked out which ones were going to be most interesting by doing a little bit of simulation. There are some analytical methods you can use to calculate the bending moment of inertia of different beam profiles, but that doesn't work out too well when you have stress concentrators at the corners and weird lopsided shapes. So, instead, I use some finite element analysis
7:46 to make it a little bit more robust. This works by breaking up our complex shape into thousands of individual points connected together by springs tuned to represent the properties of the material that we are trying to represent. Rather than spending a week manually putting this into my simulator, instead, I spent seven days writing a code to do it for me. After an additional two days of simulation, here are our results: (light electro-synth music) To plot everything, I'm putting mass on the x-axis and load on the y. Here are the Instron bending tests compared to the simulation data, we see that they mostly match up. Although the traditional beam doesn't have the greatest load capacity, it does have the greatest amount for the lowest amount of mass, which, when it comes to actually building things,
8:32 is what matters. Simulations can give us more information as well. For example, with Kablammo, we see that the outer flanges are being used while the inner ones are basically just wasting material. Each line from the origin passing through a series of points represents a series of functionally identical beams. If you really like the look of Academy Engraved but don't have the material for it, then you can scale it down to hold the same amount of load as an equivalent mass of Imbue. In a line passing through the middle, this cluster represents the rectangular I's stretching from Anybody to Hammersmith and the big black 100% filled-in box, which I used as a sort of control. Below this good/bad line, a cluster here tends to be filled with italicized
9:18 or otherwise lopsided beams where the unbalanced profile causes them to fall over prematurely. Above the line, they tend to be highly symmetric, and all of them have working flanges. Although a lot of the fonts come close to a traditional I-beam, the only one to actually beat it was BioRhyme, a Google font which exchanges its filets for wide, load-carrying flanges. Of all the typefaces that you're likely to experience with a stock operating system, the very worst was Zapfino. Not only was it a nightmare to produce, but it also fell apart almost immediately. (machinery whirring) (plastic snaps) The very best was Courier New, a typewriter-inspired font that looks and performs almost identically to a regular I-beam.
10:06 A bit of an anti-climax, but finally, I had my answer. And then I realized that the rotated H also looks almost identical to an I-beam, and if I was gonna test that out, well, then I might as well examine the rest of the alphabet. While I was there, I looked at bending, buckling, tension, and torsion as well. The simulations took an additional six months and produced 13 terabytes of data. Fortunately, I was able to boil that down to an 1,800-page text file and seven gigabytes of pretty renderings. (light electro-synth music) In bending, A's are almost always below the good/bad line. Weirdly enough, it's the old calligraphy scripts which do well, with the extra top bits providing support to balance out the base. D's are best overall,
10:53 with a good support at the top and bottom with a reasonably symmetric design. E's have good strength but wasted mass on that central stroke, which doesn't help with load. Rotated H's do significantly better than both regular H's and regular I's since they almost always have flanges in the top and bottom. Buckling failure occurs in compressive loading with beam deformation in whatever direction causes the least resistance. That means that straight-line I's are particularly susceptible to it, because although they're strong in the vertical plane, they're weak horizontally, and therefore, it's in this horizontal direction where failure ends up occurring. O's and D's end up dominating. Tension is all about area, shape doesn't matter, so everyone ends up performing pretty much the same.
11:40 Finally, in torsion, rounder profiles with more material further away can hold the most load. A's are pretty terrible; O's, especially those with a circular perimeter, do consistently better. Typefaces already designed to be round, like Honk and Monofett, do the best. Overall, considering all the forces an I-beam is likely to experience, the O from Noto Serif and rotated H from Archivo Black end up performing the best, with Kristi J and Genos Z the worst. If you want to construct something from an alphabet soup of a single font, then the thick, reliable strokes of Erica One and Gill Sans are the way to go. Using Fuggles would be a Fuggle up. As to what letter forms
12:26 we should really be making I-beam from, well, O's, D's, B's, and rotated H's are the best. The worst: J, L, F, W, and ironically, I. At the end of the day, should we be building our bridges out of thematically relevant typeface? Yes, some aren't even that bad, and they look a whole lot more exciting than the regular old alternative. If you're looking to build your own house out of your favorite font, then I've left the code, my results, and some instructions in the form of my own peer-reviewed paper, linked in the description. Let me know how it goes. Until then, this has been James Dingley from the Atomic Frontier. Keep looking up. (light thoughtful music)
