Is that effect (gradual decelerate, then accelerate again) real or an indication of some kind of systematic measurement error?

No, this is very likely a systematic measuring error, because, as you note, the two velocities are similar in shape. The reason is quite simple — it’s much easier to get the relative distance between the two runners right, than their absolute position in the track. The main source of uncertainty here is connected to the definition of a line perpendicular to the track. Because of the camera perspective, its angle with respect to the track is relatively sharp angle in the early and mid part of the race, and this makes it harder to draw it accurately, unless you already have some known reference line (such as the start line).

This line is defined once for a given frame, and applies (with some very minor rotation) to both runners. That’s why the two velocities have a similar (but unphysical) shape — there is a common systematic in these measurements. But, of course, this is taken into account in the uncertainties, and is in fact the main source of uncertainty.

For some time, we were thinking about trying to estimate the absolute position of the camera, using the observed angles between the starting line and distances between two consecutive tracks. Given that information, it should be possible to draw perpendiculars at arbitrary positions throughout the track, which at least in theory should improve the accuracy. However, it turned out to be quite a mess getting everything right, so we gave it up. But that’s another fun follow-up project for somebody else

Is that effect (gradual decelerate, then accelerate again) real or an indication of some kind of systematic measurement error?

0.05s being equivalent to 40cm does put it into perspective!

Can you see a reason for this apparent disagreement? (I realize this is all for fun — especially since Bolt is only 22 and has plenty of time to provide an empirical answer to how fast he can run 100 m. I’m just curious.)

I made a more careful comparison of the two data sets by plotting them in the same figure, and then it was clear that I was a little too quick in my assessment — there *are* indeed differences, although they are very small around 8 seconds. The main difference is that we have measured Bolt ~1-1.5 meters further ahead at ~5-6 seconds than the other data set, corresponding to a 2-3 sigma difference in that range. This is not too surprising, given that this is by far the most problematic range to measure, due to lack of robust parallel lines across the track.

Anyway, the implication of this is that we obtain a slightly higher maximum speed for Bolt than the other data set (12.5 m/s vs. 12.1 m/s) up to 8 seconds, and also a stronger deceleration after 8 seconds. And that’s where the “different” projections come from.

However, again it’s important to note that there are large uncertainties here — 0.04 seconds is a pretty substantial chunk relative to the small time differences we are talking about here. So when we are talking about differences here, we are in fact talking about two projections that agree to within 2 sigma, and that’s actually pretty good, given the crude data used for this analysis. Further, I don’t know how the other data were obtained, and what the corresponding uncertainties are; surely they are not zero. So all in all, I think the agreement is pretty good, and the small differences we do observe arise in precisely the most problematic region of the track.

If you don’t mind one more question . . . you said your analysis isn’t really inconsistent with that “Science of Sport” article linked above. But that article seems to be claiming that even if Bolt hadn’t slowed at all, he’d still finish no faster than 9.60.

Even if he had maintained his speed for the final 20m (which is unlikely – he’d probably have dropped off slightly), he would have run 0.09 seconds faster. This would give him a 9.60 second time

In contrast, you seem to be saying he could do that well just by matching Richard Thompson’s acceleration, who clearly did slow towards the end of the race (as shown in your figure 2).

Can you see a reason for this apparent disagreement? (I realize this is all for fun — especially since Bolt is only 22 and has plenty of time to provide an empirical answer to how fast he can run 100 m. I’m just curious.)

In contrast, the paper on arXiv claims Bolt was at the 81.7 m mark at 8 seconds.

Just wanted to note that the times quoted in Table 1 are not calibrated, so one has to add 0.09 seconds to all times except the first and last two. So according to our numbers, Bolt was at 81.7 m at 8.09 sec, while the other data set has him at 80 meter at 7.96 sec. Given that he runs ~ 12 m/s, he will have moved ~1.5 meter in 0.13 sec, which means that we’re quite consistent, at least to within the error bars.

The two analyses are in fact quite compatible, and the main difference lies in the assumptions used for extrapolation and interpretation. That being said, it seems somewhat unreasonable to us that the most “probable” number should be just 0.05 seconds better than what actually happened — that corresponds to ~40 cm, and that simply seems too little, given the development of the run between 60 and 80 meters. In fact, just his posture alone when crossing the finishing line is probably something like 20 cm. (He’s standing upright, or slightly backwards, but could have been leaning forwards, and it’s the chest that counts.)

But it’s important to note that these calculations have lots of uncertainties, and the assumptions about the acceleration makes noticable differences.

And, of course, the most important thing about this whole thing is that it’s fun — not that it’s millimeter precision science

There is a detailed analysis here. You can see that Bolt reached his top speed around the 60m mark. That article estimates that celebrating only cost Bolt about 0.05 seconds.

There seems to be a bit of a disagreement between that article and the paper on how far along Bolt was at 8 seconds.

The article you linked to has Bolt at 80 m after 7.96 s. That means he couldn’t possibly have been much further than the 80.5 m mark at 8 seconds.

In contrast, the paper on arXiv claims Bolt was at the 81.7 m mark at 8 seconds.

This makes a big difference. If the arXiv paper is right, Bolt “only” had to run the last 1.55 seconds with an average speed of 11.8 m/s to finish in 9.55 s. That’s possible. In contrast, if the article you cited is right, Bolt would have had to run the last 1.55 seconds with an average speed of nearly 12.6 m/s to finish in 9.55 s.

I have no idea which source has the more accurate data for how far along Bolt was at the 8 second mark, but the arXiv paper at least explains how they calculated their numbers.