The caution with visual proofs being just because it looks right doesn't mean it is right. A classic example being the "missing square puzzle". More on point examples would be a curve which ever so slightly changes to concave up for a short portion of a shallow concave down region in an overall wavy function, a curve that looks like it converges at a limit but actually doesn't, or something undefined at only 1 point that otherwise looks continuous.
What this kind of thing is really good at is giving good intuition for understanding a proved concept or thinking about a potential solution to an unproven one. It doesn't actually replace having to then do the math behind it to see if it really makes sense. Even a bog standard classroom calculus textbook will show a visual representation of e.g. Simpson's rule before dumping the actual equations and derivations on you.
Note that this doesn't apply to proofs by construction, as in Euclidean geometry.
Those are visual in nature, but the rules must be rigorously followed, if they are the resulting proofs are reliable. The steps of the proof may also be written out textually, but that's merely a translation, the construction itself is a proof. It's how we can tell the difference between a diagram which looks like a trisected angle, and a construction of a trisected angle, which has been proved impossible in the general case.
It kind of doesn't and kinda does at the same time. You can follow these visual mappings of rules to get something that is visually very convincing but subtly not actually a correct result (e.g. something very closely approximating a construction of a trisected angle). If you want to be sure the construction works you've got to have already mathematically proven how the construction method used is valid, applies in the way you used it, and really did come out with the right answer instead of something very close to the right answer which really isn't any less pure math than just doing it symbolically altogether. Something that reasonably makes sense and seems to come out to the right answer is often a great way to shortcut to finding such a validation though, it's just also sometimes a disappointment when it turns out to be arbitrarily close instead of exact.
Correctly following the rules of construction results in a valid proof. That proof is in the steps taken, not the final diagram, because there's no way to tell by looking at a diagram if the rules were in fact followed.
Part of the point I was making is that proof by construction isn't really a "visual proof" at all, it's a proof system where the steps in the proof are added to a diagram, according to exact rules which must be followed at every step.
How do you know the proof is valid for the theorem you are trying to proof? You have to check the logic of the construction, to know things like which measurements are exactly equal. So the visual part of the construction doesn't suffice. It's the same issue as the infinite chocolate triangles glitch.
Is there a more complete explanation? They state that this method fails for the quadratrix, but the chosen vectors drawn in the picture in the video seem clearly nonsensical, so it's not clear to me precisely what procedure was being followed.
Edit: I don't think the video leaves itself enough time to do a good job covering this point. But there's a clearer description of the history at Wolfson (2001) "The Crooked Made Straight: Roberval and Newton on Tangents" AMM 108(3): 206–216 https://jstor.org/stable/2695381
I've been considering reading this book - can you tell me what I might get from it in addition to what I learned in undergraduate maths/physics courses?
I found these courses allowed me to do calculations (e.g. contour integrals, conformal transformations) that were useful but always felt like black magic. I don't have an intuitive feel for the subject. I am wondering if this book will help me with that?
There was something in that book on the relationship between Taylor series and Fourier series which I haven't seen anywhere else, however I'm unable to recollect the details. Do you know what is the relation between these two structures as described in the book?
Reminds me of the simpler geometric example where you show that the area of a parallelogram is base X height, the same as a rectangle, by making a cut in the parallelogram perpendicular to the base and sliding/joining the severed piece to the other edge to show that you can create a rectangle with the same area.
Also the similar one with a bit more calculus, where you slice a circle (well, a disk to be precise) into an even number n of identical sectors and reorder them to form a parallelogram-ish shape (with "bumpy" bases) that approaches a rectangle with side lengths r and πr as n → ∞. This is an easy way to visualize why the enclosed area of a circle is πr². (You do have to take as a given that the circumference equals 2πr though.)