How a property gets deduced, and why you still check it against a real shape

0:00There are two ways to find out whether something is true about a shape. You can draw one and measure it. Draw another and measure that. Keep going. Or you can argue: start from things already settled and work forwards until the thing you wanted falls out. The first way touches paper. The second way never does. Most people assume the second is just the first done more carefully. It is not. They give you different kinds of answer, and the difference is worth more than either of them.

0:35This is about what measuring actually buys you, and what it cannot. Draw a rectangle. Draw its two diagonals. Measure where they cross. Each one cuts the other exactly in half. Try another rectangle: same answer. A tall thin one, a nearly square one, an enormous one. Every time, each diagonal is halved. You have now checked it, let us say, a thousand times, and it worked a thousand times.

1:07So here is the question that decides everything: what does that tell you about the next one? The honest answer is that it tells you what happened in the ones you did. It is very good evidence. It is not a reason the thousand and first cannot be different. That sounds like pedantry until you price it, so here are two claims that have nothing to do with shapes. First: take a whole number, square it, add the number, and add forty-one. The answer is always prime.

1:42Check it at zero, at one, at two. It holds. It goes on holding for forty numbers in a row. At the forty-first it fails, and the failure is not close: the answer is a perfect square. Now the second. Sort every prime by what it leaves when you divide it by four: some leave three, some leave one. Count them as you go, and the threes never fall behind the ones. That survives twenty-six thousand eight hundred and fifty-eight checks.

2:14Then, at twenty-six thousand eight hundred and sixty-one, the ones get ahead. Put those two side by side, because together they say something neither says alone. One false claim died after forty checks. The other walked past a thousand, past ten thousand, and died in the twenty-six thousands. So when a claim survives a thousand checks, you have not learned that it is true. You have learned that if it is false, its first failure is somewhere past a thousand.

2:49That is genuinely useful. It is also nothing like knowing. A claim you have good reason to trust and no argument for is worth having, and worth naming: it is a conjecture. A conjecture is not a guess. It is evidence without a proof, and throwing it away would be silly. So take a claim and push on it. A rectangle has four square corners and its opposite sides match.

3:19Are those two separate demands, or is the second one already inside the first? The way to find out is to try to break them apart. Draw a four-sided shape with every corner square, and with one pair of opposite sides deliberately different lengths. Try it. Ruler, set square, as much care as you like. You will not manage it, and the interesting part is what happens when you fail.

3:47Watch which corner the failure lands in. Here is the attempt made honest. Instead of drawing, walk it. Go five steps forward. Turn a square corner. Go three. Turn another square corner. Now go seven, because seven is the side you wanted to be different from the five. Stop, and join up to where you started, because a four-sided shape has to close. Two of the corners are square, exactly as you built them. The other two are not, and they are not close.

4:22The shape closed. It just closed as something else. The two corners you controlled came out right and the two you did not paid for it. One failed attempt proves nothing, so do the thing properly and hunt. Build every four-sided shape a walk like that can make, over a whole grid of side lengths and corner angles. That is seven thousand six hundred and twenty-two shapes, none of them drawn with any outcome in mind.

4:54Keep only the ones with four square corners, and then look for one with mismatched opposite sides. The count is zero. Not one shape in the sweep has four square corners and sides that disagree. And to be sure the hunt can see such a thing at all, loosen what counts as square to within two degrees. Now it finds them immediately. So the empty result is a finding about shapes, not a blind spot in the hunt.

5:26Which brings us to two sentences that sound the same and are not. I could not find one. And: there is not one. The first is a report about a search. It carries the size of the search inside it, and nothing more. The second is a statement about every shape there could ever be, including the ones nobody has drawn. No amount of the first ever turns into the second. That is what the prime race cost twenty-six thousand checks to show.

5:59To say the second sentence you need a different kind of move. And the move is small enough to fit in one line. Take any four-sided shape with all four corners square. Do not assume anything else about it. Join one corner to the corner diagonally opposite. That single line cuts the shape into two triangles. The two triangles share that line, so one of their sides is the same length in both, without measuring.

6:31Now chase the angles. The diagonal splits one square corner into two parts. One part sits in one triangle. In the other triangle, the angle at the far end has to make up the rest of a hundred and eighty. Work it through and those two angles are forced to be equal. Two angles matching and the shared side between them: the triangles cannot be anything but identical. Identical triangles means matching sides, and those sides are the opposite sides of the shape.

7:08So four square corners force the opposite sides to match. It was never a second demand. Which means you can strike a clause out of the description and lose absolutely nothing. Four right angles and matching opposite sides. Diagonals of equal length that cut each other in half. Or simply: four right angles. Run all three over the same sweep of shapes and they disagree about nothing at all. Zero shapes.

7:41But weaken either half of the middle one and it stops coinciding at once. Equal diagonals alone let in a hundred and seventy-five extra shapes; diagonals that merely halve each other let in two hundred. One more thing, because it is where this kind of argument is usually dropped. When you say two triangles are identical, the order you name their corners in is the claim. There are six ways to write the second triangle's three corners against the first triangle's three.

8:18Exactly one of them is the true correspondence. The other five assert that parts match which do not. And here is the trap. Try a scrambled order on a square, and it comes out true. A square is symmetric enough that the wrong pairing happens to work. On every other rectangle it fails. So checking your naming on the shape you happened to draw is the thousandth rectangle all over again.

8:48It would be neat to end by saying that proof beats measuring, and that is not what happened here. Measuring is what made anyone suspect the thing worth proving. The failed construction is what said where to look, and it pointed straight at the corner the diagonal splits. And once a thing is argued, you go back and check it against a real shape anyway. Not because the argument might be badly done, but because it might be a flawless argument about the wrong statement.

9:20A proof of something you did not mean is still a proof, and only contact with an actual object tells you. Argue it if you can, measure it if you cannot, and check it either way.