Pınar

A second angle that handles the "but what about a really big quadratic" objection head-on, which is the actual question he asked.

The coefficient — the 10 in 10 x squared, or a million if he likes — only shifts when the crossover happens. It never changes whether it happens.

Here is why, in a form he can check. Suppose the quadratic is a x squared for some enormous a. The exponential is 2 to the x. Now compare the two at x and at x plus 1:

• The quadratic goes up by a factor of (x+1)² / x², which gets closer and closer to 1 as x grows. By x = 100 it is about 1.02.
• The exponential goes up by a factor of exactly 2, forever.

So past some point the quadratic is growing by about two percent per step while the exponential is growing by a hundred percent per step. Whatever head start a huge coefficient bought, it gets eaten, because one side's growth rate is dying towards nothing and the other's is constant.

The slogan version: a big coefficient buys a delay, never a victory.

There is a formula, several in fact, and the reason nobody shows you is that they are infinite sums — and infinite sums are usually taught a few years after trigonometry.

The main one says that the sine of an angle x, with x measured in radians rather than degrees, equals:

x  -  x³/6  +  x⁵/120  -  x⁷/5040  +  ...

forever, with the pattern being odd powers, alternating signs, and the denominators being factorials. You can try it right now with a calculator: take x = 0.5, compute those four terms, and you will get 0.4794, which is correct to four decimal places already.

That is genuinely how it is done. It is not an approximation to some other true method — this sum is the sine, in the same way that the triangle definition is.

The catch is that this only converges usefully for small x, which is why the second half of the answer matters.

And yes, it can give the wrong answer. The construction is easy: take any function on the integers and pass a wiggly curve through the same points with its minimum between two of them. Rounding the continuous minimum then lands you on a point that is not the discrete minimum.

What makes the engineering cases work in practice is a property the textbook rarely names: the extension used is typically convex, or at least unimodal, over the range of interest.

For a convex function on an interval, the discrete minimum is guaranteed to be at one of the two integers bracketing the continuous minimum. That is a real theorem and it is the thing doing the work.

Which gives you the correct version of the recipe, and it is barely more effort than the one you were taught: differentiate, find the continuous optimum, then evaluate the original discrete function at both neighbouring integers and compare. Do not round — check. Rounding is a guess that happens to be right most of the time; checking two values is certain and costs one extra line.

My favourite from group theory: for symmetric groups — the groups of all rearrangements of n objects — every symmetry of the group is an inner one, meaning it comes from conjugating by an element of the group itself.

True for every n. Except n = 6, which has an extra symmetry that comes from nowhere and cannot be explained by conjugation.

There is no small reason for it. Six is not special in any way that shows up first; the exception is discovered by working through the proof and finding that one step fails in exactly one place. It has been connected to several other unique objects that also live at six, which suggests something coherent is going on, and it still reads as an accident.

A related one: the finite simple groups fall into a small number of infinite families, plus twenty-six sporadic groups that belong to no family at all. That is not one exception, but it is the same flavour — a clean classification with a bag of leftovers that stubbornly exist.

One intuition pump that helped me: try to find the optimal tour for fifteen randomly placed points by hand, on paper, and then try to prove to somebody that your answer is optimal.

Finding a good tour is easy — your eye does it. Proving no better one exists is where the difficulty lives, and doing it once by hand makes the theory feel obvious rather than abstract. That gap between "I found a good answer" and "I can rule out every better one" is the entire subject.

On whether a real proof has been dismissed: the honest answer is that the historical failure mode is different from what you are imagining.

What has genuinely happened is that unconventional work from outside the mainstream took a long time to be read, sometimes years, and was eventually recognised. That is a delay rather than a permanent dismissal, and it is a real cost of the system.

What has essentially not happened is a correct proof of a famous problem being rejected on the merits and then vanishing. The incentive to be the person who validates a landmark result is enormous, and correctness is checkable. The bottleneck is attention, not willingness.

The practical advice given to anyone who believes they have one is unchanged and reasonable: write it up properly, state precisely where the new idea enters, and show what your argument does on a simpler case that is already known. That last one is persuasive in a way that assertion never is.

The second reason is that it makes a genuinely contested philosophical position operational.

The claim that a probability can describe a degree of belief in a hypothesis — rather than a long-run frequency of an event — was and to some extent still is disputed. The theorem is what turns that claim into a procedure: start with a prior belief, observe data, get a posterior belief, repeat. Belief becomes something you can update mechanically and check.

That is why an entire school of statistics is named after it. The name does not mark the difficulty of the algebra; it marks which side of an argument you are on about what probability means. A frequentist and a Bayesian both accept the theorem and disagree about what it is legitimate to put into it.

The obvious first answer is the incompleteness results of the early twentieth century, and the disappointment there was as large as mathematics has ever produced.

The programme at the time was to put all of mathematics on a complete, consistent, mechanically checkable foundation. What was proved instead is that any system strong enough to describe ordinary arithmetic contains true statements it cannot prove, and cannot establish its own consistency.

That closed the programme permanently. People had not merely hoped for the other answer; a great deal of work had been organised around the assumption that the other answer was forthcoming.

In hindsight the disappointment was real and the consequences were enormously productive. The techniques developed to prove it founded computability theory, which is a large part of why computers have a theory at all. The letdown and the payoff are the same result.

The comparison you drew is the right one, and the answer starts one level lower than differential equations.

Consider the simplest possible differential equation: f′(x) = g(x). Its solution is the integral of g. So asking whether this equation has a closed-form solution is exactly asking whether g has an elementary antiderivative.

And we know that most elementary functions do not. There is a real theorem here rather than a failure of ingenuity — results in differential algebra establish that specific functions have no antiderivative expressible in elementary terms at all. The Gaussian bell curve is the famous example.

So the difficulty is present before you get anywhere near an interesting equation. Even the most trivial differential equation inherits the fact that elementary functions are not closed under integration.

That is the deep explanation you were looking for, and it is genuinely analogous to the polynomial story: in both cases, a natural operation takes you outside the class of expressions you started with.