Olympiad mathematics trains solving. Research mathematics runs on asking. These are different cognitive operations, the first is trained intensively and the second is not trained at all, and there is no stage in the pathway where anybody switches you over.

I wrote earlier this year about the differences between olympiad math and research math. That post described the gap. This one is about a specific part of it that I have thought about more since, partly because of how the selection pathway is built.

The guarantee

An olympiad problem carries a guarantee. Someone constructed it, verified a solution exists, calibrated its difficulty, and put it in front of you with a time limit. That guarantee is invisible while you are inside the system and it is doing enormous work.

Because a solution exists, being stuck is information. It means you have not found the right idea yet, so you keep going, try another decomposition, look for the standard technique this is disguising. Persistence is always correct. Giving up is always wrong.

Remove the guarantee and persistence stops being unambiguously correct. Being stuck on a research question could mean you have not found the idea. It could also mean the statement is false, or true but not provable with available tools, or true and provable but not by any route a person will find this decade, or not actually well posed. Three of those five are reasons to stop.

Nothing in olympiad training gives you any practice distinguishing them, because the environment guarantees you are always in case one.

Problem posing is the actual skill

Here is what I did not understand until I tried to do research.

In competition, the problem is the input. Your entire contribution is the solution. In research, choosing the problem is most of the contribution. A well-chosen question that is tractable, non-trivial and connected to something people care about is worth more than a clever solution to a badly chosen one, and by a wide margin.

The skills involved in choosing are almost entirely absent from competition training. Reading enough of a field to know what is open and what is folklore. Judging whether a question is hard because it is deep or hard because it is arbitrary. Estimating how long an attack will take before committing. Recognising that a question is malformed and reformulating it. Knowing when a smaller question is the right first move.

Every one of these is a judgment made under uncertainty with no verification available. Competition offers no uncertainty and immediate verification.

Timescale changes the method

An olympiad problem takes between thirty minutes and a few hours. You hold the whole thing in your head. You try approaches, discard them, and the discarded ones cost you nothing because you never wrote them down.

A research problem runs for months. You cannot hold it in your head, which means you have to externalise it, which means note-taking stops being optional and becomes the primary technical skill. What did I try in March? Why did I abandon it? Was it wrong or did I just get tired?

I have lost real time to re-deriving something I had already worked out and could not read back. Nobody warns you about this because in competition there is nothing to keep.

The other timescale effect is emotional. A hard olympiad problem produces frustration for two hours. A research problem produces it for six months, and the ability to remain functional under that is a different capacity from the one that gets you through a timed paper. It is closer to what endurance athletes describe than to what competitors describe.

What does transfer

I want to be careful not to overstate this, because the transfer is real and substantial.

Technical fluency transfers completely. Someone who has done serious olympiad work can manipulate inequalities, construct arguments, and handle combinatorial structures at a level that saves years later. That fluency is not decorative and I would not trade it.

Rigour transfers. Knowing what constitutes a proof, feeling the difference between an argument and a plausibility story, noticing when a step is being waved through. Competition trains that hard and it is exactly what research needs.

The habit of not accepting a result until you understand why it is true transfers, and it is rarer than it should be.

Decomposition transfers. Breaking a hard configuration into known lemmas is the same operation at both scales.

So the picture is not that olympiad training is useless preparation. It is that it prepares the second half of the job completely and the first half not at all.

Why nothing bridges the gap

The structural reason is that the two activities are run by different institutions with different aims.

Olympiad training is delivered by programmes optimising for competition performance, which is the correct thing for them to optimise. Research training happens in graduate school, which typically starts around age 21 or 22. Between the last olympiad and the first research problem there is a gap of three or four years occupied by undergraduate coursework, which is mostly more solving of problems that have guaranteed answers.

The first time many people encounter an unguaranteed question is when a doctoral supervisor hands them one, and the transition is often described as the hardest part of the degree. That is a system design outcome, not a fact about the difficulty of research.

Nobody owns the handover, so nobody has built it. A gap of that shape, well defined, structurally unowned, and sitting directly on top of a talent pool the country has already paid to train, is not a complaint. It is a specification for something somebody should build.

What I would do about it

If you are in this position, and I partly am, the useful moves are available now and do not need anyone’s permission.

Read papers before you can fully follow them. Not textbooks. Papers, including the introduction and the open problems section, which is where authors say plainly what they could not do. Reading the frontier is how you learn what unsolved looks like from the outside.

Try a question with no known answer. A small one. Something you noticed. Most attempts will fail and the failure is the training, because it is the first time being stuck has been genuinely ambiguous.

Write down the questions. Keep a file. Most will be trivial or malformed and you will only find out by writing them, and the discrimination improves with volume.

Keep the record. Everything tried, everything abandoned, and why. This is the habit competition never asked for and research cannot proceed without.

Talk to people doing it. Problem selection is largely tacit knowledge transmitted in conversation, which is a real argument for the value of a research group and a real disadvantage for anyone working alone.

The competition experience is a genuine asset. It is also half the preparation, the pathway will not tell you which half is missing, and nobody is coming to close the gap on your behalf. That part is yours to do.

Related: