I once watched a promising student spend an entire term chasing a beautiful conjecture, only to discover while reading a paper he really should have found in his first week that his whole approach had already been ruled out, decades earlier, by someone working in a completely different tradition. Nothing about his mathematics was weak. He simply hadn't been taught that choosing a problem is its own skill, separate from solving one.

Most advice on this topic starts with inspiration: read widely, follow your curiosity, talk to people. That's fine as far as it goes. But it answers the wrong question. Finding something interesting was never the hard part. Working out which interesting thing could actually become a finished piece of research, in the time you have, is.

Interesting and Researchable Are Not the Same Thing

Interest costs nothing. Sit with almost any area of mathematics for an afternoon and something will grab you a simply stated conjecture that refuses to fall, a pattern nobody's quite explained. The Collatz conjecture does this to people constantly. So does the idea of finally settling something about prime gaps. That pull is genuine, and for nearly everyone reading this, it's also a trap.

A question earns the word "researchable" when three things line up, and they rarely line up on the first try. The scope has to fit inside the time you actually have. It has to lean on tools you know or can realistically pick up along the way, not an entire second degree's worth of theory. And whatever you find, even a disappointing or partial answer, has to be genuinely new rather than a tidy restatement of what's already written down.

Most wasted months trace back to this gap, not to any shortage of ability. The exciting version of a question and the workable version of it are frequently two different objects wearing the same name.

Test the Smallest Case Before You Trust the Big One

There's a strong pull toward questions that sound significant. A better habit, and one worth building deliberately, is to test whether you can verify something small before you commit to anything large. Can you take the smallest, most stripped-down version of your question and work through it by hand, arriving at an answer you'd actually stake your name on?

This feels like a detour. It isn't. Research mathematics has far less to do with generating ideas than with testing them against something concrete until the fragile ones fall away. If you can't build even a toy example of your own question, that's usually the clearest sign you don't yet understand it as well as you assumed.

A practical way to check: take the general statement and strip it down smallest case, plainest structure, most degenerate example you can imagine. If that smallest case resists a genuine afternoon's effort, treat that as useful information rather than a personal failing. It usually means the full question isn't within reach yet, whatever the surrounding theory promises.

Narrowing a Broad Interest Into a Real Question

A general pull toward, say, number theory or graph theory is a starting point, not a direction. Getting from one to the other means a series of deliberate restrictions, and it can feel oddly uncomfortable the first time, as though you're trading ambition for something smaller.

Start with one specific object rather than a whole family of them not "properties of graphs" but one particular construction with a name and a definition you could write on a single page. A specific object has a finite list of properties you can actually check. A general class has thousands, and you'll drown trying to survey them.

From there, find out exactly where the known results for that object stop. This edge sits far closer than most textbooks let on, because textbooks present settled theory as though nothing else exists beyond the final chapter. Reading two or three recent papers that argue with one another citing, correcting, extending each other's work shows you the real boundary of a field faster than any amount of lecture notes, because papers disagree about what's still open in a way textbooks simply aren't built to do.

Then try loosening one condition slightly. Relax a hypothesis, add a dimension, remove a symmetry. Small, deliberate variations on established results make up a genuinely large share of published mathematics not the headline share, but the finishable share, which is the one that matters when you're the person who has to actually finish it. This is exactly the gap where solid mathematics research topics tend to live: not in famous unsolved questions, but in the quiet, well-defined space just past where someone else stopped.

A Rough Test for Whether an Idea Will Hold Up

Once you've got a candidate, push it a step further than "this sounds sensible." Try to state, in one or two plain sentences, what a positive answer would look like and separately, what a negative one would look like. If only one of those pictures comes easily, the question probably isn't sharp enough yet. It's a topic dressed up as a problem.

It also helps to ask your supervisor a sharper question than the usual one. Rather than "is this a good topic," try "has anyone tried this and failed, and do you know why." That second question tends to surface the real obstacles, the ones that never make it into a published paper because a failed attempt mostly just lives in someone's memory.

Worth noticing too: would you genuinely be satisfied with a negative result? Plenty of people unconsciously pick questions where they've already decided what the answer should be, so any evidence pointing elsewhere starts to feel like a personal mistake rather than a legitimate finding. If "no" would feel like failure rather than data, that's worth sitting with before you go much further.

Knowing When You're Stuck Versus When You're Wrong

There are two kinds of stuck, and confusing them is costly. One is ordinary technical difficulty you can see roughly where you're heading, even if the next step is genuinely hard. Keep going through that kind.

The other is not knowing what tool would even begin to address the question, where every path seems to demand an entirely different subfield. That's a different signal entirely, and it usually points back to how the question was framed in the first place, not to any gap in your ability. The right move is to narrow further or shift the angle, not to conclude, somewhere around week three, that the subject simply isn't for you.

A quieter warning sign: if explaining your problem to someone else keeps needing more and more background before it lands, the question is probably sitting too far from your current foundation, however good the eventual payoff might be.

What Actually Changes Once You Do This Properly

The real shift is about what you optimise for. Early on, most people optimise for the question that sounds most exciting to describe out loud. It's more useful, if less immediately satisfying, to optimise for the smallest question whose answer you genuinely don't know and can't easily guess. That's the one where the work teaches you something real, and where finishing is an actual possibility rather than a hope.

None of this means staying small forever. Real breadth comes later, once you've built a track record of finished, specific results and developed something closer to instinct earned slowly, through repetition, not handed down as advice for which bigger questions are within reach.

Choosing well isn't a step before the real work begins. It is the work, done early, and it's probably the single decision that decides whether the months ahead feel like progress or like walking the same unanswered question round in a circle.