Research conjectures and proof strategies

A conjecture is a proposed mathematical statement whose truth has not been established in its intended setting. Before testing it, specify the objects, allowable parameters, and order of quantifiers. For every input there exists a bound is different from there exists one bound for every input. A theorem proved only for smooth compact data is not automatically a theorem for all distributions or arbitrary boundaries. Many apparent breakthroughs arise from changing the problem midway, solving a special case, or using a lemma whose assumptions fail exactly at the difficult point.

A universal assertion can be refuted by one valid counterexample, but a list of successful examples proves only the finite statement that was actually checked. Even exact arithmetic on a billion inputs does not exhaust an infinite domain. Computation can constitute a rigorous proof when a theorem reduces the problem to a finite exhaustive search and the search is certified correctly. Interval arithmetic, proof assistants, and independently checkable certificates can control different sources of error. Their value lies in a complete logical reduction and verification, not simply the presence of a computer.

Induction proves a statement for all natural numbers by establishing a base case and a valid step from an arbitrary case to the next. The step cannot assume the desired conclusion for unrelated inputs. An invariant instead identifies a quantity preserved by each allowed operation, ruling out states that violate it. Extremal arguments choose a smallest counterexample and exploit its minimality. Compactness, energy estimates, and contradiction methods perform other jobs. A strategy is useful only when its hypotheses and the exact conclusion it supplies fit the problem being investigated.

A good research workflow alternates examples with structural reformulation. Test degenerate cases, scaling limits, boundary points, and low dimensions; compute a quantity that the conjecture would constrain; search for an invariant or a monotonicity principle; then identify the missing lemma. If a proof uses approximation, specify the topology of convergence and which operations survive the limit. Passing a nonlinear term through a weak limit is not justified merely because the approximants solve an equation. The earlier convergence lessons provide practical tools for detecting that kind of gap before the notation grows more elaborate.

A conditional theorem says that a conclusion follows if a named conjecture or hypothesis holds. It may be a fully rigorous implication without establishing that hypothesis. Distinguish this from a heuristic argument that proposes a mechanism and from a numerical experiment that observes finite behavior. As checked against the Clay Mathematics Institute's problem page in October 2026, the Riemann hypothesis remains open: the asserted real part one half concerns the nontrivial zeros of the analytically continued zeta function. Numerical agreement for many zeros is evidence, not an argument covering all of them.

When reading a claimed proof, isolate its main proposition and trace every dependency until reaching established results with matching hypotheses. Recompute a small nontrivial example, identify where infinitude or a singularity enters, and ask whether the argument gives an explicit bound or only suggests one. The quadratic lab illustrates this habit: a stable-looking iteration at one step size cannot establish convergence for every step size, and the exact recurrence supplies the missing condition. It is a teaching example of theorem testing, not an automated conjecture prover or evidence about the Riemann hypothesis.