8.4 Heuristics and Experimentation

A heuristic is a rule of thumb that guides search. It does not guarantee a solution, but it helps decide what to try next. Experimentation is the systematic use of examples, computations, diagrams, and small cases to discover patterns.

These methods are useful when the direct route is unclear. A problem may have no obvious theorem to apply. Its structure may be hidden. In that situation, experimentation gives data, and heuristics organize the search.

A simple heuristic is to test small cases. If a statement involves a natural number $n$, try $n=0$, $n=1$, $n=2$, and $n=3$. If it involves a graph, try paths, cycles, complete graphs, and disconnected graphs. If it involves a function, try constant, linear, polynomial, and discontinuous examples.

Small cases rarely prove the result, but they show what the statement is saying.

For example, suppose a problem asks for a formula involving

Computing small values gives

The pattern suggests the formula

The experiment suggests the theorem. A proof is still required.

Another heuristic is to look for invariants. If a process changes an object step by step, ask what remains unchanged. Invariants often turn a dynamic problem into a static one.

For example, in many puzzles and combinatorial problems, parity is the key invariant. A move may change many details while preserving whether a number is even or odd. If the target state has different parity, the target is impossible.

A third heuristic is to simplify the problem. Remove conditions. Study a special case. Replace a hard object by a simpler one. If the simplified problem can be solved, the solution may suggest how to return to the original problem.

This is not the same as ignoring difficulty. It is a way to locate it.

If the simplified version is still hard, the obstacle is fundamental. If the simplified version becomes easy, the removed condition may contain the main difficulty.

Diagrams are another form of experimentation. A geometric drawing, graph sketch, lattice picture, or commutative diagram can expose relationships that are hard to see symbolically.

The diagram should not be trusted blindly. It can suggest false conclusions when it represents only a special case. But it can guide the search for definitions, lemmas, and invariants.

Symbolic experimentation also matters. One may expand expressions, compute examples, factor polynomials, reduce matrices, or evaluate special values. These manipulations can reveal hidden form.

For instance, an expression may become simpler after substitution. A recurrence may become clearer after listing its first terms. A polynomial identity may become visible after factoring.

Computational tools extend this process. Computer algebra systems, numerical libraries, and small scripts can generate data quickly. They are useful for conjecture formation, counterexample search, and verification of finite cases.

But computation does not replace proof. Numerical evidence can be misleading because of rounding, insufficient cases, or hidden assumptions. Symbolic computation can also depend on simplification rules that must be interpreted correctly.

A good experiment records its conditions. What examples were tested? What assumptions were used? What pattern appeared? What cases were not covered?

This discipline prevents accidental overgeneralization.

Heuristics often work best in combination. One may test small cases, notice an invariant, make a conjecture, prove a special case, then generalize. The process is iterative.

A common workflow is:

Start with examples. Look for stable features. Guess a statement. Test boundary cases. Search for an invariant or transformation. Turn the pattern into a proof.

The boundary cases are especially important. They include zero objects, empty sets, identity maps, degenerate triangles, disconnected graphs, singular matrices, constant functions, and limiting values. These cases often reveal whether the proposed statement is correctly formulated.

For example, a theorem about matrices may need to exclude singular matrices. A theorem about division must exclude zero denominators. A theorem about compactness may require closedness or boundedness, depending on the setting.

Heuristic reasoning is provisional. It helps find the path, but it does not certify the destination. Once a promising route is found, the argument must be converted into proof.

The value of heuristics is practical. They reduce blind search. They help one decide where to look, what to test, and which structure might matter.

Experimentation makes mathematics empirical at the discovery stage. Proof makes it mathematical at the validation stage.