Must the point be chosen randomly?

No. Random choice is common, but other policies are possible.

Can the cut fall inside an encoded number?

In bit-level encodings it can, which changes the decoded value and must be understood.

Is crossover the same as mutation?

No. Crossover recombines parental material; mutation changes material according to a separate rule.

AI Glossary / Graphs and symbolic AI / Intermediate

Crossover point

Definition

A crossover point is a position in a candidate solution's representation where a crossover operation switches from material inherited from one parent to material inherited from another. In a single-point genetic algorithm crossover, one cut divides two parent strings into prefixes and suffixes.

What it means

Genetic algorithms search using a population of candidate solutions. After selecting parents, they can recombine parts of their representations to create offspring. The crossover point determines which adjacent positions remain together and which segments are exchanged.

It is a location in the encoding, not a guarantee of improvement or a place where the underlying business problem naturally divides. If useful decisions depend on one another, cutting between them can disrupt a beneficial combination. If parents contain complementary useful segments, the same operation can create a better candidate.

Different crossover operators use different notions of a cut. Two-point crossover exchanges an interval between two positions. Uniform crossover chooses a parent separately at each position and therefore does not rely on one continuous switching boundary.

The representation matters. A bit string, a sequence of real-valued parameters and a permutation of tasks have different validity rules. An operator appropriate for one may create invalid offspring in another.

How it works

For equal-length strings of length L, a common single-point convention chooses a cut k between 1 and L - 1. One child takes the first k positions from parent A and the remainder from parent B. The other child takes the complementary combination. Some implementations allow other endpoint conventions, so the indexing rule should be stated.

The cut is often sampled randomly, but crossover probability and crossover location are different settings. Crossover probability determines whether the operator is applied; the point determines where an applied operation divides the representation.

After recombination, offspring are evaluated and may undergo mutation. Mutation changes selected values rather than merely exchanging parental segments. Selection and replacement rules determine which candidates remain in the population.

A crossover point does not ensure feasibility. If the encoding represents a schedule with unique tasks, exchanging arbitrary suffixes can duplicate one task and omit another. Specialized permutation operators, repair procedures or constraint-aware encodings may be needed.

Reproducible experiments should record random seeds, operator conventions and the full evaluation budget. A favorable child from one cut is an illustration, not evidence that an operator consistently outperforms alternatives.

Three practical examples

A teaching genetic algorithm exchanges sections of binary strings and shows how the cut affects the resulting candidates.
A parameter-search prototype recombines encoded configuration choices, then validates each child against compatibility restrictions.
A scheduling experiment uses a permutation-aware crossover method because a naive cut could repeat or lose tasks.

Think of it this way

Think of joining the beginning of one draft itinerary to the end of another. The join position determines which sections are inherited from each plan.

The analogy also reveals the limitation: the joined itinerary may visit the same place twice or skip a required destination. Combining valid parents does not automatically create a valid or better child.

Worked example

A cut that helps, and a cut that breaks validity

Let parent A be 110010 and parent B be 001101. Cut after the third position. A contributes prefix 110 and B contributes suffix 101, producing child 110101. The complementary child is 001010.

Under the illustrative fitness function count the number of ones, each parent scores 3. The first child scores 4 and the second scores 2. The same crossover event therefore produces one better and one worse candidate under this particular objective.

Changing the objective could change that assessment. If adjacent equal bits are rewarded instead, a count of ones is no longer an adequate fitness calculation.

Now use task permutations A = [A, B, C, D] and B = [C, D, A, B]. A naive cut after two positions creates [A, B, A, B] and [C, D, C, D]. Both parents contain every task once, but neither child does. The defect arises from applying a string-style operator without respecting the permutation representation.

Illustrative case study

Fictional case: Fern Scheduling introduces crossover to combine candidate workshop schedules. Its first demonstration highlights an offspring with a better score but does not validate task coverage.

A review finds duplicated sessions created at the cut boundary. The team adopts a representation-aware operator, checks feasibility before comparing objective values and reports results across repeated runs with equal evaluation budgets.

The lesson is that recombination quality includes preserving the meaning of the encoding. A numerical fitness improvement is irrelevant if the candidate no longer represents an admissible plan.

Common misconceptions and mistakes

The crossover point is the probability of crossover. One is a location; the other controls whether crossover happens.
Offspring from good parents must be better. Recombination can help, harm or leave fitness unchanged.
A valid parent pair guarantees valid children. Representation constraints can be broken.
All crossover methods use one point. Multi-point, uniform and representation-specific operators differ.
Crossover alone supplies a complete optimizer. Fitness, selection, mutation and replacement policies also matter.

Business use

Understanding crossover points helps teams evaluate evolutionary optimization claims and diagnose why an encoding behaves poorly. It encourages attention to which decisions are kept together and which combinations are disrupted.

Choose operators that preserve or explicitly repair the problem's constraints. Compare methods using the same evaluation budget and include unsuccessful runs rather than reporting only the best offspring. Record whether the goal is a feasible solution, a high-scoring solution or a provably optimal one; a standard genetic algorithm does not provide an optimality proof merely by applying crossover.

Check your understanding

With parents 111000 and 000111 and a cut after position 3, what are the two complementary children?

Answer: 111111 and 000000, assuming the stated prefix-suffix exchange convention. Their fitness depends on the objective.

Related terms

Genetic algorithms
Recombination
Mutation
Fitness function
Selection
Genetic algorithm
chromosome encoding
single-point crossover
permutation encoding

Learn next

Repeat the worked bit-string crossover at every interior cut position and calculate the count-of-ones scores. Then try the same cuts on the task permutations and identify which offspring violate uniqueness. Use the contrast to explain why operator choice depends on representation.

Sources

Carnegie Mellon course material on genetic algorithms, including single-point crossover: https://www.cs.cmu.edu/~02317/slides/lec_8.pdf

The bit-string scores, permutation counterexample and fictional scheduling review are original illustrations. They demonstrate operator behaviour rather than establish comparative optimisation performance.

Previous: Cross-validation Next: Crossover study