Non-Algorithmic Reasoning and the Future of Mathematics Learning

Last Updated May 30, 2026

Mathematics is often taught as if learning means mastering procedures: solve the equation, apply the formula, follow the algorithm, compute the answer. Procedures matter. Algorithms matter. Fluency matters. But mathematical thinking is larger than procedure. Many of the most important mathematical acts are non-algorithmic: noticing structure, choosing a representation, forming a conjecture, deciding what to prove, recognizing a hidden assumption, inventing a counterexample, interpreting a result, or knowing when a method does not apply.

Non-algorithmic reasoning is not anti-algorithmic. It does not reject computation, symbolic manipulation, procedural fluency, or formal methods. Instead, it asks what must happen before, around, and after algorithms. Which problem is being solved? Which structure matters? Which assumptions are active? Which representation is useful? Which result is meaningful? Which proof is needed? Which answer is plausible? Which conclusion is responsible?

This article examines non-algorithmic reasoning as a foundation for the future of mathematics learning. It explores conceptual understanding, mathematical judgment, problem framing, representation, conjecture, proof, metacognition, creativity, AI-assisted learning, computational tools, Haskell-style structural modeling, and the educational challenge of helping students become mathematical thinkers rather than only procedural performers.

What Non-Algorithmic Reasoning Means

Non-algorithmic reasoning refers to forms of mathematical thought that cannot be reduced to following a fixed sequence of steps. It includes choosing a strategy, interpreting a problem, recognizing structure, generating examples, constructing a representation, forming a conjecture, deciding what matters, evaluating plausibility, building a proof, and reflecting on whether an answer makes sense.

An algorithm is a procedure: a finite, rule-governed sequence of operations designed to produce a result. Algorithms are essential in mathematics. Long division, Gaussian elimination, the Euclidean algorithm, gradient descent, numerical integration, graph search, sorting, symbolic simplification, and proof-checking procedures all matter. But mathematical reasoning includes decisions about when an algorithm applies, what its output means, and whether the underlying formulation is appropriate.

Non-algorithmic reasoning is visible whenever a learner asks questions such as:

• What kind of problem is this?
• What information is relevant?
• What assumptions are hidden?
• Which representation should I use?
• Does this pattern continue?
• Can I find a counterexample?
• Why does this method work?
• Does the answer make sense?
• Can this result be generalized?

These questions are not peripheral. They are central to mathematical maturity. A student who can compute but cannot interpret, justify, or adapt remains dependent on familiar templates. A student who can reason non-algorithmically can use procedures as tools rather than as scripts.

Non-algorithmic reasoning is the part of mathematical learning that helps students become flexible, critical, creative, and responsible. It is the difference between knowing a method and knowing how to think mathematically with methods.

Back to top ↑

Algorithmic Fluency and Its Limits

Algorithmic fluency is valuable. Students need procedures. They need to manipulate expressions, solve equations, compute derivatives, perform matrix operations, use statistical formulas, write code, and execute formal methods. Procedural fluency reduces cognitive load and allows attention to move toward structure, interpretation, and proof.

The problem arises when mathematics is reduced to procedure alone. A learner may know how to complete a familiar exercise without understanding why the method works, when it applies, or what the result means. This is especially common when classroom tasks are highly patterned: identify the formula, substitute values, compute, and report an answer.

Algorithmic fluency has limits because many important mathematical problems do not announce which algorithm to use. Real mathematical work often begins with ambiguity. The learner must determine the object, representation, assumptions, and goal before procedure becomes useful. Even in formal mathematics, proof is rarely a matter of blindly applying a known algorithm. One must decide what to prove, what to assume, what examples suggest, which lemmas are relevant, and where a proof strategy might fail.

The future of mathematics learning should not abandon algorithms. It should place algorithms inside a richer ecology of reasoning. Students should learn procedures, but they should also learn why the procedures work, what they assume, how they connect to structure, and when they fail.

Back to top ↑

Mathematical Judgment Beyond Procedure

Mathematical judgment is the ability to decide what matters in a mathematical situation. It includes selecting a strategy, recognizing a relevant structure, identifying hidden assumptions, deciding whether a result is plausible, and knowing when more justification is needed. Judgment is not the opposite of rigor. It is how rigor is applied intelligently.

A student may know several algorithms and still lack mathematical judgment. For example, they may apply a formula outside its domain, solve an equation by squaring both sides without checking extraneous solutions, trust a calculator output without considering scale, or treat a finite pattern as proof. The issue is not procedural ignorance. The issue is lack of judgment about meaning, assumptions, and validity.

Judgment develops through tasks that require explanation, comparison, error analysis, counterexample search, representation choice, and proof. It also develops when students encounter problems that do not fit a single obvious template. Productive struggle matters because it teaches students to make mathematical decisions.

Mathematical judgment is difficult to teach through answer-only exercises. It requires visible reasoning. Students need opportunities to explain why a method applies, compare approaches, diagnose mistakes, revise conjectures, and defend conclusions.

Back to top ↑

Conceptual Understanding and Relational Knowledge

Conceptual understanding means knowing how mathematical ideas relate. It is different from merely remembering a rule. A student with conceptual understanding knows why a procedure works, how it connects to definitions, what it preserves, what assumptions it uses, and how it can be adapted.

For example, solving \(2x+3=11\) is not only a sequence of algebraic moves. It is an exercise in preserving equality. Factoring is not only a mechanical pattern; it reveals multiplicative structure. A derivative is not only a formula to compute; it represents local rate of change, linear approximation, and limiting behavior. A matrix is not only an array of numbers; it may represent a linear transformation, a system of equations, a graph, or data.

Relational knowledge supports transfer. A student who knows only a template may struggle when the problem changes form. A student who understands the underlying structure can recognize the same idea in a new representation. This is essential in advanced mathematics, modeling, data science, engineering, economics, AI, and scientific reasoning.

The future of mathematics learning should make conceptual relationships visible. Students should encounter multiple representations, compare methods, explain reasoning, and learn to move between procedure and meaning.

Back to top ↑

Problem Framing Before Problem Solving

Problem solving begins before calculation. A learner must frame the problem: identify what is being asked, what information is given, what assumptions are implied, what constraints apply, what representation may help, and what kind of answer is expected. This framing is deeply non-algorithmic.

Many students struggle not because they cannot execute procedures, but because they do not know which procedure belongs to the situation. In real mathematical work, the method is often not announced. The learner must infer structure.

Problem framing includes translating language into mathematics, deciding what to ignore, choosing variables, defining units, identifying domains, and recognizing whether the task is asking for computation, proof, explanation, estimation, optimization, classification, or modeling.

A mathematics classroom oriented toward the future should ask students not only to solve framed problems, but to participate in the framing. This helps them become thinkers who can approach unfamiliar situations rather than only performers of familiar scripts.

Back to top ↑

Representation Choice as Mathematical Reasoning

Choosing a representation is a form of mathematical reasoning. A table, graph, equation, diagram, matrix, simulation, proof tree, or formal language each makes different structures visible. The same problem can become easier or harder depending on how it is represented.

Representation choice is non-algorithmic because there is rarely one automatic answer. A learner must decide what needs to be seen. If the issue is change, a graph or derivative may help. If the issue is dependency, a directed graph may help. If the issue is repeated structure, a recurrence may help. If the issue is proof, definitions and logical form may help. If the issue is computation, data structures may help.

Students often treat representations as teacher-provided formats. A stronger mathematics education invites students to ask which representation they need and why. This develops flexibility, abstraction, and interpretive skill.

The future of mathematics learning should treat representation as a decision, not merely a format. Students should learn to justify why a representation is appropriate and what it preserves or omits.

Back to top ↑

Conjecture, Exploration, and Productive Uncertainty

Non-algorithmic reasoning thrives in productive uncertainty. A conjecture is not yet proved. A pattern is not yet explained. A diagram suggests but does not establish. A computational result is promising but not decisive. This uncertain space is where mathematical discovery begins.

Traditional mathematics instruction often moves too quickly from problem statement to method. Students may rarely experience the stage where one must explore examples, formulate a conjecture, test special cases, and revise. Yet this stage is central to authentic mathematical thinking.

Productive uncertainty does not mean confusion without support. It means designing learning environments where students have enough structure to explore, but enough openness to make mathematical decisions. They should be asked to notice patterns, propose rules, find counterexamples, explain why a conjecture might be true, and decide what would count as proof.

Conjecture-based learning helps students see that mathematics is not only about having the right answer. It is about learning how claims become trustworthy.

Back to top ↑

Proof as Non-Algorithmic Justification

Proof is often taught as if it were a formal template: assume, manipulate, conclude. But proof is deeply non-algorithmic. The hard part is often not writing a final proof, but finding why the statement should be true and what structure can justify it.

Proof requires decisions. Which definitions matter? Should one prove the contrapositive? Is induction appropriate? Would contradiction expose an impossibility? Is a constructive witness needed? Is there an invariant? What lemmas are missing? Are there boundary cases? Does the conclusion actually follow?

Students who see proof only as formal writing may miss its exploratory dimension. Proof begins when one asks what would make the claim necessary. This often involves examples, failed attempts, diagrams, analogies, and counterexamples. The finished proof may be linear, but the discovery of the proof rarely is.

Teaching proof as non-algorithmic reasoning means teaching students to search for structure, not merely to imitate proof formats. The goal is not only that students can read proofs, but that they can ask what a proof must explain.

Back to top ↑

Metacognition, Strategy, and Mathematical Self-Monitoring

Metacognition is thinking about one’s own thinking. In mathematics, it includes monitoring progress, recognizing confusion, checking assumptions, evaluating strategies, and deciding when to change approach. It is one of the most important forms of non-algorithmic reasoning.

A student using metacognition may ask: Do I understand the problem? Is this method working? Did I use all the information? Did I introduce an assumption? Is the answer reasonable? Can I explain why this step is valid? Should I try a smaller case? Should I draw a diagram? Should I look for a counterexample?

Metacognition is especially important in unfamiliar problems. A student following a memorized procedure may not know what to do when the procedure fails. A student with strategic awareness can diagnose the difficulty and try a different representation, example, or proof method.

The future of mathematics learning should make metacognition explicit. Students should not only show work; they should learn to explain why they chose a path, where they got stuck, how they checked their answer, and what they would try next.

Back to top ↑

Creativity, Insight, and Mathematical Imagination

Mathematical creativity is the ability to see possible structure before it is fully justified. It appears when a learner invents a representation, notices a hidden pattern, finds a counterexample, asks a better question, connects two ideas, or discovers a proof strategy. Creativity is not separate from rigor; it is what gives rigor something to work on.

Non-algorithmic reasoning is where mathematical imagination lives. Algorithms execute known procedures. Creativity asks what procedure, representation, theorem, or definition might be needed. It works in the space of uncertainty, analogy, pattern, and possibility.

Mathematics learning should therefore make room for invention. Students should be asked to create examples, invent notation, compare strategies, build models, generate conjectures, and explain why one representation is better than another. These activities help students experience mathematics as a living discipline rather than a catalog of settled procedures.

Mathematical creativity is not a luxury reserved for advanced students. It is part of how students learn to own mathematical ideas. A future-oriented mathematics education should cultivate creative reasoning from the beginning.

Back to top ↑

AI, Automation, and the Future of Learning

AI and computational tools change the meaning of mathematics learning. When software can solve equations, graph functions, compute derivatives, generate examples, write code, suggest proofs, and produce explanations, education cannot depend only on tasks that machines can complete mechanically. Students still need procedural fluency, but they also need stronger judgment, interpretation, verification, and conceptual reasoning.

The central educational question is not whether students should use tools. Mathematics has always used tools: numerals, notation, diagrams, tables, slide rules, calculators, computers, symbolic algebra systems, proof assistants, and programming languages. The question is what kind of thinking tools should support.

AI makes non-algorithmic reasoning more important, not less. If a tool can perform routine steps, students must become better at deciding what to ask, how to check, what assumptions are present, whether the answer is plausible, and whether the explanation is valid. In a tool-rich environment, mathematical literacy includes verification literacy.

The future of mathematics learning should not define success as outperforming machines at mechanical tasks. It should define success as learning to reason mathematically with and beyond tools: to frame problems, choose representations, interpret outputs, verify claims, and justify conclusions.

Back to top ↑

Assessment Beyond Correct Answers

If mathematics learning values non-algorithmic reasoning, assessment must change. Correct answers matter, but they are not enough. A student may get the right answer by memorization, guessing, pattern-matching, or tool use without understanding. Another student may make a computational error while demonstrating strong reasoning, representation, and strategy.

Assessment should make reasoning visible. It should ask students to explain, justify, compare, estimate, critique, revise, and generalize. It should include tasks where the method is not obvious and where students must choose representations or evaluate claims.

Assessing non-algorithmic reasoning does not mean abandoning rigor. It means assessing the reasoning that makes rigor meaningful. Students should be evaluated not only on whether they can produce an answer, but on whether they understand, justify, and interpret the mathematics behind it.

Back to top ↑

A Mathematical Lens: Notice, Frame, Represent, Reason, Justify, Reflect

A useful lens for future mathematics learning is the sequence: notice, frame, represent, reason, justify, reflect. This sequence captures the non-algorithmic dimensions of mathematical thought that surround procedural work.

Noticing identifies a pattern or problem. Framing clarifies what is being asked. Representation gives the structure a usable form. Reasoning develops a path forward. Justification establishes why the conclusion follows. Reflection turns the experience into transferable knowledge.

This lens also helps integrate technology responsibly. A calculator, CAS, proof assistant, or AI system can assist at several stages, but it should not replace the learner’s responsibility to frame, interpret, justify, and reflect.

Back to top ↑

Computational Companion Examples

The companion repository for this article should extend the Mathematical Thinking codebase with examples focused on non-algorithmic reasoning metadata, strategy selection, representation-choice audits, misconception diagnosis, proof-status tracking, AI-output verification, Haskell algebraic data types for reasoning moves, and computational examples that distinguish procedure execution from mathematical judgment. The examples below are compact article-level previews; the repository can expand them into richer professional workflows.

Python: Reasoning Move and Strategy Audit

from dataclasses import dataclass

@dataclass(frozen=True)
class ReasoningMove:
    move_id: str
    stage: str
    question: str
    mathematical_role: str

moves = [
    ReasoningMove(
        "notice_pattern",
        "notice",
        "What pattern or structure appears?",
        "initiates conjecture and exploration"
    ),
    ReasoningMove(
        "choose_representation",
        "represent",
        "Which representation makes the structure visible?",
        "supports strategic flexibility"
    ),
    ReasoningMove(
        "check_assumptions",
        "justify",
        "What assumptions does this method require?",
        "prevents invalid procedure use"
    ),
    ReasoningMove(
        "search_counterexample",
        "reason",
        "Can the claim fail in a boundary case?",
        "tests generality"
    ),
]

def audit_solution(method: str, assumptions_checked: bool, interpretation_given: bool) -> dict:
    return {
        "method": method,
        "algorithmic_step_present": True,
        "assumptions_checked": assumptions_checked,
        "interpretation_given": interpretation_given,
        "reasoning_quality": "strong" if assumptions_checked and interpretation_given else "incomplete"
    }

print(moves)
print(audit_solution("quadratic formula", assumptions_checked=True, interpretation_given=False))

R: Assessment Rubric for Non-Algorithmic Reasoning

rubric <- data.frame(
  dimension = c(
    "problem framing",
    "representation choice",
    "strategy explanation",
    "assumption checking",
    "justification",
    "reflection"
  ),
  question = c(
    "Did the learner identify what is being asked?",
    "Did the learner choose and explain a useful representation?",
    "Did the learner explain why the strategy applies?",
    "Did the learner identify assumptions and domains?",
    "Did the learner justify the conclusion?",
    "Did the learner reflect on transfer or limitations?"
  ),
  max_score = c(3, 3, 3, 3, 4, 2)
)

sample_scores <- data.frame(
  student_id = c("A", "B", "C"),
  problem_framing = c(3, 1, 2),
  representation_choice = c(3, 2, 1),
  strategy_explanation = c(2, 1, 3),
  assumption_checking = c(2, 0, 2),
  justification = c(3, 1, 3),
  reflection = c(2, 0, 1)
)

sample_scores$total <- rowSums(sample_scores[, -1])

print(rubric)
print(sample_scores)

Julia: Procedure Output vs Plausibility Check

function quadratic_formula(a, b, c)
    discriminant = b^2 - 4a*c
    if discriminant < 0
        return nothing
    end
    return ((-b + sqrt(discriminant)) / (2a),
            (-b - sqrt(discriminant)) / (2a))
end

function verify_roots(a, b, c, roots)
    if roots === nothing
        return "no real roots"
    end
    return [a*r^2 + b*r + c for r in roots]
end

roots = quadratic_formula(1.0, -5.0, 6.0)

println("Computed roots: ", roots)
println("Residual checks: ", verify_roots(1.0, -5.0, 6.0, roots))
println("Interpretation: procedure gives candidates; verification checks whether they satisfy the equation.")

Haskell: Reasoning Moves as Algebraic Data Types

{-# OPTIONS_GHC -Wall #-}

data ReasoningStage
  = Notice
  | Frame
  | Represent
  | Reason
  | Justify
  | Reflect
  deriving (Eq, Show)

data ReasoningMove = ReasoningMove
  { moveId :: String
  , stage :: ReasoningStage
  , guidingQuestion :: String
  , mathematicalRole :: String
  } deriving (Eq, Show)

data SolutionAudit = SolutionAudit
  { methodName :: String
  , assumptionsChecked :: Bool
  , interpretationGiven :: Bool
  , justificationGiven :: Bool
  } deriving (Eq, Show)

quality :: SolutionAudit -> String
quality audit
  | assumptionsChecked audit && interpretationGiven audit && justificationGiven audit = "strong reasoning"
  | justificationGiven audit = "partially justified"
  | otherwise = "procedural or incomplete"

main :: IO ()
main = do
  let move = ReasoningMove
        "choose_representation"
        Represent
        "Which representation makes the structure visible?"
        "supports flexible mathematical reasoning"

  let audit = SolutionAudit
        "quadratic formula"
        True
        False
        True

  print move
  putStrLn (quality audit)

SQL: Mathematics Learning and Reasoning Metadata

CREATE TABLE reasoning_move (
  move_id TEXT PRIMARY KEY,
  stage TEXT NOT NULL,
  guiding_question TEXT NOT NULL,
  mathematical_role TEXT NOT NULL
);

CREATE TABLE learning_task (
  task_id TEXT PRIMARY KEY,
  title TEXT NOT NULL,
  task_type TEXT NOT NULL,
  algorithmic_component TEXT NOT NULL,
  non_algorithmic_component TEXT NOT NULL
);

CREATE TABLE assessment_dimension (
  dimension_id TEXT PRIMARY KEY,
  name TEXT NOT NULL,
  description TEXT NOT NULL,
  evidence_of_learning TEXT NOT NULL
);

CREATE TABLE solution_audit (
  audit_id TEXT PRIMARY KEY,
  task_id TEXT NOT NULL,
  method_used TEXT NOT NULL,
  assumptions_checked INTEGER NOT NULL,
  interpretation_given INTEGER NOT NULL,
  justification_given INTEGER NOT NULL,
  reflection_given INTEGER NOT NULL,
  FOREIGN KEY (task_id) REFERENCES learning_task(task_id)
);

These examples treat mathematical learning as something that can be represented beyond answer checking. Reasoning moves, strategy choices, assumptions, verification, interpretation, and reflection can be documented as structured artifacts. The goal is not to reduce learning to metadata, but to make the non-algorithmic dimensions of mathematical thinking visible.

Back to top ↑

GitHub Repository

The companion repository for this article is designed as a reproducible mathematical-thinking workspace focused on non-algorithmic reasoning, strategy selection, representation choice, proof-status discipline, assessment metadata, misconception audits, Haskell algebraic data types, AI-output verification, and computational workflows for the future of mathematics learning.

Back to top ↑

Equity, Agency, and Responsible Mathematics Education

Non-algorithmic reasoning is also an equity issue. If mathematics is taught only as rule following, students who do not immediately recognize the template may be labeled weak even when they have strong curiosity, insight, spatial reasoning, verbal reasoning, or conceptual potential. A narrow procedural model can hide mathematical intelligence.

At the same time, conceptual and exploratory teaching must not become vague or inaccessible. Students need explicit support, clear expectations, worked examples, language development, practice, feedback, and opportunities to build fluency. Non-algorithmic reasoning is not an excuse to abandon structure. It requires better structure: tasks that help students learn how to think, not only what to compute.

AI and automated tools add another ethical layer. Students with access to powerful tools may appear more fluent than they are. Students without guidance may become dependent on generated answers. Teachers may be pressured to assess easily graded outputs rather than deeper reasoning. Institutions may mistake automation for learning.

A responsible future for mathematics education should expand mathematical agency. Students should learn that they are not merely executing instructions; they are interpreting, questioning, representing, proving, modeling, and making sense. That agency is essential in a world where mathematical tools are powerful and increasingly automated.

Back to top ↑

Why Non-Algorithmic Reasoning Matters

Non-algorithmic reasoning matters because the future will not reward mathematical learning that is only mechanical. Routine procedures can increasingly be automated. What remains essential is the human capacity to frame problems, interpret outputs, judge assumptions, build arguments, recognize structure, ask better questions, and use tools responsibly.

This does not mean that computation is unimportant. It means computation must be integrated into a richer mathematics education. Students should learn algorithms, but also learn how algorithms fit into proof, modeling, representation, and interpretation. They should learn to calculate, but also to explain. They should learn to use tools, but also to question them.

Mathematics learning should prepare students not only to answer known questions, but to approach unfamiliar ones. That requires non-algorithmic reasoning: the ability to notice, frame, represent, reason, justify, and reflect.

The future of mathematics learning is not a choice between procedural fluency and conceptual creativity. It is the integration of both. Algorithms give mathematics power. Non-algorithmic reasoning gives mathematics meaning. Together, they prepare learners to think mathematically in a world where computation is everywhere, but understanding remains irreplaceable.

Back to top ↑

Related Articles

• What Is Mathematical Thinking? Pattern, Proof, Architecture, and Reason
• Patterns, Structure, and the Mathematical Imagination
• Abstraction and the Power of Generalization
• Mathematics as the Science of Patterns
• Proof and the Logic of Mathematical Justification
• Logic and the Structure of Formal Inference
• Symbols, Language, and Mathematical Representation
• Conjecture, Creativity, and Mathematical Discovery
• Mathematical Thinking and AI-Assisted Discovery
• Mathematical Thinking and Proof Assistants

Further Reading

• Boaler, J. (2016) Mathematical Mindsets. San Francisco: Jossey-Bass. Available at: https://www.wiley.com/en-us/Mathematical+Mindsets
• Devlin, K. (2012) Introduction to Mathematical Thinking. Stanford University / Coursera course materials and related publications. Available at: https://web.stanford.edu/~kdevlin/
• Lockhart, P. (2009) A Mathematician’s Lament. New York: Bellevue Literary Press. Available at: https://www.bellevueliterarypress.org/books/a-mathematicians-lament/
• Papert, S. (1980) Mindstorms: Children, Computers, and Powerful Ideas. New York: Basic Books. Available at: https://mitpress.mit.edu/9780465046744/mindstorms/
• Pólya, G. (1945) How to Solve It: A New Aspect of Mathematical Method. Princeton: Princeton University Press. Available at: https://press.princeton.edu/books/paperback/9780691164076/how-to-solve-it
• Schoenfeld, A.H. (1985) Mathematical Problem Solving. Orlando: Academic Press.
• Skemp, R.R. (1976) ‘Relational understanding and instrumental understanding’, Mathematics Teaching, 77, pp. 20–26.
• Velleman, D.J. (2019) How to Prove It: A Structured Approach. 3rd edn. Cambridge: Cambridge University Press. Available at: https://www.cambridge.org/highereducation/books/how-to-prove-it/6D2965D625C6836CD4A785A2C843B3DA

Back to top ↑

References

• Boaler, J. (2016) Mathematical Mindsets. San Francisco: Jossey-Bass. Available at: https://www.wiley.com/en-us/Mathematical+Mindsets
• Devlin, K. (2012) Introduction to Mathematical Thinking. Stanford University / Coursera course materials and related publications. Available at: https://web.stanford.edu/~kdevlin/
• Lockhart, P. (2009) A Mathematician’s Lament. New York: Bellevue Literary Press. Available at: https://www.bellevueliterarypress.org/books/a-mathematicians-lament/
• Papert, S. (1980) Mindstorms: Children, Computers, and Powerful Ideas. New York: Basic Books. Available at: https://mitpress.mit.edu/9780465046744/mindstorms/
• Pólya, G. (1945) How to Solve It: A New Aspect of Mathematical Method. Princeton: Princeton University Press. Available at: https://press.princeton.edu/books/paperback/9780691164076/how-to-solve-it
• Schoenfeld, A.H. (1985) Mathematical Problem Solving. Orlando: Academic Press.
• Skemp, R.R. (1976) ‘Relational understanding and instrumental understanding’, Mathematics Teaching, 77, pp. 20–26.
• Velleman, D.J. (2019) How to Prove It: A Structured Approach. 3rd edn. Cambridge: Cambridge University Press. Available at: https://www.cambridge.org/highereducation/books/how-to-prove-it/6D2965D625C6836CD4A785A2C843B3DA

Back to top ↑