Mathematical Thinking - Preface

MATH.Preface:1 - Problem frame - Construct mathematics for the question

You may know a formula, a programming technique or a useful physical law and still be unable to formulate the next problem. The objects may have been chosen too coarsely. Two operations may work separately but fail when combined. A plausible statement may need a proof, or a proof may leave you without a way to obtain its promised object.

Mathematical Thinking helps construct and develop the mathematics needed in such situations. Its starting repertoire forms objects and operations, tests identifications, builds arguments and witnesses, changes representations, and obtains consequences from transformations and constraints. Work can begin in mathematics itself, a physical investigation or the design of another working method.

This edition contains twenty patterns. Mathematical Thinking belongs to the Foundational Thinking DPF Suite alongside Mathematical Modeling, Physical Thinking, Computational Thinking and Notational Engineering. Together they connect these inquiries with the development of methods of work. Use the present Table of Contents to find an available method; a planned contribution still requires another source or collaborator. Pattern IDs remain stable across editions, including gaps left by withdrawn bodies; Parts group the available methods by the work they support.

Begin with the question that is blocked. If its mathematical form is still unclear, the Readme MP-FRAME entry uses FPF B.5.FM, B.5.TU and B.5.MPC to obtain the first account and locate the missing contribution. Once a mathematical operation is needed, a body here develops that operation. The worked use in :4 begins before a representation has been selected. Use the Table of Contents for other questions; each pattern states its prerequisites and conditions.

The common starting preparation is elementary sets, relations, functions and the ability to follow a short proof. Several examples need only integer arithmetic. Variation of a curve additionally uses differentiation and integration; the relevant pattern states those requirements. A collaborator can provide a mathematical contribution that you cannot yet construct yourself. Retain the inputs, conditions and result of that contribution so that the next part of the work can use it.

MATH.Preface:2 - Problem and forces - Make a usable construction

A mathematical description earns its place by making an operation, argument or question possible. Choosing names for objects is often the beginning of that work. The harder part is deciding how the objects can be formed, transformed, compared and used in a subsequent construction.

Several tensions recur:

The same question can have several satisfactory mathematical descriptions. A list of movement steps can show each change; a displacement summary can answer a final-position question with less information. A later question about visiting an intermediate position can require a distinction that the summary discarded. Choose a representation from the answer needed, its further use and the effort available.

MATH.Preface:3 - Solution - Connect constructions through what they supply

MATH.Preface:3.1 - Form objects, operations and representations

MATH.16 starts one step earlier, when several constructions seem plausible. Describe the maps the new object must support, then compare arbitrary allowed ways of supplying or processing its data. The resulting universal property distinguishes, for example, carrying both components from accepting either input, and arbitrary pairs from compatible pairs. It can also specify an object that represents a prepared function. A known construction can then supply the object and its maps.

MATH.1 starts with permitted elementary connections and constructs finite paths, identities and composition. Retaining a path can preserve the order and history that its final effect forgets. MATH.5 starts with assigned values for generators and obtains values for their composites while preserving the operations.

When several descriptions should count as the same input, MATH.2 tests whether the required operation is independent of the representative. For a partially available operation, its availability can matter as much as its output. A failed test supplies a distinction to restore.

MATH.7 addresses a reversible change of representation. It constructs the operations in the new representation and carries results back. Renaming the elements while keeping an unsuitable operation can change the problem; the transported operation supplies the repair.

MATH.17 makes the rules themselves available for construction and change. Select allowable operations, determine whether their composition stays allowable, then construct an operation that transforms them. Its required law follows the intended use: repeating a composite and repeating its stages separately can yield different answers.

MATH.18 compares descriptions with different primitives, allowed maps or equality. Construct the needed interpretations, establish what transfers, and inspect the return. A partial interpretation can be enough for one consequence; equivalence requires the corresponding comparisons in both directions.

These methods can be used separately. They also connect: form expressions from generators, interpret their operations, identify descriptions that preserve the desired answer, then choose a convenient representation for calculation. When the rule or the whole account changes, use MATH.17 or MATH.18 to construct and examine that change.

MATH.Preface:3.2 - Obtain an argument and the object it supports

MATH.19 constructs an argument when the premises and conclusion are known but the connecting steps are missing. Work backward to a sufficient claim and forward to available consequences, then prove a lemma joining them. An unsuccessful induction can require an extra parameter or a stronger intermediate statement. B.5.RA instead helps recover an argument already supplied in another description.

MATH.4 obtains a witness by following the construction of a finite input. A step may require a stronger intermediate result or another parameter. MATH.12 recovers functions, pairs, projections and branch information from proof steps, including non-inductive steps. It distinguishes the operations that produce data from a proof that some suitable data exists.

MATH.6 constructs a case in which the assumptions hold and the proposed conclusion fails. Such a case can reveal a missing premise, a misplaced quantifier or an overly narrow search. MATH.11 instead solves for a function preserved by the allowed transformations. Its value can exclude a target or determine an accumulated quantity. Equal values leave any required reachability construction to be supplied.

MATH.20 obtains a useful comparison before the whole unknown is available. A feasible path bounds a shortest length from above; inequalities covering all paths can bound it from below. The method constructs the comparison, propagates its direction through the needed operations and tightens the part that leaves a consequential gap. It can return enough for the next decision without completing an optimization.

An argument and an obtaining procedure can support each other. Given a finite list, a terminating test and a proof that some listed element passes, testing the entries obtains a witness. If the searched range becomes infinite, the finite-search argument must be reconsidered. The available result may remain a logical conclusion or a procedure for each finite portion.

MATH.Preface:3.3 - Change a construction and develop its theory

MATH.13 establishes how a transformation of the data relates to transformations of admissible candidates and answers. That relation can transfer a solution, constrain a unique answer or expose an impossible choice requirement. MATH.8 constructs the resulting orbit, removes repetitions and separates one orbit from all solutions. MATH.9 constructs a choice compatible with the transformations when the input’s own symmetries permit one.

When candidates satisfy constraints, MATH.10 constructs changes that stay within them and calculates what those changes do to a criterion. The result may be an improving candidate or a necessary condition. A minimum requires the corresponding additional argument. In a physical-action calculation the requested condition can be stationarity.

Symmetry and variation can simplify the same problem while answering different questions. One establishes a relation among transformed problems and solutions; the other investigates admissible changes and their effect on a criterion. Preserve the conclusion supplied by each.

MATH.21 constructs an object through converging or compatible approximations. The intended use selects what convergence must preserve: a finite prefix, function values, an integral or another observation. MATH.20 can provide the necessary error bound; MATH.19 can supply a missing limit or interchange argument. A limit’s existence and an effective way to obtain the requested finite information require their respective constructions.

MATH.22 changes the permitted objects or reasoning by changing assumptions. Follow the affected proof steps and constructions, retaining a conclusion when its justification survives or can be repaired. MATH.18 supplies interpretations between accounts; MATH.6 can separate a claimed consequence from what the new assumptions allow.

MATH.23 turns a variation, obstruction or unexplained regularity into a next mathematical claim and a proving or refuting operation. If equal weighting of group means fails for unequal groups, restricting all groups to equal size abandons the original need. Retaining sums and counts repairs the construction and opens a general question about combinable summaries. Proof construction, countermodels and changed axioms then provide different continuations.

MATH.Preface:3.4 - Return a construction to further work

The connecting rule is simple: name what one construction returns and what the next one uses. A set of solutions, one chosen solution, a proof of existence and an executable selection support different continuations. Enter at a contribution already available and stop when the requested mathematical result has been supplied.

For a question about another subject, use FPF C.29 to establish the correspondence through which the mathematical result answers that question. C.29.2 develops a missing computational formulation; C.29.3 connects a computation with the arrangement that prepares its inputs, performs it and exposes an interpretable result. B.5.MPC coordinates mathematical, physical and computational reasoning when their contributions must be developed together.

MATH.Preface:3.5 - Use this contribution within the wider repertoire

The Foundational Thinking Suite Reference explains where this mathematical contribution connects with model formulation, physical premises, computation, notation and Method Engineering. A result such as a function object or an interpreted path becomes useful through what the next operation can do with it. Its mathematical laws alone leave the subject interpretation and execution conditions to their respective methods.

The twenty bodies connect formation and interpretation with proof, comparison and continued theory development. MATH.17 makes operations available for change, MATH.18 compares interpretations, MATH.19 builds a missing proof, and MATH.20/.21 connect justified bounds with approximation. MATH.22/.23 develop changed theories and useful conjectures. Start with the contribution needed next and read the methods that supply its missing inputs.

For example, changing the order of a read and an update can preserve the final stored value but change the reading used by a later decision. MATH.1/.5 supply sequences and their interpretation, and MATH.2 tests the proposed identification. Method Engineering uses that distinction when deciding how work may be rearranged. The mathematical construction and its use in the working method remain separately inspectable.

MATH.Preface:3.6 - Constituent actions in ongoing work

A symbolic rewrite can be part of proving a lemma while that lemma’s proof is part of proving a larger claim. The required domain constrains the rewrite at that same moment: cancelling a factor is permissible only under the relevant algebraic conditions, and excluding zero would change a claim that is meant to include it. Knowing the symbols and the theorem goal can leave the intermediate reasoning unavailable. Recover or obtain that reasoning rather than treat the smaller calculation as proof of the whole.

FPF B.1.5.EW helps recover these constituent–whole connections; B.1.5.RS examines a proposed replacement. Use the parts of the vertical that can change the present result. A Method described here can require additional capability, available support and compatible resources at other grains.

MATH.Preface:4 - Worked use - Choose a representation after the question changes

A team receives travel logs from a cart moving along a straight line. Two runs have the same recorded distance, yet one ends at the starting point and the other does not. The immediate question is which runs returned to their start. Starting with a familiar distance formula leaves the missing information unresolved.

Recover the subject account. B.5.FM asks what can change the answer. Total distance has forgotten direction. Establish where the line’s origin and positive direction are, and what a recorded movement denotes. For this small case, each observed movement is a monotone step of one unit, either forward or backward; the cart starts at zero. This is the supplied physical account. If the log only records a commanded movement, the actual displacement remains a question for observation or a supported movement model. B.5.MPC keeps that physical contribution distinct from the subsequent calculation.

Construct and interpret the mathematics. Use MATH.1 to form finite sequences of forward and backward steps. MATH.5 constructs their displacement map by assigning +1 to a forward step, -1 to a backward step, and addition to concatenation. The empty sequence has displacement zero. Forward-then-backward and forward-then-forward both have total distance two, but their displacements are zero and two. Thus the first run returns to its starting point and the second does not, under the stated account.

The method needed an operation on sequences, not just names for two kinds of movement. MATH.5’s extension explains why adding the elementary displacements evaluates any finite sequence and respects concatenation. C.29 connects the resulting number back to the cart’s position. If a program performs the calculation, C.29.2/.3 connect that rule to its input interpretation and implementation. For example, adding absolute distances would compute a different quantity from the one now requested.

Decide what can be forgotten. For the return-to-start question, sequences with the same displacement have the same answer. Concatenating another movement sequence adds the same further displacement. MATH.2 therefore permits identification by displacement for these operations and this answer. The compact representation is useful because its retained information has been matched to the continuation.

Change the requested result. Now ask whether a run visited position +1. Forward-then-backward visits +1; backward-then-forward does not. Both end at zero. The two histories could be identified for the former question, but their equivalence no longer preserves this new answer. Return through MATH.2 to the forgotten distinction. Retain the path and compute its successive positions; compare each with the queried location. In this unit-step example the recorded endpoints and monotone steps suffice. For a longer continuous movement from 0 to 2, the endpoints already imply a visit to +1. Use the recorded positions and known conditions of movement to answer the visit question. Recover further information only when those conditions leave the answer undetermined: a run that starts and ends at 0, for example, may or may not have reached +1 in between.

Change the working method. The team can now revise its logging and analysis method: retain the movement information needed by the questions it actually asks, specify how a calculation interprets it, and return a result with that meaning. An observer can supply displacements, a mathematical contributor can define the representation and its operations, and a programmer can implement them. ME.7 helps describe those proposed contributions and their relations. ME.12 checks the method claims and the description used by their recipients. Each contributor must understand the conditions at the join where another uses the result.

Develop the next question. The failure suggests a further mathematical problem: what smaller summary, if any, preserves both displacement and the requested visit information under concatenation? B.5.QD helps turn that question into a first construction or counterexample. MATH.16 can help specify what a proposed summary must let the next operation recover; the required maps and laws still need to be established. The cart is the worked example. Selecting distinctions by their use, constructing operations, justifying a compression and restoring information after a changed question are the reusable moves.

MATH.Preface:5 - Worked use - Construct a choice after finding an obstruction

Four positions form a cycle, with costs (1,3,1,3). The required result is one cheapest position. Changing the numbering origin must rotate the selected position along with the input.

MATH.13 first establishes the transformation relation: rotation preserves the cost-minimization question. MATH.8 identifies the relevant transformed cases. A half-turn leaves this particular input unchanged but swaps its two cheapest positions.

Apply MATH.9’s choice condition. An input-preserving transformation must also preserve the answer chosen from that input. Neither cheapest position is fixed by the half-turn, so the requested deterministic rule cannot choose one of them while meeting the rotation requirement.

This obstruction gives useful ways to change the problem. If the receiver can use all minimizers, return the set of the two cheapest positions. If one position is needed and a meaningful mark is available, include that mark in the input and choose the first cheapest position clockwise from it. A joint rotation preserves distances from the mark, so the selected position rotates as required. MATH.9 gives the construction and its reason.

A chosen lowest numeral would introduce a preferred origin. It is appropriate only when that added distinction belongs to the problem. A later request for continuity of the choice raises another condition; an equivariance argument alone has not addressed it.

The result identifies the missing input distinction or output change that permits a construction. The same pattern of work appears in mathematics used for geometric learning, cyclic schedules and other settings, with each setting supplying its own acceptable outputs and additional conditions.

MATH.Preface:6 - Checks and common failures - Preserve what the next construction needs

For a use spanning several patterns, check the following relations at the joins:

1. The next operation receives objects in its domain, with every enabling condition it uses.
2. An identification retains the operation or answer that will be applied to the classes.
3. A change of representation carries the relevant operations and quantities, including the rule for returning the answer.
4. A consequence has the scope supplied by its argument: for example, impossibility, a necessary condition, a constructed witness or a minimum.
5. A computation has the data, branch decisions and termination properties claimed for its use.
6. A changed premise is followed through the constructions that depend on it.

Use the corresponding local pattern’s argument where it already answers the question. A small direct use needs only its relevant conditions. A larger claim can require more: obtaining a minimum over one orbit does not by itself settle a minimum over every solution, and a physical conservation statement requires the dynamics under which the quantity is conserved.

The movement example shows why a representation must be judged against the question it answers. Accumulated displacement answers the final-position question; it loses information needed for some visit questions. MATH.2 supplies the test for retaining an answer when cases are identified. In symmetry-dependent choice, MATH.8 and MATH.9 ask how the answer must change with the data; an invariant quantity and a choice that transforms with the input need different constructions.

Another failure arises when mathematical equality is used to reorganize work without examining execution. MATH.12’s split-and-join example returns equal values for a fixed function, while repeated evaluation can cost more. If a call changes hidden state, repeated calls can also return different values. Restore the relevant state and use C.29.3 for the execution correspondence before transferring that mathematical transformation.

MATH.Preface:7 - Consequences, biases and limits

The language makes useful intermediate results available for reuse: composable paths, valid classes, interpreted expressions, witnesses, countermodels, invariant equations, solution families and qualified improvements. Their conditions help divide a larger problem among contributors and locate the part that needs revision.

This arrangement adds the cost of constructing the mathematical account. A reusable proof or operation can repay that cost across many cases. For a small question, direct calculation or an existing result can be more economical. A failed construction remains useful when it exposes the premise or operation that must change.

The examples favour small, inspectable constructions. They make dependence and failure visible, but do not establish that every larger instance will be computationally affordable. Existence of an answer, an efficient algorithm and an implementable calculation have different requirements.

The starting repertoire concentrates on formation, proof, representation and transformations. Numerical analysis, statistical inference, signal processing and specialized algorithm design supply substantial further methods. Use those contributions when the question reaches their conditions. The present patterns can help formulate that question and carry the resulting mathematical contribution into a larger argument.

An assisting agent can propose objects, examples or proofs and execute calculations. Its contribution must be understandable at the join where another contributor uses it. The mathematical statement and its justification remain available for criticism and revision, including when a person delegates the detailed calculation.

MATH.Preface:8 - Architectural Rationale - Organize by reusable constructions

A single mathematical construction can serve many subjects. A quotient can retain the information used by a later operation; symmetry can organize solutions of a theoretical problem or constrain a learning system. This reuse makes the mathematics valuable across disciplines while leaving its specialized construction methods with mathematical practice.

The organization therefore separates three relations. Mathematical methods form objects, perform operations and establish consequences under stated assumptions. A modeling correspondence interprets an object or result in another subject. An executing arrangement performs the selected computation. These relations can be developed together, but each supplies conditions that the others need. FPF’s C.29 family makes their connections usable; the mathematical bodies develop the constructions in detail.

This also clarifies the connection to methodology. Functions or paths can model how contributions compose, and mathematical laws can expose a failed identification or changed order. The model must retain the state, interactions and outputs relevant to the working method. Method Engineering, including its composition method, concerns the method being designed. Choosing morphisms to describe it provides a mathematical account whose adequacy depends on that interpretation.

Choosing a construction from its required maps is another reusable move. Its specification can lead to a product, a compatibility construction or a function object. The corresponding realization and proof still have to be supplied. This lets MATH.16 help choose among constructions while MATH.2, MATH.5 and MATH.7 retain their detailed quotient, extension and transport methods.

The pattern boundaries follow different reusable moves. Constructing a path differs from interpreting its generators; forming a quotient differs from transporting structure through a bijection. Induction obtains witnesses by input construction, while proof extraction can also use non-inductive steps. Symmetry consequences, orbit construction and compatible choice have different results and stopping conditions. Keeping these moves addressable lets a user take the required contribution and preserve its explanation, countercases and source alternatives.

A textbook can teach these constructions through a sustained sequence. A reference arranged by recurring difficulties supports another use: enter with a blocked question, obtain the relevant construction and continue elsewhere. The pattern language complements detailed source treatments, which remain useful for deeper theory and specialized methods.

There can be further useful scales. A geometric-computation profile may reuse symmetry, transport and variation while adding its own conditions; a narrower profile may develop continuous choices for a particular representation. One pattern can contribute to several such profiles. Specialization, reuse and composition describe those relations; a chapter order only helps a reader navigate them. Profiles should add a useful difference for their narrower situation rather than replace a developed method with a broad summary.

Revise this organization when a recurring use needs a construction that no body supplies, when two bodies duplicate the same useful move, or when a changed method makes their joins misleading. Preserve still-useful operations, explanations and source qualifications during that change.

MATH.Preface:9 - Source use and currentness

For formation and composition, Fong and Spivak’s Seven Sketches in Compositionality, §3.2, supplies paths and imposed equations. The patterns adapt these constructions to explicit enabling states and interpretation of composite operations. Direct enumeration remains useful when a small set of paths already answers the question.

For specifying an object through its needed maps, Riehl’s Category Theory in Context, §§2.3, 3.1–3.2, supplies universal properties and set constructions. Fong and Spivak’s Example 3.72 supplies currying: a function with two inputs becomes a function returning a function. MATH.16 uses these constructions to clarify an undecided use; directly defining a familiar representation remains sufficient when that choice is already settled.

Burris and Sankappanavar’s A Course in Universal Algebra, Chapter II, supplies congruences, quotients, term evaluation and isomorphisms. These support MATH.2, MATH.5 and MATH.7. The partial-operation convention in MATH.2 additionally preserves whether an operation is available at the represented state; the total-algebra definition alone does not choose that convention.

For obtaining objects from arguments, Wadler’s Propositions as Types and Rijke’s Introduction to Homotopy Type Theory supply constructive rules and their mathematical setting. The maintained Lean and Rocq accounts linked in MATH.4 and MATH.12 expose the consequences of data representation, proof erasure and choice for execution. The resulting method asks which operation actually obtains the data and permits classical reasoning in a justification of an independently computable operation.

Riehl’s Category Theory in Context, §§1.1, 1.3 and 1.5, supplies composition, transformations preserving it, and equivalence through compatible isomorphisms. MATH.17 uses those structures where the requested operation needs their laws; MATH.18 combines the family-level comparison with direct interpretations of primitives and assertions. A supplied inverse map remains sufficient for a simpler representation question.

For building and changing arguments, the maintained Logic and Proof and Theorem Proving in Lean accounts provide forward and backward proof steps. MATH.19 adapts them to finding a sufficient intermediate claim, including useful generalization; a known direct theorem remains cheaper when it already closes the goal. MATH.22 combines this proof-dependency work with model and interpretation comparison, retaining the distinction between failure of one proof and failure of the theorem.

For bounds and approximations, MATH.20 derives bounds from feasible constructions, universal inequalities and residual-to-error relations, using Vandenberghe’s duality account and Higham’s sensitivity analysis. MATH.21 uses Cauchy completion, compatible finite information and operation-specific convergence. Its maintained Lean sources make the conditions for passing a limit through an operation explicit; Bauer and Kavkler’s constructive-real account supplies an implementation alternative to a fixed convergence rate. The mathematical existence argument and the method of obtaining a requested approximation remain separately available.

For developing a new question, MATH.23 combines Lakatos’s proof-and-refutation method with Georgiev, Gómez-Serrano, Tao and Wagner’s account of AI-assisted conjecture and counterexample work. The latter contributes concrete search and proof routes and their resource limits. The Mathematical Research in the Age of AI declaration raises the retention of mathematical methods as a concern; it is treated as a position in that comparison. The adopted method returns a revised claim and an attainable next operation. A person or an AI agent can supply that contribution.

The symmetry bodies compare group-action constructions with equivariant output requirements and their obstructions. MATH.9’s source account distinguishes compatible selection from the additional regularity questions studied in geometric learning. MATH.10 and MATH.13 retain the separate assumptions for variation, physical conservation and numerical evolution. Return to their source comparisons when one of those stronger conclusions matters.

For invariants, Bayarmagnai, Mohammadi and Prébet supplies a constructive comparison with polynomial-loop methods. MATH.11 uses unknown coefficients and preservation equations at their stated scope. A global polynomial identity is one useful result; a restricted-state question can call for a different construction. The finite-state countercase shows when a cheaper check is sufficient.

These sources play different roles. Established mathematics supplies definitions and arguments; maintained formal libraries and current research expose further constructions, limits and implementation choices. A publication date alone does not make one answer replace another. Revisit a choice when the receiving question changes, a relied-on assumption fails, or another method obtains a more useful result at acceptable cost. The individual bodies keep the operative source passages and the conditions of that comparison.

MATH.Preface:10 - Relations to further reasoning and work

FPF B.5.RC and B.5.RA recover an available construction or argument; B.5.RR follows a changed premise through it. Mathematical Thinking supplies specific constructions that such reasoning can recover, use and revise. B.5.QD develops a further question from an obstruction or a successful result; C.39.RO and C.40.CD support reusable operations and their development. A construction can be worth retaining because it makes another operation or question possible before its final application is known. When choosing which question to pursue, state the work its answer could enable; E.10.INT distinguishes that usefulness from other senses of interest.

For expressions, A.6.3.RT and A.6.3.RT.OE help make an operation performable in a notation. Mathematical Thinking constructs selected mathematical structures and transports their operations. Developing a notation for a new range of work can require additional notational-engineering methods.

For application, the C.29 family and B.5.MPC connect mathematical results, computational constructions and physical accounts. Method Engineering contributes when the result is used to design or revise a way of working. Other subject frameworks supply the physical, organizational or professional methods used with the mathematics.

For evaluating alternatives, reuse FPF’s characteristic, comparison and improvement methods. C.11 supports a consequential choice; C.11.DUA helps decide whether further calculation or inquiry can change that choice enough to justify its cost. Mathematical Thinking supplies the relevant construction or quantitative relation.

MATH.Preface:End