{
  "id": "undecidability",
  "title": "Undecidability",
  "lang": "en",
  "source_lang": "en",
  "summary": "Undecidability is the impossibility of solving a precisely specified decision problem by an algorithm that always terminates with the correct answer.",
  "category": "mathematics",
  "subcategory": "discrete-logic",
  "type": "concept",
  "aliases": [
    "Algorithmic undecidability",
    "Computational undecidability"
  ],
  "generated_at": "2026-11-04",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/undecidability/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/undecidability/",
      "json": "https://aiwiki.page/zh/undecidability.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Undecidability is a fundamental limitation on what can be accomplished by an algorithm. A decision problem is undecidable if no algorithm can take every admissible instance as input, terminate after finitely many steps, and correctly answer “yes” or “no.” The limitation concerns the existence of any such procedure, not merely the speed, memory requirements, or reliability of currently available software. Undecidable problems occur in mathematics, logic, and computer science, including questions about program execution, logical validity, polynomial equations, and geometric constructions.\n\nA related use of the word concerns statements within a formal theory: a sentence is undecidable in that theory if neither it nor its negation can be proved from the theory’s axioms. This is commonly called independence. Algorithmic undecidability and independence are closely connected, especially through Gödel’s incompleteness theorems, but they are not identical. One concerns the absence of a uniform computational method; the other concerns the deductive limits of a specified collection of axioms.\n\nThe modern theory emerged in the 1930s from attempts to formalize effective calculation and mechanize mathematical reasoning. Its central examples, particularly the halting problem and the decision problem for first-order logic, established rigorous limits on universal methods. Subsequent work developed techniques for proving undecidability, classifying different degrees of noncomputability, and identifying useful restricted problems that remain decidable."
    },
    {
      "heading": "Decision problems and effective procedures",
      "text": "A decision problem can be represented as a set \\(A\\) of finite strings over a finite alphabet. Each string encodes an instance, and membership in \\(A\\) means that the instance has the answer “yes.” For example, an integer may be encoded in binary, a polynomial by a finite list of coefficients and exponents, and a computer program by its source code or machine description. The problem is decidable if there is a total computable characteristic function\n\n\\[\\chi_A(x)= \\begin{cases} 1,&x\\in A,\\\\ 0,&x\\notin A. \\end{cases}\\]\n\n“Total” is essential: the procedure must halt on every input. A process that occasionally runs forever does not decide the problem, even if every answer it produces is correct.\n\nThe standard mathematical model is the Turing machine, introduced by Alan Turing in 1936. Equivalent formulations use recursive functions, lambda calculus, or ordinary programming languages with unbounded computational resources. The Church–Turing thesis identifies effectively calculable procedures with the functions computable by these models. This identification is not a theorem about an independently formalized notion of all possible effective methods; it is supported by the agreement of different formalizations and by their ability to represent familiar algorithmic procedures. Once a model is fixed, undecidability results within it are mathematical theorems.\n\nA weaker property is recognizability, also called computable enumerability or semidecidability. A computably enumerable set has a procedure that halts and accepts exactly its members. On a nonmember, that procedure may run forever. Equivalently, its elements can be listed by an algorithm, with repetitions allowed. This distinction captures many searches: a procedure may eventually find a proof, a solution, or a terminating computation without being able to certify that none exists.\n\nA set is decidable exactly when both it and its complement are computably enumerable. For the nontrivial direction, run their recognizers in parallel, alternating simulation steps. Every input belongs to one of the two sets, so eventually one recognizer accepts and determines the answer. This argument explains why a one-sided search is not enough for decidability.\n\nUndecidability ordinarily concerns an infinite family of instances. Every decision problem on an explicitly given finite domain is decidable in principle by a finite lookup table. Practical computation, however, also requires consideration of the table’s size and accessibility. Likewise, a particular instance of an undecidable problem may be easy to settle. What fails is a single effective procedure guaranteed to settle every instance."
    },
    {
      "heading": "The emergence of undecidability",
      "text": "The historical background includes the effort to place mathematics on explicit axiomatic foundations. David Hilbert sought rigorous methods for establishing the consistency of mathematical reasoning, a project now associated with Hilbert’s program. A related question, the Entscheidungsproblem, asked for a general effective method to determine the validity of formulas of first-order logic. Hilbert and Wilhelm Ackermann prominently formulated this problem in their 1928 textbook on mathematical logic.\n\nIn 1931, Kurt Gödel proved that sufficiently expressive, effectively axiomatized formal theories cannot satisfy certain combinations of completeness and consistency. His construction encoded syntactic objects and proofs as natural numbers, allowing arithmetic to express claims about its own derivations. The first incompleteness theorem showed, under the hypotheses used in its original formulation, that there are sentences for which neither the sentence nor its negation is provable. J. Barkley Rosser’s 1936 refinement reduced the relevant assumption to consistency for the standard class of theories under consideration.\n\nGödel’s result did not by itself provide the modern general definition of an algorithm. That became central in 1936, when Alonzo Church and Alan Turing independently established the unsolvability of the decision problem for first-order logic. Church used lambda definability and recursive functions. Turing introduced an idealized machine with a finite control, a tape, and elementary read-write operations. His analysis connected a precise machine model with the informal idea of computation by following explicit rules.\n\nTuring’s original paper studied questions about machines producing sequences and used those impossibility results to address the Entscheidungsproblem. The familiar halting problem is the modern standard presentation of the same general obstruction, although it is not a verbatim statement of the principal problem in his paper. Emil Post’s work on finite combinatorial processes supplied another influential route to formalizing computation. The convergence of these approaches helped establish computability theory as a distinct discipline.\n\nA second historical development was the movement from foundational questions to concrete mathematics. In the 1940s and 1950s, undecidability was established for rewriting systems and word problems in algebra. In 1970, Yuri Matiyasevich completed the work of Martin Davis, Hilary Putnam, and Julia Robinson on Hilbert’s tenth problem, proving that no general algorithm decides whether an integer-coefficient polynomial equation has an integer solution. These results showed that undecidability was not confined to explicitly self-referential statements or artificial descriptions of machines."
    },
    {
      "heading": "The halting problem and diagonalization",
      "text": "The halting problem asks whether a program \\(P\\), given an input \\(x\\), eventually terminates. Programs and inputs have finite encodings, so the problem can be represented by the set\n\n\\[K=\\{\\langle P,x\\rangle : P\\text{ halts on }x\\}.\\]\n\nThere is a straightforward recognizer for \\(K\\): simulate \\(P\\) on \\(x\\), and accept if the simulation ends. What is impossible is a procedure that always terminates and correctly distinguishes terminating from nonterminating executions.\n\nSuppose that such a procedure \\(H(P,x)\\) exists, returning \\(1\\) when \\(P(x)\\) halts and \\(0\\) otherwise. Construct a new program \\(D\\) with the following behavior on an input \\(y\\):\n\n\\[D(y)= \\begin{cases} \\text{loop forever},&H(y,y)=1,\\\\ \\text{halt},&H(y,y)=0. \\end{cases}\\]\n\nProgram descriptions can themselves be supplied as data. Let \\(d\\) be the description of \\(D\\), and consider \\(D(d)\\). If \\(H(d,d)=1\\), the definition forces \\(D(d)\\) not to halt. If \\(H(d,d)=0\\), the definition forces it to halt. Either answer contradicts the claimed correctness of \\(H\\).\n\nThis is a computational form of diagonalization. It defeats a proposed universal classifier by constructing an object that behaves oppositely to its prediction on a specially chosen input. The construction does not rely on running a computer for an exceptionally long time, nor does it assume that programmers lack ingenuity. It proves that the specification of a total, universally correct halting classifier is inconsistent with the computational model.\n\nThe complement of \\(K\\), consisting of nonhalting computations, is not computably enumerable. If it were, both \\(K\\) and its complement would be recognizable, making \\(K\\) decidable by parallel simulation. Consequently, nontermination cannot in general be established by an exhaustive effective search through certificates that exist for every nonterminating program and are absent for every terminating one.\n\nA related counting argument shows that noncomputable sets must exist. There are only countably many finite programs but uncountably many subsets of the natural numbers, by Cantor’s theorem. Therefore most such subsets have no decision algorithm. This argument establishes abundance but does not identify a particular naturally occurring undecidable problem. The halting proof does identify one, explains the obstruction explicitly, and supplies a starting point for transferring undecidability to other domains."
    },
    {
      "heading": "Reductions and Rice’s theorem",
      "text": "The main method for proving a new problem undecidable is a reduction from a problem already known to be undecidable. For sets \\(A\\) and \\(B\\), a many-one reduction\n\n\\[A\\leq_m B\\]\n\nis a total computable function \\(f\\) satisfying\n\n\\[x\\in A\\quad\\Longleftrightarrow\\quad f(x)\\in B.\\]\n\nIf a decider for \\(B\\) existed, computing \\(f(x)\\) and applying that decider would decide \\(A\\). Thus the undecidability of \\(A\\) implies the undecidability of \\(B\\).\n\nThe direction matters. Reducing a new problem to the halting problem supplies an upper bound on its computational difficulty; it does not establish that the new problem is undecidable. To establish undecidability by this method, a known undecidable problem must be reduced to the new one. The transformation must also be effective. An assignment that maps instances according to their unknown correct answers is not a valid reduction.\n\nFor example, consider the problem of deciding whether a program ever prints a designated symbol. Given a program \\(P\\) and input \\(x\\), construct a program that simulates \\(P(x)\\), printing the symbol only if that simulation terminates. The constructed program prints the symbol exactly when \\(P(x)\\) halts. A universal printing detector would therefore decide the halting problem. The reduction converts a question about termination into an apparently different question about observable behavior.\n\nRice’s theorem, proved by Henry Gordon Rice in 1953, generalizes this pattern. Every nontrivial property of the partial function computed by a program is undecidable when programs are given by their descriptions. “Nontrivial” means that some computable partial functions have the property and others do not. “Property of the function” means that programs computing the same partial function must receive the same classification, regardless of their internal organization.\n\nExamples include whether a program computes a total function, whether its computed function ever takes the value zero, and whether it computes a specified computable function. By contrast, properties such as having fewer than a thousand instructions or containing a particular syntactic construct are not properties of the computed function; they can generally be checked directly from an appropriate finite representation. Rice’s theorem does not make every question about programs undecidable.\n\nReductions need not preserve running time efficiently when used only to establish undecidability. They must terminate, but they may be computationally expensive. Complexity theory imposes additional restrictions, such as polynomial-time computability, because it compares resource requirements among problems that may all be decidable. The logical form of a reduction is similar in both settings, but its required strength depends on the question being investigated."
    },
    {
      "heading": "Logical validity, arithmetic, and independent statements",
      "text": "In first-order logic, validity means truth in every structure interpreting a sentence’s nonlogical symbols. The set of valid first-order sentences is computably enumerable: a sound, complete, effectively presented proof calculus permits systematic enumeration of all proofs. Gödel’s completeness theorem, established in 1929 and published in 1930, ensures that every logically valid sentence has such a proof. Church’s and Turing’s results show that validity is nevertheless undecidable.\n\nThere is no contradiction between completeness and undecidability. Completeness guarantees that proof search eventually succeeds for valid sentences. It does not guarantee that a search terminates when a sentence is invalid. Moreover, logical completeness concerns the relationship between validity and formal derivability in a proof calculus. It differs from the completeness of a theory, which requires that each sentence in the theory’s language be either provable or refutable from its axioms.\n\nFor a theory \\(T\\), a sentence \\(\\varphi\\) is independent of \\(T\\) when\n\n\\[T\\nvdash\\varphi \\qquad\\text{and}\\qquad T\\nvdash\\neg\\varphi.\\]\n\nIndependence is always relative to a theory. A sentence independent of one theory may become an axiom or theorem of a stronger theory. For an effectively axiomatized, syntactically complete classical theory, simultaneous searches for a proof of \\(\\varphi\\) and a proof of \\(\\neg\\varphi\\) yield a decision procedure for theoremhood. A consistent theory with an undecidable theorem set must therefore be incomplete.\n\nArithmetic supplies especially strong results. Robinson arithmetic, usually denoted \\(Q\\), is a finitely axiomatized theory capable of expressing enough elementary computation to be essentially undecidable: every consistent computably axiomatized extension of it is undecidable. Peano arithmetic is a familiar stronger example. Under consistency, Rosser’s refinement of incompleteness guarantees independent sentences for these effectively axiomatized theories. This is not a claim that every mathematical theory is incomplete or undecidable.\n\nThe set of all sentences true in the standard natural-number structure is complete, because every sentence has a truth value there. It is not computably enumerable. Thus adding all true arithmetical sentences would achieve completeness only by abandoning effective axiomatizability. Tarski’s undefinability theorem adds a related limitation: arithmetical truth cannot be defined within arithmetic by a single formula that correctly captures all arithmetical sentences.\n\nIndependence in set theory provides a distinct illustration. Gödel and Paul Cohen established the relative consistency results underlying the independence of the continuum hypothesis from Zermelo–Fraenkel set theory with the axiom of choice. Subject to the relevant consistency assumption, neither the hypothesis nor its negation follows from those axioms. Such a result concerns a particular statement and axiom system, rather than directly asserting that a specified family of inputs lacks an algorithm."
    },
    {
      "heading": "Degrees of unsolvability and definability hierarchies",
      "text": "Undecidability is not a single uniform level of difficulty. Some undecidable problems become solvable when a procedure is permitted to query solutions to another undecidable problem; others remain unsolvable even with that assistance. Turing reducibility formalizes this comparison. The notation\n\n\\[A\\leq_T B\\]\n\nmeans that a machine can decide membership in \\(A\\) when equipped with an oracle answering membership questions about \\(B\\). Unlike a many-one reduction, an oracle computation may make multiple adaptive queries.\n\nTwo sets have the same Turing degree when each is Turing reducible to the other. All decidable sets belong to the degree conventionally written \\(\\mathbf{0}\\). The halting problem has degree \\(\\mathbf{0}'\\), the first jump of the computable degree. For any oracle \\(A\\), its relative halting problem \\(A'\\) is not computable from \\(A\\). Relativizing the diagonal argument therefore yields an unending progression of stronger computational tasks.\n\nNot every undecidable computably enumerable set has the halting problem’s degree. Emil Post asked whether there were computably enumerable degrees strictly between \\(\\mathbf{0}\\) and \\(\\mathbf{0}'\\). Richard Friedberg and Albert Muchnik independently answered this positively in the 1950s using priority constructions. Their results showed that an undecidable enumerable problem need not be powerful enough, as an oracle, to solve every other enumerable problem. The degree structure also contains incomparable elements: neither problem need be computable from the other.\n\nThe arithmetical hierarchy classifies sets according to the quantifiers needed to define them over natural numbers. At the first level, a set is \\(\\Sigma^0_1\\) if membership can be written\n\n\\[x\\in A\\quad\\Longleftrightarrow\\quad \\exists s\\,R(x,s),\\]\n\nwhere \\(R\\) is a decidable relation. These are exactly the computably enumerable sets. Their complements form the class \\(\\Pi^0_1\\). The halting problem is \\(\\Sigma^0_1\\)-complete under many-one reductions, while nonhalting is \\(\\Pi^0_1\\)-complete.\n\nAlternating quantifiers produces higher levels. For example, totality of a program can be expressed as\n\n\\[\\forall x\\,\\exists s\\, [\\text{the program halts on input }x\\text{ within }s\\text{ steps}],\\]\n\nand the set of indices of total programs is \\(\\Pi^0_2\\)-complete. This classification explains why totality is not merely another recognizable version of halting. The full theory of truth in the natural numbers lies beyond every finite level of the arithmetical hierarchy.\n\nThese hierarchies distinguish the information required to solve problems, rather than their execution time. Two problems can both lack ordinary algorithms while having sharply different relative strengths, definability properties, and possibilities for partial recognition."
    },
    {
      "heading": "Undecidability in number theory, algebra, and topology",
      "text": "Hilbert’s tenth problem asked for an algorithm deciding whether a polynomial equation with integer coefficients has an integer solution. Its negative solution followed from the Davis–Putnam–Robinson–Matiyasevich theorem, completed in 1970. The theorem states that every computably enumerable subset of the natural numbers has a Diophantine representation: for a suitable integer-coefficient polynomial \\(p\\),\n\n\\[n\\in A \\quad\\Longleftrightarrow\\quad \\exists x_1,\\ldots,x_k\\in\\mathbb N\\; p(n,x_1,\\ldots,x_k)=0.\\]\n\nApplying this to an undecidable enumerable set transfers undecidability into the existence of polynomial solutions.\n\nThe versions over nonnegative integers and over all integers are connected by standard encodings. In particular, Lagrange’s four-square theorem allows a nonnegative variable to be represented by a sum of four integer squares. Integer variables can in turn be represented as differences of nonnegative variables. The conclusion concerns general polynomials with arbitrarily many variables and varying degrees, not every restricted equation family. Linear Diophantine equations remain decidable. The corresponding general problem over the rational numbers remains open.\n\nIn group theory, the word problem asks whether a word in generators and their inverses represents the identity in a group specified by generators and defining relations. Pyotr Novikov in 1955 and William Boone later in the 1950s proved the existence of finitely presented groups with undecidable word problem. Although every input word is finite, the relations can encode computations whose consequences cannot be decided uniformly.\n\nMany important groups nevertheless have decidable word problems. Free groups admit cancellation procedures, and finitely generated abelian groups can be handled using integer linear algebra. The undecidability theorem therefore does not say that symbolic manipulation in groups is generally futile. It says that finite presentations can specify structures too expressive for a universal identity-testing algorithm, and even that particular fixed finitely presented groups can have undecidable word problem.\n\nIn topology, finite combinatorial descriptions can encode similarly difficult questions. A triangulated space is described by finitely many simplices and gluing data, but finiteness of description does not ensure that its global properties are decidable. Results originating with Andrey Markov show that the homeomorphism problem for finitely presented closed manifolds is undecidable in sufficiently high dimensions, including dimensions at least five. There is no algorithm that always determines whether two such input manifolds are homeomorphic.\n\nThese examples share a mechanism: finite mathematical data can simulate arbitrary computation. Their proofs require detailed constructions translating machine behavior into equations, relations, or geometric objects. Once that translation is established, a hypothetical decision algorithm in the target domain would solve an already undecidable computational problem."
    },
    {
      "heading": "Formal languages, rewriting systems, and tilings",
      "text": "Finite symbolic systems provide especially direct settings for undecidability. A rewriting system consists of finitely specified rules replacing one string or expression by another. Individual rule applications are mechanical, but unrestricted chains of replacements can simulate machine computations. Consequently, questions about whether one expression can eventually be transformed into another, or whether every permitted transformation sequence terminates, become undecidable in sufficiently general rewriting formalisms.\n\nThe Post correspondence problem, introduced by Emil Post in 1946, is a compact example. An instance contains finitely many pairs of words\n\n\\[(u_1,v_1),\\ldots,(u_m,v_m).\\]\n\nThe question is whether there is a nonempty sequence of indices \\(i_1,\\ldots,i_k\\), with repetition permitted, such that\n\n\\[u_{i_1}u_{i_2}\\cdots u_{i_k} = v_{i_1}v_{i_2}\\cdots v_{i_k}.\\]\n\nEvery proposed solution can be checked by finite concatenation and comparison. Exhaustively enumerating index sequences therefore recognizes positive instances, but there is no decision algorithm for all instances.\n\nFormal-language theory contains sharp boundaries between decidable and undecidable questions. For finite automata, language emptiness, equivalence, and inclusion are decidable by finite graph constructions. For context-free grammars, membership and emptiness are also decidable, but equivalence of two grammars is undecidable. Ambiguity—whether some generated word has more than one parse tree—is undecidable for general context-free grammars. The grammar is finite; the unbounded collection of derivations supplies the computational difficulty.\n\nGeometric local rules lead to another classic example. A Wang tile is a square with colored edges. Tiles may be placed on the integer grid without rotation, and touching edges must have matching colors. The domino problem asks whether a given finite tile set can tile the entire plane. Robert Berger proved its undecidability in 1966, answering a question posed by Hao Wang. The proof encodes computational constraints into local matching conditions.\n\nAn important distinction separates finite patches from infinite tilings. Testing whether a particular finite region can be tiled is a finite search and hence decidable. Tiling the whole plane requires compatible configurations at arbitrarily large scales. By a compactness argument, existence of suitable tilings of all finite squares yields an infinite tiling. Failure therefore has a finite obstruction, although no computable uniform bound on the necessary search can decide all tile sets.\n\nAperiodicity is connected to this history but is not identical to undecidability. A tile set is aperiodic when it tiles the plane but admits no periodic tiling. Such sets rule out a procedure that could always settle positive instances simply by searching through periodic patterns. Their existence reveals how finite local constraints can enforce global structures of unbounded complexity."
    },
    {
      "heading": "Consequences for program analysis and verification",
      "text": "Undecidability establishes limits on formal verification, but those limits depend on the class of systems and the exact property being checked. For arbitrary programs with unbounded memory, termination and many forms of reachability are undecidable. A termination detector would solve the halting problem directly. A reachability detector could do so by checking whether a designated final control location is ever reached during a simulation.\n\nProgram equivalence provides another important example. There is no general algorithm deciding whether two arbitrary programs compute the same partial function. One reduction compares a program constructed from a halting instance with a program that never terminates. The two compute the same nowhere-defined function exactly when the simulated computation never halts. Equivalence therefore cannot be settled by a universally terminating procedure on unrestricted program descriptions.\n\nFinite-state systems differ fundamentally. If a system has an explicitly finite set of states, reachability can be decided by graph search. Model checking exploits this fact to determine whether a state-transition model satisfies a logical specification. A physically fixed digital device has finitely many configurations in an appropriate abstraction, so exhaustive exploration is possible in principle. The obstacle may instead be state explosion: the number of states grows exponentially with the number of stored bits or interacting components.\n\nStatic analyses of unbounded programs commonly trade precision for guaranteed termination. Abstract interpretation replaces concrete program states with abstract descriptions, such as intervals of possible numerical values. A sound overapproximation includes every actual behavior but may include impossible ones. It can therefore establish that a bad state is unreachable, while sometimes reporting a possible error that no real execution can produce. Undecidability explains why no effective analysis can be simultaneously terminating and exact for every program on sufficiently expressive classes of properties.\n\nOther methods limit what must be decided. Bounded verification checks executions up to a specified length or resource bound; this is a different problem from checking every possible execution. Interactive proof systems allow users to supply invariants, lemmas, and termination arguments. Checking a fully specified proof can be decidable even when discovering a suitable proof is not guaranteed to terminate. An automatic prover may also return “unknown,” avoiding a false claim of universal completeness.\n\nThe simultaneous requirements are what create the barrier: unrestricted input programs, an exact semantic property, guaranteed termination, and correct binary answers on every input. Removing or weakening one requirement can yield useful methods. Successful verification tools operate within such mathematically specified boundaries rather than contradicting the underlying impossibility results."
    },
    {
      "heading": "Decidable fragments, complexity, and unresolved questions",
      "text": "Undecidability often disappears when the allowed language or input structure is restricted. Presburger arithmetic, introduced by Mojżesz Presburger in 1929, is the first-order theory of natural numbers with addition and order. It is decidable. By contrast, adding unrestricted multiplication produces enough expressive power for the undecidability of arithmetic. The difference shows that even elementary-looking operations can radically change the algorithmic status of a theory.\n\nThe first-order theory of real closed fields is also decidable, through methods associated with Alfred Tarski and later algorithms for quantifier elimination. Sentences about real numbers using polynomial equalities, inequalities, and quantifiers can therefore be decided. This contrasts with the undecidability of polynomial solvability over integers. The coefficient language may look similar, but requiring solutions to lie in a discrete domain changes what computations the equations can encode.\n\nDecidability does not imply practical efficiency. A terminating algorithm may require prohibitive time or memory. Conversely, a problem described as computationally hard is not necessarily undecidable. The traveling salesperson problem on a finite weighted graph can be solved by enumerating tours, however inefficiently. The unresolved \\(P\\) versus \\(NP\\) question concerns efficient computation for decidable problems; it does not ask whether those problems possess algorithms at all.\n\nAn open problem is likewise not automatically an undecidable one. A problem may lack a known solution because the appropriate algorithm or impossibility proof has not been discovered. Hilbert’s tenth problem over the rationals is an example whose decidability status remains unresolved. A conjecture such as the Riemann hypothesis is a single mathematical assertion, not by itself an infinite input-output decision problem. Independence from a specified theory would require a separate proof, not merely prolonged failure to establish the conjecture.\n\nClaims that physical systems exhibit undecidability also depend on their mathematical formulation. A model may allow an unbounded lattice, exact real-valued parameters, indefinitely long evolution, or an arbitrarily detailed initial configuration. An undecidability theorem for that model identifies a precise obstruction under those assumptions. It does not automatically show that a particular finite experiment cannot be predicted to a specified finite accuracy.\n\nOrdinary randomization does not remove the basic barrier when a randomized procedure has the standard uniform correctness guarantees of a decision algorithm. Nor does ordinary quantum computation supply solutions to Turing-undecidable problems under its usual effective formulation. Oracle machines and proposed forms of hypercomputation change the permitted computational resources. Their analysis is meaningful, but an oracle for the halting problem is an added mathematical assumption, not an algorithmic construction of the missing decider."
    }
  ],
  "links": [
    "algorithm",
    "mathematics",
    "logic",
    "computer-science",
    "godels-incompleteness-theorems",
    "turing-machine",
    "church-turing-thesis",
    "david-hilbert",
    "hilberts-program",
    "first-order-logic",
    "kurt-godel",
    "halting-problem",
    "cantors-diagonal-argument",
    "natural-number",
    "godels-completeness-theorem",
    "peano-axioms",
    "set-theory",
    "continuum-hypothesis",
    "zermelo-fraenkel-set-theory",
    "polynomial",
    "rational-number",
    "group-theory",
    "topology",
    "formal-verification",
    "model-checking",
    "alfred-tarski",
    "traveling-salesman-problem",
    "riemann-hypothesis",
    "quantum-mechanics"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}