PRH | Essay | 7.19 • The Reals as Blur–Completion of the Rationals

The Reals as Blur–Completion of the Rationals Cauchy, Cuts, and Coherent Blur Profiles Aleksandar Perišić November 2025 Abstract We show that the classical construction of R from Q via Cauchy sequences and Dedekind cuts admits a clean blur formulation: the real line is the sharp limit of a canonical blur diagram built from rational ε –views. At each fixed blur scale ε > 0, we group rationals into finitely many indistinguishability classes inside a bounded window and regard an “ ε –real” as the choice of one such class. A blur profile is then a family of compatible choices across all scales. We prove that: (i) readable blur profiles are in bijection with real numbers, (ii) the usual metric on Ris the sharp limit of the blur metrics, and (iii) completeness is precisely the statement that every Cauchy blur profile has a sharp reading. Conceptually, this parallels the earlier result that a successor orbit together with a canonical Abel blur is equivalent to the existence of N at the finite–infinite interface. Here, the rationals plus Cauchy blur are equivalent to the existence of R at the discrete–continuous interface. In the final section we sketch how both examples fit into the same blur–idealization pattern: an object is obtained as the limit of its blur approximants, and blur is the universal mediator between the finite description and the infinite ideal. 2 Introduction The standard story says that the reals Rarise from the rationals Qby •completing Qin the metric d(q, r) = |q−r|(Cauchy completion), or •filling the “gaps” in Qvia Dedekind cuts. Both constructions are rigorous, but conceptually they hide a common theme: we build R by taking Q , looking at it through ever finer lenses, and reading off the object that lives behind all these blurred views. This paper makes that idea explicit. We introduce a simple blur scheme on the rational line, show that a real number is nothing but a coherent family of rational blur-classes across all scales, and prove that the usual construction of R is equivalent to the sharp limit of this blur diagram. The guiding analogy is the result for the natural numbers: in earlier work we showed that on a successor orbit the existence of a canonical Abel–blur is equivalent to the existence of N itself at the finite–infinite interface. Probability/distribution sits exactly where the finite string of successors passes to the infinite tail. Here we take the same viewpoint for Q⊂R: •blur corresponds to a finite observational precision ε > 0, •an ε–real is what you can distinguish at blur εusing only rationals, and •a true real number is a compatible choice of such ε–reals for all scales. The main contribution is not a new set-theoretic construction of R (we stay entirely within standard analysis), but a conceptual rephrasing: The real line is the blur completion of the rationals. It is the unique space you get by insisting that Cauchy coherence across all blur scales has a sharp reading. 1 We work concretely and keep the categorical packaging light; the very last section sketches how this fits into the general Category of Blur alongside the Abel–blur story for N. 2.1 Notation We write Q for the rationals, R for the reals, and N = { 0 , 1 , 2 ,...} . The standard metric on Q is d(q, r) = |q−r|. The closed ball of radius εabout qis Bε(q) := {r∈Q:|r−q| ≤ ε}. All blur scales will be positive rationals ε > 0; for convenience we often restrict to the discrete net ε= 2−n. 3 Blur scales on Q We begin by making precise what it means to “look at Qthrough a blur ε > 0”. 3.1 Local indistinguishability at scale ε Fix ε > 0. Heuristically, two rationals q, r ∈Q are ε –indistinguishable on a bounded window [−M, M]if |q−r|≤εand |q|,|r|≤M. However, the raw condition |q−r| ≤ ε is not transitive on Q , hence it does not define an equivalence relation and cannot be used directly to form a quotient into “cells.” To obtain an honest quotient (a genuine partition into cells), we fix a gauge (anchor) and partition the line into half-open buckets of length 2 ε . This is the minimal extra choice needed to turn the blur intuition into a well-defined ε –view. This bucket model agrees with the raw metric blur away from grid boundaries; the only discrepancy is a harmless boundary artifact (which disappears in a gauge-invariant formulation, or by restricting to readable profiles). For a moment it is useful to think in terms of a fixed observation window [ −M, M ]; later we let M→ ∞ and package everything systematically. Definition 3.1 ( ε –cells on a bounded window).Fix M > 0and ε > 0, and fix the gauge (anchor) a= 0. Define an equivalence relation on QM:= Q∩[−M, M] by q∼ε,M r⇐⇒ jq−a 2εk=jr−a 2εk. An equivalence class [q]ε,M is called an ε–cell in the window [−M, M], and the quotient set Qε,M := QM/∼ε,M is the set of ε–views of Qon that window. The order on Qε,M is the natural one induced by the line: [q]ε,M ≤[q′]ε,M :⇐⇒ ∀x∈[q]ε,M ∀y∈[q′]ε,M , x ≤y. Thus Qε,M is what an observer with resolution ε sees if they only care about rationals between −M and M . Concretely, each ε –cell is the intersection of QM with a half-open interval of length 2ε: [q]ε,M =QM∩a+ 2εk, a + 2ε(k+ 1)for the unique k∈Zdetermined by q. 2 In particular, inside the window every cell is a short interval of length at most 2 ε in which all rationals are observationally indistinguishable. Remark (gauge dependence). One could allow other anchors a∈ [0 , 2 ε ), producing translated grids; in this paper we fix a= 0 for concreteness. Remark 3.2 (Finite vs. infinite).For fixed ( ε, M )there are only finitely many cells: essentially ∼ 2 M/ε . As we let M→ ∞ and ε→ 0the number of cells explodes. A real number will be encoded by which cell it belongs to at each scale. 3.2 Coarsening maps between scales If ε′≤ε , then ε′ –indistinguishability is finer than ε –indistinguishability, so one expects a “forgetful” map from ε′ –views to ε –views. Because our ε –cells depend on a chosen gauge (anchor) for the 2ε–grid, we phrase this correctly as an existence statement: once we choose compatible gauges across scales, the coarsening maps exist and behave functorially. Lemma 3.3 (Coarsening maps (gauge-compatible version)).Fix M > 0. Let 0 < ε′≤ε . Assume moreover that ε/ε′∈N (equivalently: each 2 ε –bucket is a union of 2 ε′ –buckets). Choose gauges aε=aε′= 0 getting automatically aε′≡aε(mod 2ε′).(1) (Equivalently: every boundary of the 2 ε –grid is also a boundary of the 2 ε′ –grid, i.e. the 2 ε′ –grid refines the 2ε–grid.) Then there exists a well-defined surjection πε′,ε,M :Qε′,M →Qε,M ,[q]ε′,M 7−→ [q]ε,M . Moreover, if 0 < ε′′ ≤ε′≤ε since the gauges are chosen compatibly then the coarsening maps satisfy πε,ε,M = id, πε′′ ,ε,M =πε′,ε,M ◦πε′′ ,ε′,M . Proof. By definition of our cells (with gauge aη= 0), q∼η,M r⇐⇒ jq−aη 2ηk=jr−aη 2ηk. Assume q∼ε′,M r . Then q and r lie in the same 2 ε′ –bucket. Under the compatibility aε′≡aε ( mod 2 ε′ ), each 2 ε –bucket is a union of 2 ε′ –buckets (refinement), hence q and r must also lie in the same 2 ε –bucket. Therefore q∼ε,M r , so [ q ] ε,M = [ r ] ε,M and the assignment [ q ] ε′,M 7→ [ q ] ε,M is well-defined. Surjectivity is immediate: every ε –cell contains rationals and hence has a preimage ε′–cell. For the compatibility identities, note that πε,ε,M is the identity by definition, and the composition law follows because “forgetting from ε′′ to ε ” is the same as “forget from ε′′ to ε′ and then from ε′to ε” whenever the grids are nested. Remark 3.4 (A convenient directed scale set).A simple way to enforce (1) without thinking is to work on a fixed refinement chain, e.g. I={εn:= 2−n:n∈N}, and choose a single anchor a = 0 once and for all, then set aε = a = 0 for every ε∈I . Then for ε′≤ε we have aε′ = aε holds automatically, hence aε′≡aε ( mod 2 ε′ ), so the coarsening maps πε′,ε,M are well-defined. Along the dyadic chain every 2 εn+1 –grid refines the 2 εn –grid, so the maps πεn+1,εn,M exist (with this fixed anchor) without further choices. 3 Thus for each fixed Mwe obtain a blurred diagram B(M):I→Set, ε 7→ Qε,M ,(ε′≤ε)7→ πε′,ε,M , where I can be taken as the full directed poset of positive rationals equipped with a compatible gauge choice, or simply the chain I={2−n:n∈N}. 3.3 Removing the window To recover the full line we let M→ ∞ . There are (at least) two equivalent ways to formalize this; we choose the one most faithful to the blur philosophy: we keep the same cell mechanism, but drop the cutoff. Definition 3.5 (Global ε–cells (with a fixed gauge)).Fix ε>0and fix once and for all gauge (anchor) aε aε= 0. Define an equivalence relation ∼ε on Q by declaring that q∼εr iff q and r fall into the same 2ε–bucket of the anchored grid, i.e. q∼εr⇐⇒ jq−aε 2εk=jr−aε 2εk. Let Qε:= Q/∼ε be the set of global ε–cells. Now let 0 < ε′≤ε and assume we chose gauges aε′ = aε = 0 which are instantly compatible in the refinement sense aε′≡aε(mod 2ε′). Then the coarsening map πε′,ε :Qε′→Qε,[q]ε′7−→ [q]ε is well-defined and surjective. This is the M→ ∞ version of the bounded-window construction: we now ignore any cutoff on |q|and blur the entire rational line at resolution ε. Remark 3.6 (Why we do not use |q−r|≤ε globally).On Q the raw relation |q−r|≤ε is not an equivalence relation (it fails transitivity), so it cannot be used to define a quotient Qε without further modification. One can repair it in several equivalent ways: • (Anchored buckets) Fix a gauge and partition Q into half-open intervals of length 2 ε as above. • (Transitive closure) Take the transitive closure of |q−r| ≤ ε , which again produces intervallike cells. •(Chains) Declare q≈εriff there is a finite chain q=q0, . . . , qk=rwith |qj+1 −qj| ≤ ε. All of these yield “blur cells” of diameter ≤ 2 ε and differ only by harmless conventions (choice of endpoints / anchors). For definiteness, we keep the anchored-bucket model because it makes coarsening across scales transparent. With compatible gauge choices across scales, the family ( Qε, πε′,ε )forms a projective system (inverse system) over the directed set of blur scales (e.g. ε= 2−n). We now study its limit. 4 Remark 3.7 (Standing convention: scales and compatible gauges).From here on we fix a directed set of scales I and a compatible choice of gauges {aε = 0 }ε∈I so that whenever ε′≤ε in Iwe immediately have aε′≡aε(mod 2ε′), hence the coarsening maps πε′,ε are well-defined. For concreteness (and because it is sufficient to recover R) one may take the dyadic chain I={2−n:n∈N}therefore a2−n= 0 for all n. All projective limits below are taken over this fixed system. 4 Blur profiles and their limit Intuitively, an ε–real is a choice of one ε–cell. A blur profile is a consistent choice of ε–cells for all ε>0. Definition 4.1 (Blur profile).(Under the standing convention of Remark 3.7.) A blur profile is a family (cε)ε∈Iwith cε∈Qεfor each ε∈I, such that πε′,ε(cε′)=cεwhenever ε′≤εin I. The set of blur profiles is the projective limit b Q:= lim ←− ε∈I Qε. Every real number x∈R yields such a profile: at scale ε we simply choose the unique ε –cell that contains x , or at least a rational approximant of x inside that cell. Conversely, a consistent profile tells us, for each ε , which short interval our hypothetical real number must lie in. Intersection of all these nested intervals picks out a unique point. Let us make this precise. 4.1 Real numbers produce coherent profiles Because each Qε is built from an anchored grid on the line, a real number x∈R determines its ε–cell simply by asking which bucket of the grid contains x. For ε∈I, let the (half-open) ε–bucket intervals in Rbe Jε,k := aε+ 2εk, aε+ 2ε(k+ 1)(k∈Z). Each class c∈Qε corresponds to exactly one such k (namely the common value of ⌊ ( q−aε ) / (2 ε ) ⌋ for qin that class), and we write Jε(c)for the associated interval. Definition 4.2 (Canonical ε –cell of a real).For x∈R and ε∈I , define cε ( x ) ∈Qε to be the unique class whose associated interval contains x, i.e. x∈Jε(cε(x)). (With half-open buckets this is unambiguous even when xlies on a grid boundary.) 5 4.2 Profiles determine a real number Lemma 4.3 (Nested bucket intervals).Let ( cε ) ε∈I be a blur profile. For each ε∈I let Jε := Jε ( cε ) ⊂R be its associated bucket interval, and let Jε denote the closure of Jε in R . Then: (a) if ε′≤εin I, then Jε′⊆Jε(nestedness); (b) the diameter satisfies diam(Jε)≤2ε, hence tends to 0as ε→0along I. Proof. (a) Coherence says πε′,ε ( cε′ ) = cε . Under our bucket model this means exactly: the ε′ –bucket corresponding to cε′ is contained in the ε –bucket corresponding to cε . Taking closures preserves inclusion. (b) Each bucket interval has length 2ε, so its closure has diameter ≤2ε. Proposition 4.4 (Existence of a limiting point).Let ( cε ) ε∈I be a blur profile and let ( Jε ) ε∈I be the nested closed intervals from Lemma 4.3. Then there exists a unique point x∈Rsuch that \ ε∈I Jε={x}. Proof. Choose a decreasing sequence εn↓ 0in I (e.g. εn = 2 −n ). Then ( Jεn )is a nested sequence of nonempty closed intervals with lengths ≤ 2 εn→ 0. By completeness of R , the intersection TnJεn consists of a single point x . Nestedness implies x lies in every Jε with ε∈I , hence the full intersection over Iis {x}. Definition 4.5 (Reading map).Define the reading map Ψ : b Q→R by sending a blur profile (cε)ε∈Ito the unique x∈Rgiven by Proposition 4.4. Definition 4.6 (Readable blur profiles).Let ( cε ) ε∈I be a blur profile and let x = Ψ(( cε )) be its reading. We say (cε)is readable if x∈Jε(cε)for every ε∈I. Write b Qread := {(cε)∈b Q: (cε)is readable}. Remark 4.7 (Boundary pathology and why we ignore it).A coherent profile may converge to a point that sits on the right boundary of its half-open buckets at infinitely many scales. Such a profile has the same closure-reading as the adjacent profile on the other side, so Ψon all of b Q need not be injective. The readable subspace b Qread removes exactly this grid-boundary artifact (and depends only on the chosen half-open convention). Remark 4.8 (The anchor point causes no boundary pathology).In our dyadic system with anchor a = 0, the point 0lies on a grid boundary for every scale ε . However, since we use left-closed, right-open buckets Jε,k = [a+ 2εk, a + 2ε(k+ 1)), the boundary point a is always included, i.e. a∈Jε,0 = [ a, a + 2 ε )for every ε . In particular, 0 produces a readable blur profile. More generally, boundary ambiguity can only occur at right endpoints of the half-open buckets, exactly as described above. Lemma 4.9. For each x∈R , the family cε ( x ) ε∈I is a blur profile and is readable. Hence the map Φ:R→b Qread,Φ(x) := cε(x)ε∈I, is well-defined. 6 Proof. By definition of cε ( x )we have x∈Jε ( cε ( x )) for every ε∈I , so the family is readable. It remains to check coherence. Let ε′≤ε in I . By gauge compatibility (Remark 3.7), the ε′ –grid refines the ε –grid, so every ε′ –bucket is contained in a unique ε –bucket. Since x lies in the ε′ –bucket Jε′ ( cε′ ( x )), it follows that x also lies in the unique containing ε –bucket, which is precisely Jε(cε(x)). Therefore πε′,εcε′(x)=cε(x), so the family is coherent. 4.3 Equivalence between reals and blur profiles We now show that Φand Ψare inverse isomorphisms. Theorem 4.10 (Blur completion of the rationals).The maps Φ:R→b Qread,Ψ : b Qread →R are inverse bijections: Ψ ◦ Φ = idR and Φ ◦ Ψ = idb Qread . In particular, the real line R is canonically isomorphic to the readable blur-limit b Qread. Proof. Let x∈R . By definition, Φ( x ) = ( cε ( x )) ε∈I where x∈Jε ( cε ( x )) for every ε . Thus x∈Jε ( cε ( x )) for every ε , so the nested intersection defining Ψ(Φ( x )) contains x . By uniqueness of the intersection point (Proposition 4.4), we get Ψ(Φ(x)) = x. Conversely, let (cε)ε∈Ibe a blur profile and put x= Ψ((cε)). Then by construction x∈Jε ( cε )for every ε , and since ( cε )is readable we also have x∈Jε ( cε ) for every ε . By definition of cε ( x )as the unique bucket containing x , we conclude cε ( x ) = cε for all ε∈I, hence Φ(Ψ((cε))) = (cε). Remark 4.11 (Metric structure).Because cells come from a fixed anchored grid, there is an unavoidable boundary effect: two points can be arbitrarily close yet lie in adjacent ε –buckets. What is always true is the scale control: if cε ( x ) = cε ( y )then x, y lie in the same interval of length 2 ε , hence |x−y|< 2 ε ; and if |x−y|> 2 ε then cε ( x )  = cε ( y ). Thus the usual metric is recovered as the sharp limit of blur views up to this grid-boundary convention (and it becomes exact if one works in a gauge-invariant formulation, i.e. allowing all translations of the grid). 5 Cauchy, cuts, and blur coherence The blur completion just constructed is, unsurprisingly, equivalent to the classical Cauchy and Dedekind constructions. The interest lies in how naturally these fit into the blur language. 5.1 Cauchy sequences as blur data Recall that a sequence (qn)n∈Nof rationals is Cauchy if ∀ε>0∃N∀m, n ≥N:|qm−qn|≤ε. Lemma 5.1 (Cauchy data induces a blur profile).Let ( qn ) n∈N be a Cauchy sequence in Q with (classical) limit x∈R . For each ε∈I , since ( qn )is Cauchy, there exists N ( ε )such that all qn with n≥N(ε)lie in a single ε–bucket. Let cε(q) := [qN(ε)]ε∈Qεbe that eventual ε–cell. Then (cε(q))ε∈Iis a blur profile and its reading is Ψ(cε(q))ε∈I=x. In general this induced profile need not equal the canonical (readable) profile Φ( x ), but it always has the same reading. 7 Proof. For each fixed ε , Cauchy-ness implies ( qn )is eventually trapped in one ε –bucket, so cε ( q ) is well-defined. Compatibility under coarsening follows because finer buckets sit inside unique coarser buckets (on the dyadic chain with compatible gauges). Let Jε := Jε ( cε ( q )). Since qn→x and Jε is the bucket that contains all sufficiently large qn , we must have x∈Jε for every ε . Hence x lies in the nested intersection Tε∈IJε , so by Proposition 4.4 the reading is x. Conversely, given a blur profile ( cε ), we can choose a sequence of scales εn↓ 0and pick a representative qεn for each cεn . The nested-interval argument shows that ( qεn )is Cauchy in Q and converges to the real Ψ((cε)). Proposition 5.2 (Cauchy completion and readable profiles).Every Cauchy sequence in Q determines a real limit x∈R , hence a canonical readable profile Φ( x ) ∈b Qread . This identifies the usual Cauchy completion of Qwith b Qread ∼ =R. Proof. Standard analysis identifies the Cauchy completion of Q with R via limits. By Theorem 4.10,R∼ =b Qread via x7→ Φ(x). Composing gives the claimed identification. Thus the blur picture is not a rival to the Cauchy completion but a repackaging: the phrase “Cauchy consistent across all blur scales” is literally the definition of a blur profile. 5.2 Dedekind cuts as blur shadows A Dedekind cut is a downward-closed A⊂Q with no greatest element, representing the set of rationals below some real. From the blur viewpoint, we can associate to each profile ( cε )the subset A:= {q∈Q:∃ε∈Iwith [q]εlying strictly to the left of cε}. This gives a cut whose corresponding real is precisely Ψ(( cε )). We do not spell out the full correspondence here; the key point is that the usual algebraic/topological structure of R can be read entirely through blur-consistency of rational information. 6 Completeness as blur readability One of the main virtues of R is completeness: every Cauchy sequence converges. In the blur language, this is nothing but the statement that every coherent blur profile has a sharp reading. Principle 6.1 (Blur readability).Every blur profile ( cε ) ∈b Q has a (unique) reading x = Ψ((cε)) ∈Robtained from the nested closed buckets. Moreover, the restriction Ψ : b Qread →R is bijective, with inverse Φ : R→b Qread. Theorem 4.10 is precisely this principle made formal. Remark 6.2 (Why blur is the right level).Instead of keeping track of individual Cauchy sequences, which are heavily redundant and depend on parametrization, blur profiles work directly at the level of “what is distinguishable at scale ε ”. This is closer to how physical measurements behave: we never see the exact rational value, only an interval of possibilities. The completeness of R then says: if these interval-valued observations are coherent across all scales, there really is a unique underlying point. 8 Remark 6.3 (Compatibility with nonstandard analysis).In nonstandard analysis, real numbers can also be seen as equivalence classes of hyperrational sequences q∗ n with infinitesimal differences. Our blur is more modest: it stays entirely inside the standard rationals, and the “infinitesimal” is encoded by the blur scale ε rather than by a new number. Philosophically, both approaches formalize the same intuition; the blur version has the advantage of not leaving the original universe and being directly tied to observational resolution. 7 Comparison with the Abel–blur for N In previous work at the finite–infinite interface, we studied a successor orbit O:= {x, Sx, S2x,...} with S an injective map (successor) and showed that the existence of a canonical Abel blur on O is equivalent to the existence of the natural numbers themselves. Concretely, the Abel blur at parameter q∈(0,1) attaches to xthe distribution µq= (1 −q)X n≥0 qnδSnx, and one proves that the family ( µq ) q→1− is coherent in a sense similar to our blur profiles. The limit object encodes the infinite tail of the successor orbit, i.e. the structure of N. The present paper shows that the passage from Q to R is structurally the same: instead of a successor map we have the metric, and instead of geometric tails we have Cauchy tails. In both cases: • there is a family of blur views indexed by a scale parameter (Abel parameter q or metric scale ε), •aprofile is a compatible choice across all scales, •each profile has a unique sharp reading, and •the collection of those readings is precisely the usual infinite object (Nor R). This is the sense in which the reals are the blur completion of the rationals: they are the unique space in which every blur profile arising from Q has a sharp, coherent interpretation—and nothing more. 8 Conclusions and outlook We have shown that the usual construction of R from Q can be written entirely in terms of blur: •At each fixed scale ε > 0, we form ε–cells on Qand the quotient Qε. • A real number is a readable coherent blur profile: a consistent choice of an element of Qε for all ε, compatible under coarsening. • The map that reads off the unique point living in the intersection of the associated nested intervals is a bijection b Qread ∼ = −−→ R. This matches, one level up, what happens for N with the Abel blur: there, a successor orbit plus a canonical distributional blur gives you the natural numbers; here, the rational line plus a metric blur gives you the reals. In both cases blur is not an optional decoration but the exact mechanism that bridges a finite description with an infinite idealization. 8.1 Categorical packaging (sketch) In the language of the Category of Blur, one can summarize both constructions as follows: • Ablurred object is a diagram BX : I→ C assigning to each blur scale σ∈I an approximant BσXand to each refinement σ′⪯σa coarsening map Bσ′X→BσX. 9