--- title: "OMEGA as a Reference Language" subtitle: "Representation, relationships, exact recovery, and necessary limits" date: "24 October 2026" author: "Mathematical development prepared in conversation with Matthew Alan Newman" lang: en --- \clearpage # The statement being proved The proposal considered here is that a known OMEGA geometry can supply a common reference for representing other things, expressing their relationships, and recovering information after changes of representation. There are two different universality statements that can be proved. **Geometric universality within the reference plane.** Three noncollinear reference points already present in the marked OMEGA construction locate every point of its plane from exact distances. Consequently they determine every Euclidean relation computed from those recovered coordinates. **Universality for finite formal information.** Scaled copies of one fixed normalized OMEGA kernel can carry every finite binary record injectively. Every finitely presented finite relational structure can therefore be represented and recovered. Relations, operations, and first-order truth can be transported exactly through a lossless representation, with quantifiers restricted to its image. Every finite simple graph also has a direct geometric realization in which adjacency is a distance predicate. Finite Euclidean configurations can be recovered up to reference-preserving symmetry when their required mutual distances are supplied. Time histories can supply additional separating measurements when their observation map has sufficient rank. These statements concern information supplied through measurements or an encoding. The known shape does not supply the unknown measurements, interpretations, or relation tables. This distinction is part of the theorem, because a reference system and the state represented in that system play different mathematical roles. The note gives complete written arguments, including explicit obstructions to stronger statements. Its methods include standard Euclidean reconstruction, polynomial interpolation, transport of structure, logarithmic coordinates, and witness completion. No priority or novelty claim is made for these methods or their use here. This is a new development in this conversation, not an amendment of a published OMEGA paper or a certification of the entire website. # The reference geometry Write \[ \varphi=\frac{1+\sqrt5}{2},\quad \varepsilon_0=\frac{\pi}{648000},\quad s_0=\sin(\varepsilon_0/2),\quad c_0=\cos(\varepsilon_0/2). \] For the general geometric family let $0<\varepsilon<\pi$, $s=\sin(\varepsilon/2)$, and $c=\cos(\varepsilon/2)$. Thus $00$, and $s^2+c^2=1$. Define \[ A=s+\varphi^2=s+\varphi+1,\qquad B=s+1, \qquad x=(-A,-B,-s,s,B,A). \] The six normalized baseline coordinates are distinct, sum to zero, and have consecutive gaps \[ \varphi,\quad 1,\quad 2s,\quad 1,\quad\varphi. \] Let $o$ be the circle center, $a>0$ its radius, and $(u,v,n)$ an oriented orthonormal frame, with $n=u\times v$. Here $u$ points toward the middle of the opening and $v$ along the ordered baseline. The six physical marks are \[ p_i=o+a(cu+x_i v),\qquad i=1,\ldots,6. \tag{1} \] The circular body is the retained major arc \[ \gamma(t)=o+a(\cos t\,u+\sin t\,v), \quad \varepsilon/2\leq t\leq2\pi-\varepsilon/2, \] and the feet are the two outward baseline intervals whose consecutive added lengths are $a$ and $a\varphi$. The middle seed chord is a measurement, not an added segment of the arc-and-feet curve. The canonical reference fixes $a=1$, $\varepsilon=\varepsilon_0$, and a declared pose. Scaled or translated instances are descriptions relative to that fixed reference. Their variable parameters carry information; the fixed reference itself is not a variable message. The strict endpoint restriction above is needed for the rank-two landmark statement below. At $\varepsilon=\pi$, $c=0$ and the center-augmented landmark Gram rank drops. The probability-family statement later can include $s=1$ as an algebraic boundary without claiming rank two there. ## Theorem G1: reconstruction of the marked kernel Given the center $o$, the ordered circle seed points $d,b$, their common radius $a$, their separation $0<\varepsilon<\pi$, an ambient orientation in three dimensions, and the declared extensions, the marked kernel and retained major arc are uniquely determined. **Proof.** Put $r_d=d-o$ and $r_b=b-o$. Equal radii imply \[ (r_b+r_d)\cdot(r_b-r_d)=\|r_b\|^2-\|r_d\|^2=0. \] The sum and difference have norms $2ac$ and $2as$, both positive. Hence \[ u=\frac{r_b+r_d}{2ac},\qquad v=\frac{r_b-r_d}{2as},\qquad n=u\times v \] give the required frame. The two seed equations are $d=o+a(cu-sv)$ and $b=o+a(cu+sv)$. Extending successively by $a$ and $a\varphi$ gives exactly (1), since $1+\varphi=\varphi^2$. The circle, seed ordering, and selection of the major arc specify the circular component. Every construction step has a unique result. If the absolute pose is omitted, uniqueness is up to the corresponding orientation-preserving Euclidean isometry. $\square$ Distances and angles among these points follow from the same data. For instance, \[ \|p_i-p_j\|=a|x_i-x_j|, \quad (p_i-o)\cdot(p_j-o)=a^2(c^2+x_ix_j). \] Thus the center-angle cosine is \[ \frac{c^2+x_ix_j}{\sqrt{c^2+x_i^2}\sqrt{c^2+x_j^2}}. \] The normalized landmark Gram matrix is $c^2\mathbf1\mathbf1^T+xx^T$. Since $\sum x_i=0$, $c>0$, and $x\ne0$, its rank is two: the two orthogonal columns $c\mathbf1$ and $x$ are nonzero. These are consequences of the construction data, not extra independent measurements. # Locating other things relative to OMEGA ## Theorem G2: every point in the reference plane is determined Use the three reference points \[ o,\qquad p_-=o+a(cu-Av),\qquad p_+=o+a(cu+Av). \] They are the circle center and two actual outer marks. For a point $P$ in the reference plane, let \[ R_0=\|P-o\|^2,\quad R_-=\|P-p_-\|^2, \quad R_+=\|P-p_+\|^2. \] Then $P=o+Xv+Yu$, where \[ \boxed{X=\frac{R_--R_+}{4aA}}, \qquad \boxed{Y=\frac{2R_0+2a^2(c^2+A^2)-R_--R_+}{4ac}}. \tag{2} \] **Proof.** Expanding the squared distances gives \[ \begin{aligned} R_0&=X^2+Y^2,\\ R_-&=X^2+Y^2+a^2(A^2+c^2)+2aAX-2acY,\\ R_+&=X^2+Y^2+a^2(A^2+c^2)-2aAX-2acY. \end{aligned} \] Their difference yields the first formula; the sum together with $R_0$ yields the second. Both denominators are positive. Any two planar points with the same three measurements therefore have the same $X,Y$ and are equal. $\square$ The theorem proves injectivity on the actual measurement image. It does not assert that every arbitrary triple of nonnegative numbers is a consistent distance triple. To validate a candidate triple, compute (2) and verify the three defining squared-distance equations. ## Corollary G3: all planar Euclidean relations are recoverable For two represented points $P,Q$, \[ \|P-Q\|^2=(X_P-X_Q)^2+(Y_P-Y_Q)^2. \] Dot products, incidence, collinearity, angles where defined, and every other relation specified as a function or predicate of the recovered planar coordinates can therefore be evaluated from the records. A planar point set or a parameterized planar curve is reconstructed pointwise from its complete records. A finite list of samples reconstructs only that list unless a further interpolation model is specified. This is a precise sense in which one known OMEGA reference supports all locations and geometric relationships in its plane. The unknown point's three measured distances remain input data. ## Conditioning of the measurements If each squared-distance error has absolute value at most $\eta$, while the reference parameters are exact, (2) gives \[ |\Delta X|\leq\frac{\eta}{2aA},\qquad |\Delta Y|\leq\frac{\eta}{ac}. \] Using the inner seed pair in place of the outer pair replaces $A$ by $s$ in the first bound. The ratio of those bounds is $A/s$. Thus a very small canonical opening does not force the same poor horizontal conditioning when the outer reference marks are available. This bound is for errors in squared distances; a bound for raw distances must first account for squaring and for their magnitudes. ## Theorem G4: planar references have an exact higher-dimensional ambiguity Let $H$ be the reference plane in any real inner-product space. Write \[ P=o+Xv+Yu+w,\qquad w\perp H. \] Distances from $P$ to every point of the planar OMEGA determine $X,Y$ and $\|w\|$, but in general do not determine the direction of $w$. **Proof.** For every $q\in H$, \[ \|P-q\|^2=\|o+Xv+Yu-q\|^2+\|w\|^2. \tag{3} \] Formula (2) still recovers $X,Y$, because the extra $\|w\|^2$ cancels. Then $\|w\|^2=R_0-X^2-Y^2$. Every orthogonal transformation of $H^\perp$ leaves (3) unchanged. In three dimensions, $w=Zn$ and $w=-Zn$ are indistinguishable when $Z\ne0$. In larger dimensions there are further indistinguishable perpendicular directions. $\square$ This ambiguity can affect relationships between represented objects. Take $P=o+n$ and either $Q_+=o+2n$ or $Q_-=o-2n$. The two choices of $Q$ have identical distances to every planar reference point. Nevertheless, \[ \|P-Q_+\|=1,\qquad \|P-Q_-\|=3. \] The full planar distance descriptions of the individual objects therefore do not determine all their mutual spatial relationships. An additional off-plane probe $q_3=o+an$ removes the three-dimensional sign ambiguity. If $R_3=\|P-q_3\|^2$, then \[ Z=\frac{R_0-R_3+a^2}{2a}. \] This probe is constructible from a retained oriented frame, but it is additional measurement data and is not a point of the original planar kernel. ## Corollary G5: the infinite-dimensional version requires more probes In a separable real Hilbert space with a specified complete orthonormal basis $(e_j)$, the distances to $o$ and $o+ae_j$ recover every coordinate: \[ x_j=\frac{\|P-o\|^2+a^2-\|P-(o+ae_j)\|^2}{2a}. \] Then $P=o+\sum_jx_je_j$, with convergence in the Hilbert norm. The OMEGA plane can provide two basis directions; a complete additional basis is not implied by that plane. With only finitely many probes, unseen orthogonal directions can remain. For the first $N$ basis probes, the points $o+e_k$, $k>N$, all have the same distance profile to those probes and to $o$. ## Theorem G6: complete finite configurations are determined up to reference-preserving isometry Suppose finitely many points $P_i$ in a common Euclidean space have their distances to the three reference anchors and all their mutual distances supplied. Those data determine the configuration up to an orthogonal motion that fixes the reference plane pointwise. **Proof.** Formula (2) determines each planar projection $q_i=o+X_iv+Y_iu$. Put $w_i=P_i-q_i$. The center distances give $\|w_i\|^2$. The mutual distances give \[ \|w_i-w_j\|^2=\|P_i-P_j\|^2-\|q_i-q_j\|^2, \] and hence the full perpendicular Gram matrix \[ G_{ij}=\frac{\|w_i\|^2+\|w_j\|^2-\|w_i-w_j\|^2}{2}. \] If another configuration has the same data, its perpendicular vectors $w'_i$ have the same Gram matrix. The assignment $T(\sum_i a_iw_i)=\sum_i a_iw'_i$ is well defined: a combination is zero exactly when its squared norm $a^TGa$ is zero, which holds for the other combination as well. The same Gram equality shows that $T$ preserves inner products and maps the two spans onto each other. In the same finite-dimensional ambient space, extend orthonormal bases of these spans to bases of the perpendicular space. This extends $T$ to an orthogonal map there. Adjoin the identity on the reference plane. The resulting motion fixes the entire OMEGA reference and sends every $P_i$ to its counterpart. $\square$ In a fixed Hilbert ambient space with only finitely many represented points, the same extension argument applies with a complete orthonormal basis. The remaining symmetry is a common symmetry of the entire configuration, not a separate arbitrary motion of each point. If oriented handedness outside the reference plane is required, distances alone may leave a reflection and an oriented measurement is needed. This theorem uses the supplied mutual relations to resolve exactly the ambiguity identified by G4. # Relations as a complete metric description ## Theorem M1: a complete distance profile is isometric For any metric space $(X,d)$ and anchor $o\in X$, define the bounded function \[ \Phi(x)(t)=d(x,t)-d(o,t),\qquad t\in X. \] Then \[ \boxed{\|\Phi(x)-\Phi(y)\|_\infty=d(x,y)}. \tag{4} \] **Proof.** The triangle inequality gives $|\Phi(x)(t)|\leq d(x,o)$, so each profile is bounded. It also gives $|d(x,t)-d(y,t)|\leq d(x,y)$, proving the upper bound in (4). At $t=y$, the absolute difference is exactly $d(x,y)$. This proves equality and injectivity. $\square$ Thus relations to a sufficiently rich reference family can determine an object and all pairwise metric distances. The reference family here is all of $X$. Equation (4) does not assert that an arbitrary fixed finite shape supplies all those probes. ## Theorem M2: a finite net gives a quantitative approximation For a nonempty reference subset $A\subseteq X$, set \[ d_A(x,y)=\sup_{t\in A}|d(x,t)-d(y,t)|. \] Always $d_A(x,y)\leq d(x,y)$. If some $t\in A$ satisfies $d(x,t)\leq\delta$, then \[ d(x,y)-2\delta\leq d_A(x,y)\leq d(x,y). \] **Proof.** The upper bound is the reverse triangle inequality. For the chosen probe, \[ d(y,t)-d(x,t)\geq d(x,y)-2d(x,t)\geq d(x,y)-2\delta. \] Taking the supremum of absolute differences proves the result. In particular an entire $\delta$-net gives the bound for all pairs. A subset that fails to separate points gives only a pseudometric. $\square$ # The exact criterion for a reference language ## Theorem R1: when every relationship can be recovered Let $E:X\to Y$ be a representation and $Y_0=E(X)$. The following are equivalent: 1. $E$ is injective. 2. There is a decoder $D:Y_0\to X$ with $D(E(x))=x$. 3. Every function $h:X\to Z$, for every target set $Z$, has a transported version on $Y_0$ satisfying $h=\widetilde h\circ E$. 4. Every finitary relation on $X$ has an exactly equivalent relation on $Y_0$. **Proof.** Injectivity makes the inverse on $Y_0$ well defined, proving 1 implies 2. Under 2, define $\widetilde h=h\circ D$ and define \[ \widetilde R(y_1,\ldots,y_k) \iff R(D(y_1),\ldots,D(y_k)). \tag{5} \] These prove 3 and 4. Under 3 take $h=\operatorname{id}_X$, obtaining a decoder. A decoder implies injectivity by applying it to $E(x)=E(x')$. Finally, under 4 transport the equality relation. If $E(x)=E(x')$ for distinct points, the same encoded pair would have to represent both the true statement $x=x$ and the false statement $x'=x$. This is impossible. $\square$ The criterion concerns the information in the complete representation, including retained context. It says neither that a convenient formula for every transported relation is already known nor that every relation is computable. ## Theorem R2: the exact criterion when only selected observables matter For a particular function $h:X\to Z$, an exact factorization $h=\widetilde h\circ E$ exists if and only if $h$ is constant on each fiber of $E$. **Proof.** A factorization gives identical $h$ values whenever the $E$ values agree. Conversely, define $\widetilde h(y)$ to be $h(x)$ for any $x$ with $E(x)=y$. Fiber constancy makes this independent of the chosen representative. $\square$ A representation can therefore be adequate for an explicitly chosen class of questions while discarding other information. Its scope is determined by those questions, not by a blanket label of losslessness. ## Theorem R3: operations and composition transport faithfully Under R1, for a map $f:X\to X$, define \[ \widetilde f=E\circ f\circ D:Y_0\to Y_0. \] Then $\widetilde f(E(x))=E(f(x))$, \[ \widetilde{g\circ f}=\widetilde g\circ\widetilde f, \qquad \widetilde{\operatorname{id}_X}=\operatorname{id}_{Y_0}. \] If $f$ is bijective, $\widetilde f$ is bijective with inverse $E\circ f^{-1}\circ D$. The same formulas hold for operations of any finite arity and for partial operations when the transported domain is retained. **Proof.** Substitute the definitions and cancel $D\circ E=\operatorname{id}_X$ at each adjacent pair. The domain restriction ensures that every occurrence of $D$ is defined. $\square$ ## Theorem R4: first-order truth is preserved Let a first-order structure $\mathcal A$ have domain $X$. Transport its constants, functions, and relations through an injective $E$ as above, obtaining a structure $\widetilde{\mathcal A}$ with domain $Y_0$. For every first-order formula $\theta$ and assignment $\nu$, \[ \mathcal A\models\theta[\nu] \iff \widetilde{\mathcal A}\models\theta[E\circ\nu]. \tag{6} \] **Proof.** First use induction on terms: variables and constants have their encoded values, and an encoded function applied to encoded arguments gives the encoded original value by R3. Equality is preserved by injectivity. Every other atomic relation is preserved by (5). Negation and Boolean connectives preserve equivalence. For an existential quantifier, a witness $x\in X$ gives the witness $E(x)\in Y_0$, and every witness in $Y_0$ has a unique decoded original. Universal quantification follows in the same way, or by negation and existential quantification. Structural induction proves (6) for all formulas. $\square$ This is a preservation theorem for formally specified meaning: an interpretation is already given and is transported. Quantifying over all of $Y$ instead of $Y_0$ can introduce spurious elements and invalidate the result. # A constructive universal finite carrier ## Theorem U1: every finite binary record has an OMEGA representation For a finite bitstring $w$ of length $\ell$, let $\operatorname{val}_2(w)$ be its value, allowing leading zeros and assigning the empty word value zero. Define \[ N(w)=2^\ell+\operatorname{val}_2(w),\qquad E_\Omega(w)=N(w)x^{(0)}, \tag{7} \] where $x^{(0)}$ is the fixed canonical normalized six-mark vector. This is a scaled canonical OMEGA baseline. It has a unique inverse on its image. **Proof.** The binary expansion of $N(w)$ is the leading sentinel bit $1$ followed by the entire word $w$. Removing that first bit recovers $w$, including its length and any leading zeros. The inner unit gap of $x^{(0)}$ is \[ x^{(0)}_3-x^{(0)}_2=(-s_0)-(-s_0-1)=1. \] Therefore an encoded word $y=E_\Omega(w)$ gives \[ \boxed{N(w)=y_3-y_2}. \tag{8} \] The sentinel decoder then recovers $w$. Different words have different positive integer scales. The corresponding entire scaled arc-and-feet kernel is also determined from that scale and the fixed reference. $\square$ For a byte string of length $m$, the equivalent integer code is $2^{8m}+\operatorname{int}_{256}(w)$. Its image contains only integers whose post-sentinel bit count is divisible by eight. The byte decoder checks that condition. The use of scale in (7) is deliberate. Encoding the record only as an additive translation and then applying a centered softmax would erase that record. A translation parameter must be retained separately if it is used as data. Equation (7) uses an unbounded family of possible scaled states and exact coordinates. It does not place arbitrarily many distinguishable messages in one fixed finite-capacity state. The scale $N$ has $\ell+1$ binary digits. A bounded alternative, such as $a=1/(N+1)$, moves the demand into measurement precision instead of removing it. ## Theorem U2: every finite relational structure can be represented Fix a finite relational signature with ordered relation symbols and their finite arities. For a structure with labeled domain $\{0,\ldots,m-1\}$, encode $m$ by a self-delimiting positive-integer code, followed by the truth table of each relation in lexicographic tuple order. Each table has the known finite length $m^k$ for its arity $k$. This produces an unambiguous finite bitstring and therefore an OMEGA carrier by U1. **Proof.** Read the self-delimiting header to recover $m$. The retained signature and order determine every remaining table length and tuple position. Read those bits and reconstruct exactly the original relation tables. There is no choice at any decoding step. Applying R4 to the resulting labeled domain representation preserves every first-order formula. $\square$ If the signature itself varies, its names, arities, and ordering must also be encoded or retained as explicit context. Distinct labelings are distinct records unless an isomorphism-canonicalization rule is separately adopted. This construction is a lossless representation of supplied relations; it obtains its relation data from the record. ## Corollary U3: effective countable domains and computable operations Suppose a domain $X$ has an explicit injective finite-record encoding $c$ with an effective decoder on its image. Then $E_\Omega\circ c$ represents $X$ injectively. If an operation or predicate on $X$ is computable, decoding, applying it, and re-encoding implements its transported form. Validity checks and partial domains remain part of that algorithm. Countability alone supplies no particular executable encoding for an arbitrary abstract set. ## Theorem U4: every finite simple graph is directly representable by geometric distances Let $A$ be the symmetric zero-diagonal adjacency matrix of a simple graph on $m\geq1$ labeled vertices. Set $\alpha=1/(2m)$ and $G=I+\alpha A$. There exist $m$ unit vectors $v_i$ in a real Euclidean space such that \[ \|v_i-v_j\|^2= \begin{cases} 2-2\alpha,&i\ne j\text{ and }A_{ij}=1,\\ 2,&i\ne j\text{ and }A_{ij}=0. \end{cases} \tag{8a} \] Thus adjacency is recovered by the single distance predicate $\|v_i-v_j\|^2<2-\alpha$ for distinct vertices. **Proof.** For every real vector $x$, the inequality $2|x_ix_j|\leq x_i^2+x_j^2$ gives $|x^TAx|\leq(m-1)\|x\|^2$. Consequently \[ x^TGx\geq(1-\alpha(m-1))\|x\|^2>0 \] when $x\ne0$. Therefore $G$ is positive definite and is a Gram matrix. To see the constructive factorization, take the positive square root of its first diagonal entry; subtract the corresponding rank-one Gram contribution. The remaining Schur complement is positive definite, because testing the original quadratic form with first coordinate chosen to minimize it gives exactly the Schur-complement quadratic form. Induction factors $G$ into a real Gram matrix of vectors. For rational $G$ this construction uses finitely many rational operations and positive square roots. Its diagonal entries are one, so the vectors are unit vectors. Their squared distances are $G_{ii}+G_{jj}-2G_{ij}$, giving (8a). $\square$ The vectors can be expressed relative to a frame extending the OMEGA reference frame. They initially lie in $\mathbb R^m$; centering gives affine dimension at most $m-1$. This is not a claim that all graphs admit the same two-distance representation inside OMEGA's plane. The adjacency data determine the geometric construction; geometry preserves those supplied relations. The exact squared-distance separation is $2\alpha$, so measurement error must respect that separation if the predicate is used numerically. # A single reference can be a generator The phrase "one known reference" also has a precise structural interpretation that does not require the reference to contain the target's data internally. ## Theorem L1: a unit line from OMEGA separates all linear maps Take the real line $L$ determined by one of OMEGA's unit baseline segments, with a chosen zero and unit vector $e$. In the category of real vector spaces and linear maps, $L$ is a generator: for linear $f,g:V\to W$, \[ f\circ h=g\circ h\text{ for every linear }h:L\to V \quad\Longrightarrow\quad f=g. \] **Proof.** For each $v\in V$, the map $h_v(te)=tv$ is linear. Evaluating the assumed equality at $e$ gives $f(v)=g(v)$. This holds for every $v$, hence the maps agree. $\square$ Every vector is recoverable as $h_v(e)$. Thus a fixed known line together with all its linear probes can describe every vector and distinguish every linear transformation. The values and collection of those probes carry the target information. The claim uses the full real-linear structure and the specified class of maps; it is not a statement about arbitrary similarities of the bare arc-and-feet curve. ## Theorem L2: full relational descriptions determine categorical objects For objects $X,Y$ in a locally small category, write $h_X(A)=\operatorname{Hom}(A,X)$. Any natural transformation $\eta:h_X\to h_Y$ has the form \[ \eta_A(u)=f\circ u \] for exactly one map $f:X\to Y$. If $\eta$ is a natural isomorphism, then $X$ and $Y$ are isomorphic. **Proof.** Let $f=\eta_X(\operatorname{id}_X)$. Naturality with respect to $u:A\to X$ says precisely that $\eta_A(u)=f\circ u$. Hence $f$ determines the transformation and is uniquely determined by it. Conversely every such $f$ defines a natural transformation by associativity of composition. If a natural inverse is induced by $g:Y\to X$, the two composite transformations induce $g\circ f=\operatorname{id}_X$ and $f\circ g=\operatorname{id}_Y$, so $f$ is an isomorphism. $\square$ This is the fully faithful representable case of the Yoneda principle. It depends on the coherent maps from all test objects. Restricting to one test object needs an independently established generator or stronger reconstruction property. L1 establishes the required separation property for real linear maps; it does not assert that every category or every empirical domain has the same property. # Explicit changes of mathematical description on the site ## Theorem C1: centered coordinates and positive probabilities are equivalent Let $H=\{z\in\mathbb R^6:\sum_i z_i=0\}$ and let $\Delta^\circ$ be the strictly positive probability simplex. For fixed $\beta>0$, define \[ S_\beta(z)_i=\frac{e^{\beta z_i}}{\sum_j e^{\beta z_j}}, \qquad L_\beta(p)_i=\frac{\log p_i-\frac16\sum_j\log p_j}{\beta}. \] Then $L_\beta S_\beta=\operatorname{id}_H$ and $S_\beta L_\beta=\operatorname{id}_{\Delta^\circ}$. **Proof.** Writing $Z=\sum_j e^{\beta z_j}$ gives $\log p_i=\beta z_i-\log Z$. The mean of these logarithms is $-\log Z$, since $z$ is centered. Subtraction and division by $\beta$ recover $z_i$. Conversely, exponentiating the centered logarithms produces $p_i$ multiplied by one common positive factor; normalization cancels that factor because $\sum_i p_i=1$. $\square$ The site declares the choice $\beta=\lambda$. The theorem holds for every fixed $\beta>0$, so it does not establish that this particular parameter is forced. On uncentered vectors, $S_\beta(z+t\mathbf1)=S_\beta(z)$, so the additive mean is lost. Centering restricts to a unique representative; retaining the mean as an extra field permits recovery of an uncentered source. ## Corollary C2: a specified metric is preserved Give the simplex the metric \[ d_\beta(p,q)=\frac1\beta \left(\sum_i\left[\log p_i-\log q_i -\frac16\sum_j(\log p_j-\log q_j)\right]^2\right)^{1/2}. \] Then $d_\beta(S_\beta(z),S_\beta(w))=\|z-w\|_2$. This follows immediately from the inverse formula. Raw Euclidean distances between probability vectors are not being identified with coordinate distances; the explicitly transported metric is the one that preserves them. ## Theorem C3: scaling corresponds to normalized powering For $r>0$, define $M_r(z)=rz$ and $P_r(p)_i=p_i^r/\sum_jp_j^r$. Then \[ S_\beta\circ M_r=P_r\circ S_\beta, \quad P_rP_t=P_{rt},\quad P_1=\operatorname{id},\quad P_r^{-1}=P_{1/r}. \] **Proof.** Substitute $p_i=e^{\beta z_i}/Z$. Every term in the powered normalization has the common factor $Z^{-r}$, which cancels, leaving $e^{\beta r z_i}/\sum_j e^{\beta r z_j}$. Applying the same cancellation twice gives the exponent product $rt$. The identity and inverse laws follow. Positivity and fixed ordering ensure the formulas and real powers are defined throughout. $\square$ ## Theorem C4: recognize and recover the six-mark family in probabilities For fixed $\beta>0$ and $p\in\Delta^\circ$, put $g_i=\log(p_{i+1}/p_i)$, $i=1,\ldots,5$. Then $p=S_\beta(ax)$ for some $a>0$ and $00,\quad g_1=g_5=\varphi g_2, \quad00$, then \[ F(rz)=D_r(F(z)). \] **Proof.** The finite geometric-series identity gives $\sum_{k=0}^5e^{2\pi i(j-l)k/6}=6$ when $j=l$ and zero otherwise. Substitution into the inverse DFT therefore yields $w_j$ exactly. Centering gives $W_0=6c_0$. At every nonzero frequency the constant contribution vanishes, and the remaining expression is linear in $z$, proving the scaling identity. $\square$ For $k\ne0$, $W_{6-k}=-\overline{W_k}$. Keeping the whole coefficients retains phase. Keeping only magnitudes is lossy: $z$ and $-z$, for nonzero centered $z$, give the same coefficient magnitudes, including the common DC magnitude, but are distinct sources. Also the fixed offset $c_0$ is part of this chart's convention; it is not a claim that every variable-angle geometric realization has the same physical baseline offset. # Compatibility of geometric and affine contraction ## Theorem D1: the affine and geometric descriptions commute Fix a homothety center $b$ in a Euclidean space and $r>0$. Define \[ H_{b,r}(P)=b+r(P-b). \] Every point of a marked OMEGA kernel transforms by this map into the kernel with center $H_{b,r}(o)$, scale $ra$, and unchanged frame and opening. For any linear functional $\ell$, define $\chi(P)=1+\ell(P-b)$. Then \[ \chi(H_{b,r}(P))=1+r(\chi(P)-1). \tag{10} \] **Proof.** Apply $H_{b,r}$ to (1) and distribute the scalar: the result is $H_{b,r}(o)+ra(cu+x_iv)$. The same calculation applies to every arc and foot point. Linearity of $\ell$ proves (10). Since $r>0$, $H_{b,r}^{-1}=H_{b,1/r}$. $\square$ For $\lambda=\varphi/(\varphi+\pi)$, positivity gives $0<\lambda<1$. Taking $r=\lambda$ and iterating (10) proves \[ \Omega^n(v)=1+\lambda^n(v-1),\qquad \Xi^n(w)=1+\lambda^{-n}(w-1), \] and exact inversion for every finite integer $n\geq0$. The fixed point is uniquely $1$, since $v=1+\lambda(v-1)$ implies $(1-\lambda)(v-1)=0$. A single $\chi$ is generally many-to-one on a plane or a higher-dimensional space. The commuting equation alone does not make it a complete state representation. A full set of independent coordinate functionals supplies an injective chart for the corresponding finite-dimensional space. The selected geometric readouts satisfy $G(a)=a\varphi$, $C(a)=a\pi$, and \[ G(a)/(G(a)+C(a))=\lambda. \] This proves scale invariance of the selected ratio and the compatibility of the chosen $\lambda$-homothety. Every positive homothety factor preserves the shape family. No uniqueness of an evolution rule follows from shape preservation alone. The ratio also omits the scale and opening, so it is not a complete reconstruction record for the family. # Time-dependent observations ## Theorem O1: a history of observations recovers a state exactly when it separates states For state evolution $T$ and observation $h$, the finite observation history is \[ \mathcal O_k(x)=(h(x),h(Tx),\ldots,h(T^{k-1}x)). \] It has an exact inverse on its image if and only if it is injective, by R1. In the linear case $T(x)=Ax$, $h(x)=Cx$, the criterion is full column rank of \[ \mathbf O_k=\begin{bmatrix}C\\CA\\\vdots\\CA^{k-1}\end{bmatrix}. \] **Proof.** The observation history is multiplication by $\mathbf O_k$. It is injective exactly when its kernel is zero, equivalently when it has full column rank. In that case $(\mathbf O_k^T\mathbf O_k)^{-1}\mathbf O_k^T$ is a left inverse. If the rank is deficient, adding a nonzero kernel vector changes the state without changing any recorded observation. $\square$ For a homothety $A=\lambda I$, the rows of all $CA^j=\lambda^jC$ stay in the row space of $C$. Repeating one scalar observation, even for arbitrarily many times, cannot recover an unrestricted state of dimension greater than one. A finite horizon of 23 does not change that rank. In contrast, for $A=\operatorname{diag}(\lambda,\mu)$ and $C=(1,1)$ with $\mu\ne\lambda$, two observations $y_0=x_1+x_2$ and $y_1=\lambda x_1+\mu x_2$ recover \[ x_1=\frac{\mu y_0-y_1}{\mu-\lambda},\qquad x_2=\frac{y_1-\lambda y_0}{\mu-\lambda}. \] This is a specified two-mode observation model. Its additional dynamics and observation contract supply the extra information; it is not implied by a single scalar contraction. # Witnesses make information loss explicit ## Theorem W1: the exact witness-completion criterion Let $f:X\to Y$ and let $h:X\to W$ be a retained witness. The map $x\mapsto(f(x),h(x))$ is injective if and only if $h$ is injective on each fiber of $f$. **Proof.** A collision of the pair consists exactly of two source points in the same $f$ fiber with the same witness. Excluding all such collisions is precisely fiberwise injectivity. $\square$ For finite fibers of maximum size $M$, a fixed-length binary witness needs at least $\lceil\log_2M\rceil$ bits, when the receiver already knows $f(x)$. There are at most $2^k$ different $k$-bit witness values, proving necessity. An index in a declared enumeration of each fiber proves sufficiency. This is a worst-case conditional bound, not a statement about average entropy or the optimal size of a particular serialized record. ## Theorem W2: Euclidean quotient-remainder completion Let $00\}. \] Any two labeled coordinates determine $t,a$ uniquely. Distinct words have Hamming distance at least five, and this bound is attained. **Proof.** Given coordinates $y_i,y_j$, solve \[ a=\frac{y_j-y_i}{x_j-x_i},\qquad t=\frac{x_jy_i-x_iy_j}{x_j-x_i}. \] The difference between two words is evaluation of a nonzero polynomial of degree at most one. It has at most one root among six distinct coordinates. Thus at least five positions differ. To attain five, choose two distinct positive scales and adjust their offsets to agree at one chosen coordinate. They cannot agree at another. $\square$ ## Corollary E2: errors and erasures With $e$ arbitrary coordinate errors and $r$ known erasures, unique correction is guaranteed when \[ \boxed{2e+r\leq4}. \] **Proof.** If two candidate words fit the received data with at most $e$ errors each, they differ in at most $2e$ of the retained positions and at most $r$ erased positions. Their distance would be at most $2e+r\leq4$, contradicting E1. Existence follows from the actual transmitted word under the stated corruption model. A constructive decoder enumerates pairs of retained positions and chooses the unique affine candidate with at most $e$ disagreements; the bound leaves at least two correct retained positions. $\square$ This is a guarantee about exact symbols at labeled positions. It does not automatically correct small rounding perturbations in all six measured coordinates. The centered positive-scale subfamily used in U1 has distance six, since every $x_i$ is nonzero; its narrower prior information permits a stronger $2e+r\leq5$ uniqueness bound. # Binary gates and computation ## Theorem B1: Boolean operations can be transported into the reference carrier Encode Boolean symbols by any two distinct valid OMEGA carrier states. On decoded bits $b,c\in\{0,1\}$, \[ \neg b=1-b,\quad b\wedge c=bc,\quad b\oplus c=b+c-2bc,\quad\operatorname{NAND}(b,c)=1-bc. \] These operations define exact operations on the encoded states by R3. Every finite Boolean truth table has a finite circuit in these operations. **Proof.** The equations can be checked on the four possible input pairs. For the general finite-circuit statement, for each satisfying input assignment form the conjunction of literals that selects exactly that assignment. The disjunction of these conjunctions realizes the truth table. NOT, AND, and OR can in turn be formed from NAND: NOT is NAND with identical inputs, AND is NOT of NAND, and OR is NAND of the two negated inputs. R3 preserves each gate and every circuit composition. $\square$ This proves a representation-level connection to binary gates. It does not establish a transistor implementation, a performance advantage, or a change in the meaning of Boolean logic. ## Theorem B2: reversible completion of a finite computation For a function $f:\{0,1\}^n\to\{0,1\}^m$, define \[ U_f(x,z)=(x,z\oplus f(x)). \] Then $U_f$ is its own inverse. **Proof.** A second application leaves $x$ unchanged and gives $z\oplus f(x)\oplus f(x)=z$. Retaining $x$ is what makes recovery possible even when $f$ itself is many-to-one. $\square$ Finite machine configurations and programs also have finite descriptions. When their encoding and one-step transition are effective, U3 and R3 give an encoded simulation of each step. This is transport of a specified computation. It is not a proof that repeated application of the one-dimensional affine contraction alone implements every program. # Limits that any universal claim must respect ## Theorem N1: finite capacity cannot encode every finite record at once An injective representation of $M$ possible source states needs at least $M$ distinguishable target states. A target with at most $B$ bits has at most $2^B$ states and therefore requires $M\leq2^B$. **Proof.** This is the pigeonhole principle. There are more than $2^B$ finite binary words in total, so a fixed $B$-bit state cannot be injective on the set of all finite words. $\square$ U1 avoids the contradiction by allowing unbounded integer sizes or equivalently unbounded exact precision, not by compressing arbitrarily much information into a fixed finite resource. The set of finite binary descriptions is countable, while the set of infinite binary sequences is uncountable by the diagonal argument: from a proposed list form a sequence differing from its $k$th member at position $k$. Such sequences embed injectively into the reals by $b\mapsto\sum_{j\geq1}2b_j/3^j$: at the first differing position $k$, the leading difference $2/3^k$ exceeds the maximum tail difference $1/3^k$. Thus all arbitrary real states cannot be encoded by finite descriptions. Ideal exact-real mathematical coordinates and finite executable records are different domains. ## Theorem N2: a smooth exact decoder cannot recover more independent dimensions Let $U\subseteq\mathbb R^d$ be open. Suppose $E:U\to\mathbb R^m$ and a decoder defined on a neighborhood of $E(U)$ are differentiable and satisfy $D\circ E=\operatorname{id}_U$. Then $d\leq m$. **Proof.** The chain rule gives $J_D(E(x))J_E(x)=I_d$. The product rank is at most $m$, whereas the identity rank is $d$. Hence $d\leq m$. $\square$ A centered fixed-angle six-mark family has one real scale parameter; the variable-angle centered family has two. A smooth exact encoding into those parameter families cannot recover an unrestricted open set of more independent real degrees of freedom without extra coordinates. The discrete finite-record encoding of U1 does not satisfy the open-domain hypotheses and is not contradicted by this theorem. ## Theorem N3: faithful geometric encoding does not make every question decidable There is no total computable procedure that decides whether every encoded program halts on every encoded input. **Proof.** Suppose $H(P,x)$ were such a decision procedure. Construct a program $T$ that, on program description $z$, loops forever if $H(z,z)$ predicts halting and halts if it predicts nonhalting. Let $t$ be the finite description of $T$. If $H(t,t)$ predicts halting, $T(t)$ loops; if it predicts nonhalting, $T(t)$ halts. Both possibilities contradict correctness. If a total geometric decoder-decider could solve these questions for an effective lossless OMEGA encoding, composing the ordinary encoder with that decider would supply the impossible $H$. $\square$ An exact universal representation preserves computable questions and undecidable questions alike. It changes their description, not this decision boundary. The diagonal proof is a modern formulation of the classical limitation associated with Turing's work; it is not a quotation from that paper. ## Theorem N4: syntax alone does not choose an external interpretation An encoded formal record can be held fixed while its external denotation is changed. Therefore the record alone does not select that denotation without an interpretation rule or grounding data. **Proof.** Take two external referents $A,B$ and one formal name $q$. One interpretation assigns $q$ to $A$, another assigns it to $B$. The same bitstring, geometric carrier, and arithmetic transformations can represent the name in both interpretations. Any function of those unchanged formal data has the same value in both cases and cannot return two different intended referents. Additional interpretation data distinguish the cases. $\square$ R4 establishes preservation of a specified formal interpretation. It does not infer an unspecified natural-language interpretation or prove a historical claim that mathematics preceded every other language. # Exact constants and numerical residuals The affine identities are algebraic identities in a nonzero scalar. Their floating-point implementations are numerical experiments. Increasing a working precision leaves it finite. For exact inverse evaluation and a forward-output perturbation $e$, \[ \Xi^n(\Omega^n(v)+e)-v=e/\lambda^n. \tag{12} \] The implemented inverse can add further rounding error. Formula (12) proves an amplification factor, not monotonic growth of the measured error at each depth. It contains no exceptional transition at 23. A framework may choose that horizon; its primality supplies no additional floating-point guarantee. The rational floor implementation uses $q=Q/S$, a distinct number from $\lambda$. The checker encloses the constants with exact rational arithmetic and certifies \[ \frac QS<\lambda<\frac{Q+1}{S}. \] Here is the mathematical basis of that certificate. With $M=10^{100}$ and $j=\lfloor\sqrt{5M^2}\rfloor$, integer squaring gives $j/M\leq\sqrt5<(j+1)/M$. The arctangent series at $01/5$. Hence $4\theta-\eta=\pi/4$. Combining lower and upper arctangent sums gives rational bounds for $\pi$. Since $\phi/(\phi+p)$ increases with positive $\phi$ and decreases with positive $p$, the corresponding rational bounds enclose $\lambda$. The code checks the displayed rational inequality by integer arithmetic. It does not claim a first floor-disagreement threshold or certify an entire real orbit from these constant bounds alone. # What has been established The strongest combined statement proved here is: **A fixed known OMEGA reference determines exact planar coordinates from an explicit separating set of measurements. Its scaled marked family also admits an injective representation of every finite formal record. Lossless representations transport all supplied relations, operations, and first-order truth on their stated image domains. The site's centered-logarithmic and full Fourier descriptions give concrete mathematical translations, and quotient-remainder witnesses complete the rational floor dynamics into an exactly recoverable record. These properties compose when their data, ordering, and domain contracts are retained.** The proofs also establish the boundaries: planar distance probes have perpendicular symmetries; finite capacity limits distinguishable records; smooth recovery needs enough independent coordinates; computational undecidability survives recoding; and external interpretation requires its own grounding. Those boundaries specify where additional information is mathematically necessary. This supports OMEGA as an explicit mathematical reference language for the covered domains. The universality of finite formal encoding is shared by other adequate coding systems. No proof here assigns OMEGA a unique or universally optimal status among reference geometries. # Sources and relationship to the existing archive The live pages below were read in the browser during this conversation. They supply the reference definitions and chart conventions. The new general arguments and counterexamples are written out in this note; a page's status label or a previous checker log is not substituted for those arguments. - [OMEGA marked reference geometry](https://omega-naught.grok.me/geometry): construction data, master point formula, Gram data, and explicit transfer scope. - [TGA 2.1](https://omega-naught.grok.me/tga): typed relations, preservation contracts, and separate support labels. - [TGA 2.1 derivations](https://omega-naught.grok.me/tga/2.1/PROOFS.html): centered log-ratio inversion, normalized powering, marked Fourier scaling, and six-mark gap membership. The known tolerance-name erratum concerns the numerical policy; this note's chart proofs are exact-domain arguments. - [Relative representation and recoverable information](https://omega-naught.grok.me/geometry/relative-representation): existing compositional-reconstruction thesis and its declared checker scope. That page was read; this work is not represented as a line-by-line audit of the entire thesis PDF or its unseen companions. - C. E. Shannon, [A Mathematical Theory of Communication](https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf), 1948, corrected reprint. Relevant background for reproduction of messages and finite-state information capacity. The elementary capacity bound used here is proved directly in N1. - A. M. Turing, [On Computable Numbers, with an Application to the Entscheidungsproblem](https://www.cs.virginia.edu/~robins/Turing_Paper_1936.pdf), 1936/1937. Relevant foundational context for computability. N3 supplies its own diagonal argument and does not depend on an unexamined assertion about a program run. # Executed verification The final checker was executed on 24 October 2026 with Python 3.12.14, using the standard library only. Both the normal and optimized commands exited zero. Their stdout files are byte-identical, and both stderr files are empty. All 19 declared groups passed. The exact polynomial checks use a sparse polynomial ring over the field defined by $\varphi^2=\varphi+1$. The reference sine is retained as a formal parameter; identities proved in that ring hold under substitution of the canonical sine. The planar-location and error-correction examples instead use the explicitly auxiliary exact family member $s=3/5$, $c=4/5$. They are not reported as an evaluation of the canonical sine. The finite-message carrier retains formal $s$ and therefore does not substitute the auxiliary seed. The 120-digit Decimal and binary64 DFT group is a numerical diagnostic. The other groups use exact rational, integer, golden-field, or polynomial operations. The general reconstruction, logical-induction, generator, all-length encoding, all-horizon, and impossibility statements are established by the written proofs, not by adding up finite checks. No Lean, Rust, Julia, or proof-assistant kernel execution was performed for this development. ```text OMEGA REFERENCE CHECKER 1.0 Python 3.12.14; standard library only PASS check_field: 49 exact field elements; 2401 operation pairs PASS check_symbolic_geometry: formal s retained; 15 two-mark identities; planar and 3D inverse identities PASS check_planar_measurements: 363 exact planar locations from three squared distances PASS check_reflection_obstruction: same planar distance profiles; mutual squared distances 1 and 9 PASS check_metric_profiles: 64 full-profile isometries; 64 finite-net bounds; one nonresolving collision PASS check_joint_gram_data: 144 exact perpendicular Gram entries and common orthogonal transforms PASS check_graph_distances: 1024 simple graphs; exact positive LDL factors; 10240 distance predicates PASS check_temporal_observability: 441 two-mode state recoveries; uniform-contraction observation collision PASS check_finite_payloads: 326 payload pipelines; 511 bitstrings; longest ledger 1316 PASS check_relations_and_logic: 512 labeled directed graphs; all 9 relation entries; 2048 formula comparisons PASS check_affine_correction: 168 admissible error/erasure patterns; one beyond-bound ambiguity PASS check_ledgers: 32768 seed reversals and first-residue inversions; extinction <= 10 PASS check_fibers: 1000 quotient fibers; sizes 2 or 3; two-bit conditional index sufficient PASS check_telescoping: 23 free-state polynomial telescoping identities; general induction in note PASS check_boolean_lift: all Boolean input pairs and eight reversible lifted states PASS check_normalized_powers: 32 exact positive-rational power-law instances; real-parameter proof in note PASS check_dimension_obstruction: rank-one Jacobian product has determinant zero; cannot equal the 2D identity PASS check_constants: rational sqrt/atan enclosures certify Q/S < lambda < (Q+1)/S PASS check_numerical_charts: 120-digit Decimal charts and binary64 DFT samples; numerical evidence only TOTAL 19 GROUPS PASS SOURCE SHA256 deb36432d39ec4ee12319364e5c24260968239aef3cbab27b6fa4c8c9b8fda55 ``` Checker source SHA-256: ```text deb36432d39ec4ee12319364e5c24260968239aef3cbab27b6fa4c8c9b8fda55 ``` Normal and optimized stdout SHA-256: ```text 5f74147fc4cd530a38f50b1f9cc79df5b4b250788ee42812c860409b10cee6be ``` # Reproduce the checker The companion file `omega_reference_verify.py` is included in the downloadable source package. It requires Python 3.11 or later and only the standard library. Run: ```bash python3 omega_reference_verify.py python3 -O omega_reference_verify.py ``` The two runs are expected to produce the same transcript on the declared interpreter version. Runtime version text can differ on another interpreter. The source digest identifies the exact checker bytes. It is not a mathematical correctness certificate.