Universal Generative Deformation Calculus

Download original file

This is the original TeX manuscript source. Use the paper’s PDF for typeset equations.

Included manuscript sections

• analytic_selected.tex
• example_summary.tex
• result_macros.tex
• walk_growth.tex
• worked_realizations.tex

Read the typeset paper

1 \pdfinfoomitdate=1
2 \pdftrailerid{}
3 \pdfsuppressptexinfo=15
4 \documentclass[11pt,openany]{report}
5 \usepackage[letterpaper,margin=0.82in,headheight=14pt]{geometry}
6 \usepackage[T1]{fontenc}
7 \usepackage[utf8]{inputenc}
8 \usepackage{lmodern}
9 \usepackage{microtype}
10 \usepackage{amsmath,amssymb,amsthm,mathtools}
11 \usepackage{booktabs,longtable,tabularx,array}
12 \usepackage{graphicx,float,placeins}
13 \usepackage{xcolor}
14 \usepackage{enumitem}
15 \usepackage{fancyhdr}
16 \usepackage{listings}
17 \usepackage[hidelinks]{hyperref}
18 \usepackage[nameinlink,noabbrev]{cleveref}
19 \usepackage{needspace}
20 \allowdisplaybreaks
21 
22 \graphicspath{{figures/}}
23 \input{tables/result_macros.tex}
24 
25 \hypersetup{
26   pdftitle={Universal Generative Deformation Calculus: Visual Atlas and Worked Examples},
27   pdfauthor={Komorebi P.},
28   pdfsubject={Infinite perturbation forests, path quotients, branch completion, and family discovery}
29 }
30 
31 \pagestyle{fancy}
32 \fancyhf{}
33 \fancyhead[L]{\small UGDC: Visual Atlas and Worked Examples}
34 \fancyhead[R]{\small Komorebi P.}
35 \fancyfoot[C]{\thepage}
36 \renewcommand{\headrulewidth}{0.4pt}
37 
38 \setlength{\parindent}{0pt}
39 \setlength{\parskip}{0.55em}
40 \raggedbottom
41 
42 \newtheorem{theorem}{Theorem}[chapter]
43 \newtheorem{proposition}[theorem]{Proposition}
44 \newtheorem{lemma}[theorem]{Lemma}
45 \newtheorem{corollary}[theorem]{Corollary}
46 \newtheorem{conjecture}[theorem]{Conjecture}
47 \theoremstyle{definition}
48 \newtheorem{definition}[theorem]{Definition}
49 \newtheorem{example}[theorem]{Worked example}
50 \newtheorem{algorithm}[theorem]{Algorithm}
51 \theoremstyle{remark}
52 \newtheorem{remark}[theorem]{Remark}
53 \newtheorem{boundary}[theorem]{Boundary statement}
54 
55 \newcolumntype{L}[1]{>{\raggedright\arraybackslash}p{#1}}
56 \newcommand{\C}{\mathbb C}
57 \newcommand{\R}{\mathbb R}
58 \newcommand{\N}{\mathbb N}
59 \newcommand{\Z}{\mathbb Z}
60 \newcommand{\Tcal}{\mathcal T}
61 \newcommand{\Dcal}{\mathcal D}
62 \newcommand{\Ocal}{\mathcal O}
63 \newcommand{\Fcal}{\mathcal F}
64 \newcommand{\Pcal}{\mathcal P}
65 \newcommand{\Gcal}{\mathcal G}
66 \newcommand{\Xcal}{\mathcal X}
67 \newcommand{\Ecal}{\mathcal E}
68 \newcommand{\Path}{\operatorname{Path}}
69 \newcommand{\End}{\operatorname{End}}
70 \newcommand{\cost}{\operatorname{cost}}
71 \newcommand{\depth}{\operatorname{depth}}
72 \newcommand{\ev}{\operatorname{ev}}
73 \newcommand{\Reach}{\operatorname{Reach}}
74 \newcommand{\Inv}{\operatorname{Inv}}
75 \newcommand{\cl}{\operatorname{cl}}
76 \newcommand{\id}{\mathrm{id}}
77 \newcommand{\dd}{\,\mathrm d}
78 \newcommand{\norm}[1]{\left\lVert #1\right\rVert}
79 \newcommand{\abs}[1]{\left\lvert #1\right\rvert}
80 
81 \lstset{
82   basicstyle=\ttfamily\footnotesize,
83   breaklines=true,
84   frame=single,
85   showstringspaces=false,
86   columns=fullflexible,
87   keepspaces=true
88 }
89 
90 \title{\textbf{Universal Generative Deformation Calculus}\\[0.42em]
91 \large Visual Atlas and Worked Examples Edition\\[0.30em]
92 \normalsize Infinite Perturbation Forests, Path Quotients, Branch Completion, and Certified Family Discovery}
93 \author{Komorebi P.}
94 \date{August 2026\\Version 1.1.0}
95 
96 \begin{document}
97 \maketitle
98 
99 \begin{abstract}
100 A deformable object does not produce one neighboring object. It produces a direction space, and every resulting object carries a new direction space of its own. Repeating this rule creates an infinite tree of perturbations of perturbations. This paper makes that construction precise. The raw object is a typed path forest whose vertices are finite admissible routes. Relations identify routes with equivalent endpoints, giving a quotient deformation graph. A declared topology then adds limits of convergent infinite branches. Continuous direction spaces are handled by certified finite covers with propagated error bounds. Multi-input deformations replace the ordinary tree by an operadic hyperforest. Projectors on route cylinders give exact refinement and transfer identities. Growth entropy and weighted path zeta functions separate formal path complexity from genuinely distinct object complexity. A family certificate distinguishes repeated rediscovery from a parameterized family with proved closure laws. This visual-atlas edition adds twenty Python-generated figure pairs and eleven finite realizations. The examples cover integer routes, polynomial and analytic operators, sequence transforms, Gaussian probability laws, matrix operations, affine geometry, unlabeled graphs, projective qubits, congruence lifting, branch completion, and associative multi-input products. Every example records raw routes, quotient objects, failed operations, invariants, and reproducible tests.
101 \end{abstract}
102 
103 \begin{center}
104 \fbox{\begin{minipage}{0.92\textwidth}
105 \textbf{Status.} The finite and analytic statements are proved under the hypotheses printed below. The software verifies the declared finite examples through \BuiltInChecks\ built-in checks and \TestCount\ independent tests. Every displayed figure is generated by the accompanying Python source. The phrase ``every direction'' always means every direction in a fixed admissible grammar. It does not mean every imaginable target-dependent map. Historical priority and external novelty are separate questions and are not asserted by this manuscript.
106 \end{minipage}}
107 \end{center}
108 
109 \tableofcontents
110 \listoffigures
111 \listoftables
112 \clearpage
113 
114 \chapter{The object being constructed}
115 
116 \section{The central idea}
117 
118 Fix an object $x_0$. At $x_0$ there is a collection of admissible abstract directions. Applying one direction gives a child object. The child is not forced to inherit the same direction list. It is queried again, and its own admissible directions create the next generation. This process continues without a preset depth.
119 
120 The word direction is intentionally broad. It may mean a tangent vector, a parameter change, a differential operator, a rewrite, a projection, an inversion, a quotient, a coordinate transport, a physical interaction, a refinement, or a functorial move. What matters is that the direction belongs to a grammar declared before the target object is known.
121 
122 There are three mathematical layers.
123 
124 \begin{enumerate}[label=(\arabic*)]
125 \item The \emph{raw path forest} remembers every finite sequence of choices.
126 \item The \emph{quotient object graph} identifies paths that represent the same object or the same object up to a declared symmetry.
127 \item The \emph{completed boundary} adds limits represented by infinite compatible routes.
128 \end{enumerate}
129 
130 Confusing these layers causes two opposite errors. Counting paths as objects exaggerates discovery because many routes may collapse to one endpoint. Keeping only endpoints loses construction history, multiplicity, and interference between routes.
131 
132 \begin{figure}[H]
133 \centering
134 \includegraphics[width=0.98\textwidth]{00_universal_overview.pdf}
135 \caption{The complete object retains five distinct layers: a seed and its local direction fiber, the raw route forest, the relation quotient, the completed boundary, and certified families inside the completed universe.}
136 \label{fig:three-layers}
137 \end{figure}
138 
139 \section{Why a fixed direction grammar is necessary}
140 
141 Given any two objects $x$ and $y$, one can invent a one-step map that sends $x$ to $y$. Such a map does not reveal a generative structure. It hides the target inside the operation.
142 
143 A meaningful construction therefore fixes the allowed direction rule before exploration begins. The grammar may be large, parameterized, or typed. It may even be countably infinite. It must still be independent of the particular endpoint one later hopes to find.
144 
145 \begin{definition}[Anti-encoding condition]
146 A deformation grammar satisfies the anti-encoding condition when the following data are declared independently of the target endpoint:
147 \begin{enumerate}[label=(\alph*)]
148 \item the object types
149 \item the admissible direction fibers
150 \item the evaluation rules
151 \item the path relations and symmetries
152 \item the topology or completion rule
153 \item the cost and validity rules used by the explorer
154 \end{enumerate}
155 \end{definition}
156 
157 The condition does not require every domain to use the same operations. It requires every claim of generation to name the operations that made generation possible.
158 
159 \chapter{Typed deformation data}
160 
161 \section{State-dependent direction fibers}
162 
163 Let $\mathsf T$ be a set of types and let
164 \[
165 \Xcal=\coprod_{t\in\mathsf T}X_t
166 \]
167 be the disjoint union of the corresponding object spaces.
168 
169 \begin{definition}[Direction bundle]
170 A direction bundle over $\Xcal$ is a map
171 \[
172 \pi:\Dcal\longrightarrow\Xcal.
173 \]
174 The fiber
175 \[
176 \Dcal_x=\pi^{-1}(x)
177 \]
178 is the complete set of directions admitted by the declared grammar at $x$.
179 \end{definition}
180 
181 \begin{definition}[Elementary deformation law]
182 An elementary deformation law is a map
183 \[
184 \Phi:\Dcal\longrightarrow\Xcal
185 \]
186 with the understanding that $d\in\Dcal_x$ starts at $x$ and ends at $\Phi(d)$. We also write
187 \[
188 x\star d=\Phi(d).
189 \]
190 \end{definition}
191 
192 The fiber $\Dcal_x$ may depend on every property of $x$. A matrix near a spectral collision may expose different directions from a diagonal matrix. A function with zeros may forbid reciprocal deformation. A singular algebraic point may have a larger tangent cone than a smooth point.
193 
194 \begin{figure}[H]
195 \centering
196 \includegraphics[width=0.98\textwidth]{01_direction_fiber_gallery.pdf}
197 \caption{A geometric direction bundle in action. Each deformed shape is queried again and can expose directions that were not useful or admissible at the root.}
198 \label{fig:direction-fibers}
199 \end{figure}
200 
201 \begin{definition}[Typed generative deformation datum]
202 A typed generative deformation datum is a tuple
203 \[
204 \boxed{
205 \Gcal=
206 (\mathsf T,\Xcal,\pi:\Dcal\to\Xcal,\Phi,\mathcal R,\tau,\kappa,\mathcal V),
207 }
208 \]
209 where $\mathcal R$ is a path relation system, $\tau$ is a completion structure, $\kappa:\Dcal\to[0,\infty)$ is a direction cost, and $\mathcal V$ is a validity predicate.
210 \end{definition}
211 
212 The data separate existence from computation. The full fiber $\Dcal_x$ defines the mathematics. A finite explorer may inspect only a subset, a finite cover, or a best-first frontier.
213 
214 \section{Direction completeness}
215 
216 \begin{definition}[Direction-complete expansion]
217 An expansion rule is direction-complete relative to $\Gcal$ at $x$ when it exposes every element of $\Dcal_x$. It is direction-complete through depth $n$ when this holds at every endpoint of every admissible path of length less than $n$.
218 \end{definition}
219 
220 For a finite fiber, direction completeness is literal enumeration. For a countable fiber, it is fair lazy enumeration. For a continuous fiber, it is represented through parameter families and convergent covers. These cases must not be treated as if they were the same finite loop.
221 
222 \chapter{The raw perturbation forest}
223 
224 \section{Finite paths}
225 
226 Fix a root $x_0\in\Xcal$.
227 
228 \begin{definition}[Admissible path]
229 An admissible path of length $n$ from $x_0$ is a sequence
230 \[
231 \gamma=(d_1,\ldots,d_n)
232 \]
233 for which
234 \[
235 x_k=x_{k-1}\star d_k,
236 \qquad
237 d_k\in\Dcal_{x_{k-1}},
238 \qquad
239 1\le k\le n.
240 \]
241 Its endpoint is $\ev(\gamma)=x_n$. The empty path $\varnothing$ has endpoint $x_0$.
242 \end{definition}
243 
244 Write $\Tcal_n(x_0)$ for the set of length-$n$ paths and
245 \[
246 \Tcal_{\mathrm{fin}}(x_0)=\coprod_{n\ge0}\Tcal_n(x_0).
247 \]
248 Every nonempty path has a unique parent obtained by deleting its final direction. Therefore $\Tcal_{\mathrm{fin}}(x_0)$ is a rooted tree even when different paths have the same endpoint.
249 
250 \begin{theorem}[Exact reachability by depth]
251 For every $n\ge0$,
252 \[
253 \ev(\Tcal_n(x_0))
254 =
255 \Reach_n(x_0),
256 \]
257 where $\Reach_n(x_0)$ is the set obtained by applying exactly $n$ successive admissible directions.
258 \end{theorem}
259 
260 \begin{proof}
261 The statement is immediate at depth zero. Assume it holds at depth $n$. Every path of length $n+1$ is a path of length $n$ followed by one direction from the fiber over its endpoint. Thus its endpoint lies in the $(n+1)$-step reachable set. Conversely, every object in the $(n+1)$-step reachable set is produced by an $n$-step reachable object and one admissible direction. The induction closes.
262 \end{proof}
263 
264 \section{Universal property of the raw tree}
265 
266 The raw tree is free in the precise sense that it imposes no relation beyond admissibility.
267 
268 \begin{theorem}[Free rooted path property]
269 Let $Y$ be a rooted directed graph with root $y_0$. Suppose each admissible pair $(x,d)$ is assigned an edge operation on $Y$ in a way that depends only on the current image vertex and the label $d$. Then there is a unique root-preserving map
270 \[
271 F:\Tcal_{\mathrm{fin}}(x_0)\longrightarrow Y
272 \]
273 that follows those assigned edges along every path.
274 \end{theorem}
275 
276 \begin{proof}
277 Set $F(\varnothing)=y_0$. If $F(\gamma)$ has been defined and $d$ is admissible at $\ev(\gamma)$, define $F(\gamma d)$ by following the assigned $d$-edge from $F(\gamma)$. Every path has one parent, so the definition is unambiguous. Any root-preserving map with the required edge behavior must obey the same recursion, which proves uniqueness.
278 \end{proof}
279 
280 This theorem is simple, but it fixes the object correctly. The tree stores construction histories before any identities are imposed.
281 
282 \section{Costs and path metadata}
283 
284 The cost of a finite path is
285 \[
286 \cost(\gamma)=\sum_{k=1}^n\kappa(d_k).
287 \]
288 A path record may also store error bounds, provenance, validity certificates, symbolic assumptions, and the first place at which an invariant changed. None of these fields alters reachability. They determine how a finite explorer prioritizes and audits the infinite object.
289 
290 \chapter{Relations and the quotient deformation graph}
291 
292 \section{Path congruence}
293 
294 Two different paths may represent the same endpoint. They may also represent endpoints related by a symmetry that the domain regards as immaterial.
295 
296 \begin{definition}[Extension-compatible path relation]
297 An equivalence relation $\sim$ on finite paths is extension-compatible when the following implication holds. If $\gamma\sim\eta$ and a direction label $d$ is admissible after both paths with compatible meaning, then
298 \[
299 \gamma d\sim\eta d.
300 \]
301 The relation may also identify endpoints under a declared symmetry group, rewrite system, canonical form, gauge, or quotient law.
302 \end{definition}
303 
304 \begin{definition}[Quotient deformation graph]
305 The quotient deformation graph is
306 \[
307 \boxed{
308 \Ocal(x_0)=\Tcal_{\mathrm{fin}}(x_0)/\sim.
309 }
310 \]
311 Its vertices are relation classes of paths. An edge is induced by one admissible extension.
312 \end{definition}
313 
314 \begin{theorem}[Well-defined quotient extension]
315 If $\sim$ is extension-compatible, then the edge induced by an admissible direction does not depend on the representative of a path class.
316 \end{theorem}
317 
318 \begin{proof}
319 Choose representatives $\gamma$ and $\eta$ of one class. By extension compatibility, every common admissible extension satisfies $\gamma d\sim\eta d$. Therefore both representatives determine the same child class.
320 \end{proof}
321 
322 The quotient need not remain a tree. Relations can create merges and cycles. The correct object is generally a directed graph, a category, or a groupoid.
323 
324 \section{Canonical keys and exact deduplication}
325 
326 A computational realization uses a canonical key
327 \[
328 K:\Xcal\longrightarrow\mathcal K
329 \]
330 with
331 \[
332 K(x)=K(y)
333 \quad\Longleftrightarrow\quad
334 x\sim y
335 \]
336 for the declared finite example. Exact polynomial coefficients, reduced rational numbers, canonical symbolic expressions, and normalized matrices are suitable examples. Floating point rounding alone is not an exact quotient relation.
337 
338 \begin{boundary}[Canonicalization boundary]
339 A finite program can certify the quotient only to the strength of its canonicalizer. A key collision that is not a mathematical equivalence destroys soundness. A failure to identify equivalent objects preserves soundness but overcounts quotient vertices.
340 \end{boundary}
341 
342 \section{Exact example: the integer walk}
343 
344 Let the root be $0\in\Z$. Every integer has two directions,
345 \[
346 +1,
347 \qquad
348 -1.
349 \]
350 A path of length $n$ is a word in these two labels.
351 
352 \begin{theorem}[Raw and quotient growth of the integer walk]
353 At exact depth $n$, the raw path count is $2^n$. The distinct endpoint count is $n+1$.
354 \end{theorem}
355 
356 \begin{proof}
357 There are two choices at every step, giving $2^n$ words. If a path contains $k$ negative steps, its endpoint is
358 \[
359 (n-k)-k=n-2k.
360 \]
361 As $k$ ranges from $0$ to $n$, these are $n+1$ distinct integers. Every path with the same value of $k$ has the same endpoint.
362 \end{proof}
363 
364 \input{tables/walk_growth.tex}
365 
366 \begin{figure}[H]
367 \centering
368 \includegraphics[width=0.98\textwidth]{02_raw_tree_to_quotient.pdf}
369 \caption{A finite route tree and its quotient graph are shown side by side. Repeated endpoints merge, while route multiplicities and the path archive remain available.}
370 \label{fig:raw-quotient}
371 \end{figure}
372 
373 This example shows why the word tree must be used carefully. The path object is a binary tree. The object quotient is a line with repeated directed traffic.
374 
375 \chapter{Continuous direction spaces}
376 
377 \section{Why literal enumeration is impossible}
378 
379 A tangent sphere, a parameter interval, or a cone of admissible perturbations may be uncountable. No finite or countable loop can visit every real parameter individually. The correct representation keeps the direction family as one mathematical object and explores it through finite covers.
380 
381 Let each fiber $\Dcal_x$ carry a metric $\rho_x$. A finite set $N_\epsilon(x)\subseteq\Dcal_x$ is an $\epsilon$-net when every $d\in\Dcal_x$ lies within $\epsilon$ of some $\widehat d\in N_\epsilon(x)$.
382 
383 \section{Propagated cover theorem}
384 
385 Assume the object space carries a metric $d_X$.
386 
387 \begin{theorem}[Pathwise cover error]
388 Consider an exact route
389 \[
390 x_k=x_{k-1}\star d_k
391 \]
392 and a sampled route
393 \[
394 \widehat x_k=\widehat x_{k-1}\star\widehat d_k.
395 \]
396 Suppose at step $k$,
397 \[
398 d_X(x\star d,y\star e)
399 \le
400 L_k d_X(x,y)+M_k\rho(d,e),
401 \]
402 and
403 \[
404 \rho(d_k,\widehat d_k)\le\epsilon_k.
405 \]
406 Then
407 \[
408 \boxed{
409  d_X(x_n,\widehat x_n)
410  \le
411  \left(\prod_{k=1}^nL_k\right)d_X(x_0,\widehat x_0)
412  +
413  \sum_{j=1}^n
414  M_j\epsilon_j
415  \prod_{k=j+1}^nL_k.
416 }
417 \]
418 \end{theorem}
419 
420 \begin{proof}
421 Set $e_k=d_X(x_k,\widehat x_k)$. The hypothesis gives
422 \[
423 e_k\le L_ke_{k-1}+M_k\epsilon_k.
424 \]
425 Repeated substitution yields the displayed formula.
426 \end{proof}
427 
428 \begin{corollary}[Convergent direction covers]
429 If $L_k\le1$, $M_k\le M$, and
430 \[
431 \sum_{k\ge1}\epsilon_k<\infty,
432 \]
433 then the total endpoint error of the sampled route is bounded uniformly in depth by
434 \[
435 M\sum_{k\ge1}\epsilon_k
436 \]
437 when the roots coincide.
438 \end{corollary}
439 
440 \begin{figure}[H]
441 \centering
442 \includegraphics[width=0.98\textwidth]{03_continuous_direction_covers.pdf}
443 \caption{A continuous direction family is approximated by nested finite covers. The right panel records the certified geometric decrease of the direction-cover radius.}
444 \label{fig:cover-error}
445 \end{figure}
446 
447 \section{Parameterized edges rather than sampled edges}
448 
449 Sampling is an execution method, not the definition of the continuous forest. A continuous family is stored symbolically as an edge bundle
450 \[
451 \Ecal_x
452 =
453 \{(d,x\star d):d\in\Dcal_x\}.
454 \]
455 Finite nets approximate this bundle. Refinement replaces one coarse parameter cell by smaller cells. The cell itself remains a first-class object with a radius and a coverage certificate.
456 
457 \chapter{Infinite branches and completion}
458 
459 \section{The boundary of the raw forest}
460 
461 An infinite route is a sequence
462 \[
463 \omega=(d_1,d_2,\ldots)
464 \]
465 whose every finite prefix is admissible. The set of infinite routes is the path boundary $\partial\Tcal(x_0)$.
466 
467 An infinite route does not automatically determine an object of $\Xcal$. A topology, metric, inverse limit, direct limit, or weak convergence rule is required.
468 
469 \begin{definition}[Convergent route]
470 A route $\omega$ is convergent in a metric realization when its endpoint sequence
471 \[
472 x_0,x_1,x_2,\ldots
473 \]
474 converges in the declared completion $\widehat\Xcal$.
475 \end{definition}
476 
477 \begin{theorem}[Summable-step convergence]
478 Suppose $\widehat\Xcal$ is complete and
479 \[
480 d_X(x_{n+1},x_n)\le a_n
481 \]
482 with
483 \[
484 \sum_{n\ge0}a_n<\infty.
485 \]
486 Then $(x_n)$ is Cauchy and the route has a unique limit in $\widehat\Xcal$.
487 \end{theorem}
488 
489 \begin{proof}
490 For $m>n$,
491 \[
492 d_X(x_m,x_n)
493 \le
494 \sum_{k=n}^{m-1}d_X(x_{k+1},x_k)
495 \le
496 \sum_{k=n}^{m-1}a_k.
497 \]
498 The tail tends to zero. Completeness supplies a unique limit.
499 \end{proof}
500 
501 \section{Dyadic route example}
502 
503 Starting at $0$, let the directions at depth $n$ be
504 \[
505 \pm2^{-(n+1)}.
506 \]
507 Every infinite sign route converges because
508 \[
509 \sum_{n\ge0}2^{-(n+1)}=1.
510 \]
511 The boundary therefore maps into $[-1,1]$. Different sign sequences can represent the same endpoint only at the usual binary-expansion ambiguity, which can be handled by a path relation.
512 
513 \begin{figure}[H]
514 \centering
515 \includegraphics[width=0.98\textwidth]{04_branch_completion.pdf}
516 \caption{A summable perturbation branch converges inside the declared completion. Finite prefixes, tail bounds, and the limiting object remain separate parts of the certificate.}
517 \label{fig:branch-convergence}
518 \end{figure}
519 
520 \section{Inverse and direct limits}
521 
522 Metric completion is not the only possibility. In arithmetic deformation, a branch may be a compatible tower
523 \[
524 x_1\leftarrow x_2\leftarrow x_3\leftarrow\cdots
525 \]
526 whose object is an inverse limit. In nested function spaces, a branch may enter a direct limit. In weak probability or operator topologies, convergence may be defined only against a test class.
527 
528 The calculus therefore treats completion as declared data. It never replaces one mode of convergence with another silently.
529 
530 \chapter{Nonlinear and multi-input perturbations}
531 
532 \section{Why unary routes are not enough}
533 
534 Some directions combine several previously generated objects. Addition, multiplication, tensoring, composition, gluing, and nonlinear interaction are multi-input operations. Their natural combinatorial object is not an ordinary rooted path tree.
535 
536 For every $r\ge2$, let
537 \[
538 \Psi_r:
539 \Xcal^r\times\Dcal^{(r)}
540 \longrightarrow
541 \Xcal
542 \]
543 be a typed $r$-input deformation law.
544 
545 \begin{definition}[Operadic perturbation hyperforest]
546 The operadic perturbation hyperforest generated by a seed family is the set of all finite rooted typed trees whose leaves are seeds and whose internal vertices are labeled by admissible operations $\Phi$ or $\Psi_r$. Evaluation proceeds from leaves to root.
547 \end{definition}
548 
549 \begin{figure}[H]
550 \centering
551 \includegraphics[width=0.98\textwidth]{05_operadic_hyperforest.pdf}
552 \caption{Unary routes and multi-input operations form different generative geometries. A binary product joins two parents, while an operadic hyperforest retains the complete input arity and syntax.}
553 \label{fig:operadic}
554 \end{figure}
555 
556 \section{Free operadic property}
557 
558 \begin{theorem}[Free interaction construction]
559 Before relations are imposed, every finite expression built from the declared seeds and multi-input operations has a unique rooted syntax tree. Any interpretation that assigns compatible concrete operations to the primitive labels extends uniquely to an evaluation of all such trees.
560 \end{theorem}
561 
562 \begin{proof}
563 Every finite expression is parsed recursively into its outermost operation and its input expressions. This gives a rooted tree. Evaluation is defined from the leaves upward. Uniqueness follows because any compatible interpretation must evaluate each internal vertex by the operation written at that vertex.
564 \end{proof}
565 
566 Relations such as associativity, commutativity, distributivity, gauge equivalence, and analytic identities then quotient the free syntax trees.
567 
568 \section{Repeated quadratic substitution}
569 
570 Consider
571 \[
572 u=s+\lambda u^2.
573 \]
574 Repeated substitution expands $u$ into rooted binary trees. If only shape is counted, the number of binary trees with $n$ internal vertices is the Catalan number $C_n$.
575 
576 \begin{proposition}[Catalan interaction growth]
577 The counts satisfy
578 \[
579 C_0=1,
580 \qquad
581 C_{n+1}=\sum_{k=0}^nC_kC_{n-k}.
582 \]
583 \end{proposition}
584 
585 \begin{proof}
586 At the root of a binary tree with $n+1$ internal vertices, the left subtree has $k$ internal vertices and the right subtree has $n-k$. Summing over all $k$ gives the recurrence.
587 \end{proof}
588 
589 The first values are
590 \[
591 1,1,2,5,14,42,132,429,\ldots.
592 \]
593 The rapid growth is not an implementation defect. It is the intrinsic size of the free nonlinear syntax before identities, symmetries, truncation rules, or cost weights are applied.
594 
595 \chapter{Projectors on the perturbation forest}
596 
597 \section{Path Hilbert spaces}
598 
599 For a finite or countable depth layer, define
600 \[
601 \mathcal H_n=\ell^2(\Tcal_n(x_0)).
602 \]
603 The canonical basis vector associated with a path $\gamma$ is written $\lvert\gamma\rangle$.
604 
605 For a prefix $\alpha$ of length $k\le n$, define the cylinder projector
606 \[
607 P^{(n)}_\alpha
608 \lvert\gamma\rangle
609 =
610 \begin{cases}
611 \lvert\gamma\rangle,&\alpha\text{ is a prefix of }\gamma,\\
612 0,&\text{otherwise}.
613 \end{cases}
614 \]
615 
616 \begin{theorem}[Exact cylinder refinement]
617 For every prefix $\alpha$ and every depth $n>\abs{\alpha}$,
618 \[
619 \boxed{
620 P^{(n)}_\alpha
621 =
622 \sum_{d\in\Dcal_{\ev(\alpha)}}P^{(n)}_{\alpha d},
623 }
624 \]
625 with strong convergence when the child set is countable.
626 \end{theorem}
627 
628 \begin{proof}
629 The descendant cylinder of $\alpha$ is the disjoint union of the descendant cylinders of its one-step children. Orthogonal projection onto a disjoint union of basis supports is the sum of the corresponding orthogonal projectors.
630 \end{proof}
631 
632 This identity is the exact packet refinement law for perturbation routes.
633 
634 \section{Mixed-depth cuts}
635 
636 A complete cut is a prefix-free set $\Gamma$ meeting every infinite route exactly once. Its projectors satisfy
637 \[
638 \sum_{\alpha\in\Gamma}P_\alpha=I.
639 \]
640 Thus one branch may be resolved deeply while another stops early. Adaptive exploration can be represented without discarding the unresolved region.
641 
642 \section{Transfer blocks between route packets}
643 
644 Let $A:\mathcal H\to\mathcal K$ be a bounded operator and let $\{P_\alpha\}$ and $\{Q_\beta\}$ be complete route resolutions. Define
645 \[
646 A_{\beta\alpha}=Q_\beta A P_\alpha.
647 \]
648 Then
649 \[
650 A=\sum_{\alpha,\beta}A_{\beta\alpha}.
651 \]
652 If $B$ follows $A$, then
653 \[
654 (BA)_{\gamma\alpha}
655 =
656 \sum_\beta B_{\gamma\beta}A_{\beta\alpha}.
657 \]
658 Every operator composition becomes a sum over compatible intermediate perturbation packets.
659 
660 \chapter{Growth, entropy, and path zeta functions}
661 
662 \section{Two growth sequences}
663 
664 Define
665 \[
666 a_n=\#\Tcal_n(x_0)
667 \]
668 for raw paths and
669 \[
670 q_n=\#\bigl(\ev(\Tcal_n(x_0))/\sim\bigr)
671 \]
672 for distinct quotient endpoints at exact depth $n$.
673 
674 The raw branching entropy and object entropy are
675 \[
676 h_{\mathrm{raw}}
677 =
678 \limsup_{n\to\infty}\frac1n\log a_n,
679 \qquad
680 h_{\mathrm{obj}}
681 =
682 \limsup_{n\to\infty}\frac1n\log q_n.
683 \]
684 Always
685 \[
686 h_{\mathrm{obj}}\le h_{\mathrm{raw}}.
687 \]
688 The gap measures how much formal construction redundancy is removed by identities and symmetries.
689 
690 \section{Weighted partition function}
691 
692 Define
693 \[
694 Z_{x_0}(s)
695 =
696 \sum_{\gamma\ne\varnothing}
697 \exp\bigl(-s\cost(\gamma)\bigr).
698 \]
699 This is a path partition function. It weights short, inexpensive routes more strongly than long, expensive routes.
700 
701 \begin{theorem}[Geometric convergence bound]
702 Suppose every vertex has at most $B$ children and every direction costs at least $c>0$. Then
703 \[
704 Z_{x_0}(s)
705 \le
706 \sum_{n\ge1}B^ne^{-scn}.
707 \]
708 Consequently, $Z_{x_0}(s)$ converges whenever
709 \[
710 sc>\log B.
711 \]
712 In a full regular $B$-ary tree with constant cost $c$, it diverges whenever $sc\le\log B$.
713 \end{theorem}
714 
715 \begin{proof}
716 There are at most $B^n$ paths of length $n$, and each has cost at least $cn$. The upper bound is a geometric series with ratio $Be^{-sc}$. In the regular constant-cost case, equality holds at every depth, giving the divergence statement.
717 \end{proof}
718 
719 \begin{figure}[H]
720 \centering
721 \includegraphics[width=0.98\textwidth]{06_growth_phase_diagram.pdf}
722 \caption{The regular-tree phase diagram separates convergent and divergent weighted path sums. Branching and route cost compete through the threshold $sc=\log B$.}
723 \label{fig:partition}
724 \end{figure}
725 
726 \section{Why the quotient zeta function may be better}
727 
728 A path zeta function measures construction complexity. A quotient zeta function measures distinct object complexity. Depending on the application, one may define
729 \[
730 Z_{\mathrm{obj}}(s)
731 =
732 \sum_{[x]\in\Ocal(x_0)}e^{-sC([x])},
733 \]
734 where $C([x])$ is the least path cost reaching the class. The two functions answer different questions and should be stored together.
735 
736 \chapter{Certified discovery of mathematical families}
737 
738 \section{From isolated endpoints to families}
739 
740 An explorer may generate thousands of endpoints without discovering one coherent family. A family requires more than repeated resemblance.
741 
742 \begin{definition}[Generated family certificate]
743 A generated family certificate is a tuple
744 \[
745 \Fcal=(\Theta,F,\{\sigma_d\}_{d\in A},\mathcal Q,\mathcal I),
746 \]
747 where
748 \begin{enumerate}[label=(\alph*)]
749 \item $\Theta$ is a parameter space
750 \item $F:\Theta\to\Ocal(x_0)$ assigns an object class to every parameter
751 \item $A$ is a collection of closure directions
752 \item $\sigma_d:\Theta\to\Theta$ satisfies
753 \[
754 F(\sigma_d(\theta))=d\cdot F(\theta)
755 \]
756 for every declared closure direction
757 \item $\mathcal Q$ is a set of exact relations or recurrences
758 \item $\mathcal I$ is a set of invariants shared or predictably transformed across the family
759 \end{enumerate}
760 \end{definition}
761 
762 \begin{theorem}[Inductive soundness of a family certificate]
763 Suppose the seed member $F(\theta_0)$ is valid, every closure identity is proved, and every relation in $\mathcal Q$ is preserved by the parameter actions. Then every finite parameter word in the maps $\sigma_d$ produces a valid member of the certified family.
764 \end{theorem}
765 
766 \begin{proof}
767 The root parameter gives a valid member. If a valid member has parameter $\theta$, applying a closure direction $d$ gives $F(\sigma_d(\theta))$ by the closure identity. The preserved relations remain true after the step. Induction over the parameter word proves the claim.
768 \end{proof}
769 
770 \begin{figure}[H]
771 \centering
772 \includegraphics[width=0.99\textwidth]{07_family_certificate_pipeline.pdf}
773 \caption{Generated endpoints become a mathematical family only after a parameterization, closure actions, exact identities, recurrences, and invariants have been verified. Failed closure steps are retained as negative evidence.}
774 \label{fig:family-certificate}
775 \end{figure}
776 
777 \section{Discovery is not historical novelty}
778 
779 A family certificate proves internal structure. It does not prove that the family is absent from the literature. It also does not prove that two differently written certificates define different families. A separate comparison must check equivalence under reparameterization, analytic continuation, known transforms, and classical special cases.
780 
781 \begin{boundary}[Novelty boundary]
782 The software uses the word discovery for a new relation or family relative to the explored archive and declared canonicalizers. It does not use finite computation as a certificate of historical priority.
783 \end{boundary}
784 
785 \section{Recurrence certificates}
786 
787 For a sequence $a_0,a_1,\ldots$, the implementation can search for a constant-coefficient relation
788 \[
789 a_n=c_1a_{n-1}+\cdots+c_ra_{n-r}
790 \]
791 using exact rational linear algebra. A proposed recurrence is accepted only when it reproduces every supplied term exactly. The built-in Fibonacci example recovers
792 \[
793 (c_1,c_2)=(1,1).
794 \]
795 This is a small example of the certificate principle. Fitting alone is not enough. The relation must be verified and then proved or bounded on the intended domain.
796 
797 \chapter{Finite exploration of the infinite object}
798 
799 \section{Breadth-first relation-reduced exploration}
800 
801 The reference explorer stores raw paths and quotient nodes simultaneously.
802 
803 \begin{algorithm}[Audited breadth-first exploration]
804 Given a deformation system, roots, a depth bound, and resource limits:
805 \begin{enumerate}[label=\arabic*.]
806 \item Insert each root as a raw path and a quotient node.
807 \item Remove the oldest path from the frontier.
808 \item Query the direction fiber of its endpoint.
809 \item Apply every exposed direction.
810 \item Validate the child and record failures explicitly.
811 \item Compute the canonical relation key.
812 \item Reuse an existing quotient node or create a new one.
813 \item Record the directed edge and its path multiplicity.
814 \item Insert the new raw path even when its endpoint was already known.
815 \item Repeat until the frontier, depth, or resource limit is exhausted.
816 \end{enumerate}
817 \end{algorithm}
818 
819 Keeping the new path after endpoint deduplication is essential. It preserves the raw tree count and route multiplicity.
820 
821 \begin{figure}[H]
822 \centering
823 \includegraphics[width=0.99\textwidth]{08_exploration_frontier.pdf}
824 \caption{A finite computation is an audited frontier through an infinite forest. Expanded nodes, quotient nodes, deferred parameter cells, and failed moves are retained explicitly.}
825 \label{fig:pipeline}
826 \end{figure}
827 
828 \begin{theorem}[Finite-depth completeness]
829 If every direction fiber reached before depth $N$ is finite and the breadth-first explorer has no active resource cap below the resulting path count, then it records every admissible path of length at most $N$.
830 \end{theorem}
831 
832 \begin{proof}
833 Breadth-first exploration inserts every root. Assume every path of length $n<N$ is eventually removed from the frontier. The algorithm queries its complete finite direction fiber and inserts every one-step extension. Therefore every path of length $n+1$ is inserted. Induction proves the result.
834 \end{proof}
835 
836 \section{Fair enumeration of countable fibers}
837 
838 When a fiber is countably infinite, exhausting one vertex before visiting another may starve entire branches. A fair explorer dovetails local enumerators. For example, at global stage $m$ it may inspect the first $m$ directions of every path among the first $m$ frontier paths.
839 
840 \begin{proposition}[Fairness consequence]
841 If every local countable fiber has a fixed enumeration and the global scheduler is fair, then every finite admissible path is eventually generated.
842 \end{proposition}
843 
844 \begin{proof}
845 A finite path uses finitely many local direction indices. Fairness eventually reaches the required index at the root, then at the selected child, and so on through the finite route.
846 \end{proof}
847 
848 \section{Best-first search and its limitation}
849 
850 Cost-ordered search may reach useful objects much earlier than breadth-first search. It is not complete under an infinite budget unless the scheduling rule prevents starvation. A finite beam search is deliberately incomplete. Its output is a searched prefix plus an unexplored residual, not the entire forest.
851 
852 \chapter{Stability of perturbations of perturbations}
853 
854 \section{Two systems being compared}
855 
856 Let $\Phi_k$ and $\widetilde\Phi_k$ be exact and perturbed deformation laws along corresponding routes. Suppose
857 \[
858 d_X(\Phi_k(x,d),\widetilde\Phi_k(y,e))
859 \le
860 L_kd_X(x,y)+M_k\rho(d,e)+\eta_k,
861 \]
862 where $\eta_k$ measures model error.
863 
864 \begin{theorem}[Recursive deformation stability]
865 If the direction mismatch at step $k$ is at most $\epsilon_k$, then the endpoint error satisfies
866 \[
867 e_n
868 \le
869 \left(\prod_{k=1}^nL_k\right)e_0
870 +
871 \sum_{j=1}^n
872 (M_j\epsilon_j+\eta_j)
873 \prod_{k=j+1}^nL_k.
874 \]
875 \end{theorem}
876 
877 \begin{proof}
878 The one-step inequality gives
879 \[
880 e_k\le L_ke_{k-1}+M_k\epsilon_k+\eta_k.
881 \]
882 Iterate the recurrence.
883 \end{proof}
884 
885 This theorem shows why repeated perturbations are not automatically unstable. Stability depends on the product of state sensitivities and the accumulated local errors.
886 
887 \begin{figure}[H]
888 \centering
889 \includegraphics[width=0.98\textwidth]{18_stability_regimes.pdf}
890 \caption{The same local approximation error behaves differently under contractive, neutral, and expansive deformation laws. Stability is a property of the iterated sensitivities, not of depth alone.}
891 \label{fig:branch-stability}
892 \end{figure}
893 
894 \section{Relation stability}
895 
896 A quotient relation can change discontinuously under noise. Two exact eigenvalues may split. A zero symbolic identity may become a small residual. Therefore approximate quotienting must carry a margin or conditioning certificate.
897 
898 \begin{boundary}[No threshold without a margin]
899 A numerical tolerance does not define a stable equivalence relation by itself. A reliable approximate quotient requires a separation gap, an interval certificate, exact arithmetic, or another proof that the classification is unchanged under the declared perturbation.
900 \end{boundary}
901 
902 \chapter{Worked domain realizations}
903 
904 The calculus is useful only when the abstract layers survive contact with concrete objects. The examples below use different state spaces, direction grammars, quotient relations, and completion rules. They are not presented as interchangeable models. Their common feature is the same audited production sequence:
905 \[
906 \text{seed}\longrightarrow\text{directions}\longrightarrow\text{raw routes}
907 \longrightarrow\text{quotient classes}\longrightarrow\text{certificate or completion}.
908 \]
909 
910 \input{tables/worked_realizations.tex}
911 
912 \section{Polynomial operators and an exact commuting square}
913 
914 Let
915 \[
916 p(x)=x^2+1,
917 \qquad
918 (Tp)(x)=p(x+1),
919 \qquad
920 (Dp)(x)=p'(x).
921 \]
922 The two primitive directions $T$ and $D$ satisfy
923 \[
924 D(Tp)=T(Dp)=2x+2.
925 \]
926 The raw path tree therefore contains two length-two routes,
927 \[
928 p\xrightarrow{T}Tp\xrightarrow{D}DTp,
929 \qquad
930 p\xrightarrow{D}Dp\xrightarrow{T}TDp,
931 \]
932 but the quotient graph contains one endpoint class because $DTp=TDp$ exactly.
933 
934 \begin{figure}[H]
935 \centering
936 \includegraphics[width=0.98\textwidth]{09_polynomial_atlas.pdf}
937 \caption{Polynomial descendants are displayed as actual functions. Distinct operator routes are retained, while exact symbolic identities merge their endpoint classes.}
938 \label{fig:polynomial-square}
939 \end{figure}
940 
941 The complete finite grammar used by the package also contains reflection, multiplication by $x$, and the forward finite difference. Starting from $x^2+1$, the depth-four experiment records \PolynomialRawPaths\ raw paths and \PolynomialQuotientNodes\ exact polynomial classes under coefficient canonicalization.
942 
943 \section{Analytic functions from one hyperbolic seed}
944 
945 Start from
946 \[
947 f_0(x)=\cosh x.
948 \]
949 The declared directions are differentiation, reciprocal formation, logarithmic differentiation, squaring, even projection, odd projection, and a restricted argument-doubling direction. The first three give
950 \[
951 D(\cosh x)=\sinh x,
952 \qquad
953 R(\cosh x)=\operatorname{sech}x,
954 \qquad
955 L(\cosh x)=\tanh x.
956 \]
957 Applying the same grammar again produces $\operatorname{csch}x$, $\coth x$, powers, parity components, and doubled arguments. Reciprocal and logarithmic derivative are rejected at the zero expression and the failed attempts remain in the audit.
958 
959 \begin{figure}[H]
960 \centering
961 \includegraphics[width=0.99\textwidth]{10_analytic_atlas.pdf}
962 \caption{Analytic descendants are shown as a relation-reduced family of function profiles. Poles, parity projections, powers, derivatives, and logarithmic derivatives remain auditable.}
963 \label{fig:analytic-graph}
964 \end{figure}
965 
966 \input{tables/analytic_selected.tex}
967 
968 The recorded depth-three exploration contains \AnalyticRawPaths\ raw paths, \AnalyticQuotientNodes\ quotient objects, and twenty-four explicitly audited invalid attempts.
969 
970 \section{Sequence transforms and a binomial subfamily}
971 
972 The executable atlas starts from the eight-term Fibonacci prefix
973 \[
974 a=(0,1,1,2,3,5,8,13).
975 \]
976 Its fixed grammar contains first difference, partial summation, the binomial transform, sign alternation, and reversal. For a finite sequence $a=(a_0,\ldots,a_{N-1})$, define
977 \[
978 (Da)_0=a_0,
979 \qquad
980 (Da)_n=a_n-a_{n-1},
981 \]
982 \[
983 (Ca)_n=\sum_{j=0}^{n}a_j,
984 \qquad
985 (Ba)_n=\sum_{j=0}^{n}\binom{n}{j}a_j.
986 \]
987 The two triangular operators satisfy
988 \[
989 DC=CD=I
990 \]
991 on every finite sequence under the stated boundary convention. Hence the raw routes
992 \[
993 a\xrightarrow{D}Da\xrightarrow{C}CDa,
994 \qquad
995 a\xrightarrow{C}Ca\xrightarrow{D}DCa
996 \]
997 return to the seed class exactly.
998 
999 A second, closed subfamily starts from the impulse $\delta=(1,0,0,\ldots)$. Repeated summation gives
1000 \[
1001 (C^k\delta)_n=\binom{n+k-1}{k-1}
1002 \qquad(k\ge1),
1003 \]
1004 by induction and the hockey-stick identity. This provides a genuine family certificate: perturbation depth $k$ is the parameter, $C$ advances the parameter, $D$ reverses it away from the boundary, and the displayed coefficient formula is an exact invariant relation.
1005 
1006 \begin{figure}[H]
1007 \centering
1008 \includegraphics[width=0.99\textwidth]{11_sequence_atlas.pdf}
1009 \caption{The finite Fibonacci seed is transformed by difference, summation, binomial weighting, sign alternation, and reversal. A separate impulse route yields the closed binomial subfamily $C^k\delta$.}
1010 \label{fig:sequence-family}
1011 \end{figure}
1012 
1013 The depth-three finite exploration records \SequenceRawPaths\ raw paths, \SequenceQuotientNodes\ exact sequence states, and \SequenceQuotientEdges\ quotient edges. The visual uses the Fibonacci seed because it exposes all five transforms in one compact finite example. The binomial formula supplies the accompanying infinite parameterized example.
1014 
1015 \section{Gaussian probability laws}
1016 
1017 Represent a one-dimensional Gaussian law by the exact parameter pair $(\mu,v)$, where $v=\sigma^2>0$. The grammar contains shifts by $\pm1$, reflection, convolution with $N(0,1)$, centering, scaling by two, and standardization. Their parameter actions are
1018 \[
1019 \begin{aligned}
1020 (\mu,v)&\mapsto(\mu\pm1,v),
1021 & (\mu,v)&\mapsto(-\mu,v),\\
1022 (\mu,v)&\mapsto(\mu,v+1),
1023 & (\mu,v)&\mapsto(0,v),\\
1024 (\mu,v)&\mapsto(2\mu,4v),
1025 & (\mu,v)&\mapsto(0,1).
1026 \end{aligned}
1027 \]
1028 Standardization is state dependent: it is exposed only away from the canonical seed. Reflection is invisible at centered laws, while convolution and scaling change entropy through the variance coordinate.
1029 
1030 One explicit route is
1031 \[
1032 (0,1)\xrightarrow{\text{shift }+1}(1,1)
1033 \xrightarrow{\ast N(0,1)}(1,2)
1034 \xrightarrow{\text{scale }2}(2,8)
1035 \xrightarrow{\text{standardize}}(0,1).
1036 \]
1037 The final edge is a many-to-one canonicalization direction. It produces a strong quotient collapse without erasing the route that created the noncanonical law. The differential entropy
1038 \[
1039 h(N(\mu,v))=\frac12\log(2\pi e v)
1040 \]
1041 is unaffected by shifts and reflection, increases under unit convolution, and changes by $\log 2$ under scaling by two.
1042 
1043 \begin{figure}[H]
1044 \centering
1045 \includegraphics[width=0.99\textwidth]{12_gaussian_atlas.pdf}
1046 \caption{The Gaussian atlas is plotted in mean and standard-deviation coordinates. Color records minimum perturbation depth. A shortest-route forest retains construction history without obscuring the exact parameter geometry.}
1047 \label{fig:gaussian-atlas}
1048 \end{figure}
1049 
1050 The recorded prefix contains \GaussianRawPaths\ raw paths, \GaussianQuotientNodes\ exact parameter pairs, and \GaussianQuotientEdges\ quotient edges.
1051 
1052 \section{Graph deformation modulo isomorphism}
1053 
1054 Begin with the path graph $P_3$ on three vertices. The finite grammar contains graph complement, line-graph formation, leaf attachment, edge subdivision, and a relabeling direction. Raw states are labeled adjacency matrices. For a graph with $n\le6$ vertices, the quotient key is the lexicographically least adjacency bit string over all $n!$ vertex permutations.
1055 
1056 This example distinguishes two kinds of movement. Complement and line-graph formation can create genuinely new graph classes. A relabeling changes the raw adjacency matrix but not the quotient object. Subdivision and leaf attachment also change the number of vertices, so the direction fiber depends on the current graph rather than only on the original seed.
1057 
1058 For the seed $P_3$, the first descendants include
1059 \[
1060 \overline{P_3},
1061 \qquad
1062 L(P_3)\cong P_2,
1063 \qquad
1064 P_3\text{ with one attached leaf},
1065 \qquad
1066 P_4\text{ obtained by subdivision}.
1067 \]
1068 If $\rho$ reverses the vertex labels, then
1069 \[
1070 \operatorname{key}(G)=\operatorname{key}(\rho G\rho^{-1})
1071 \]
1072 exactly, although the two raw paths and raw adjacency matrices remain distinct.
1073 
1074 \begin{figure}[H]
1075 \centering
1076 \includegraphics[width=0.99\textwidth]{16_graph_atlas.pdf}
1077 \caption{Graph directions include complement, line graph, subdivision, leaf addition, and relabeling. Separated cards keep each graph legible, while exact canonicalization merges labeled routes representing the same unlabeled graph.}
1078 \label{fig:graph-quotient}
1079 \end{figure}
1080 
1081 The recorded prefix contains \GraphRawPaths\ raw paths, \GraphQuotientNodes\ quotient graph classes, and \GraphQuotientEdges\ directed quotient edges.
1082 
1083 \section{Finite matrix deformation atlas}
1084 
1085 Let
1086 \[
1087 A_0=\begin{pmatrix}1&1\\0&1\end{pmatrix}.
1088 \]
1089 The declared matrix directions are transpose, negation, squaring, conjugation by the coordinate swap, symmetrization, commutation with
1090 \[
1091 J=\begin{pmatrix}0&-1\\1&0\end{pmatrix},
1092 \]
1093 and adjugation. Every result is stored as an exact integer matrix. The metadata include trace, determinant, rank, symmetry, and squared Frobenius norm.
1094 
1095 The first calculations are explicit:
1096 \[
1097 A_0^{\mathsf T}=\begin{pmatrix}1&0\\1&1\end{pmatrix},
1098 \qquad
1099 A_0^2=\begin{pmatrix}1&2\\0&1\end{pmatrix},
1100 \qquad
1101 \operatorname{adj}(A_0)=\begin{pmatrix}1&-1\\0&1\end{pmatrix}.
1102 \]
1103 They satisfy
1104 \[
1105 (A^{\mathsf T})^{\mathsf T}=A,
1106 \qquad
1107 A\operatorname{adj}(A)=\det(A)I,
1108 \qquad
1109 \det(SAS)=\det(A)
1110 \]
1111 for the swap matrix $S$. Thus involution, adjugate, and conjugacy relations generate quotient identifications, while squaring and commutation usually move to new invariant values.
1112 
1113 \begin{figure}[H]
1114 \centering
1115 \includegraphics[width=0.99\textwidth]{15_matrix_atlas.pdf}
1116 \caption{Each matrix descendant is displayed with exact trace and determinant. Transposition, squaring, adjugation, symmetrization, conjugation, and commutation create an audited operator atlas.}
1117 \label{fig:matrix-quotient}
1118 \end{figure}
1119 
1120 The recorded prefix contains \MatrixRawPaths\ raw paths, \MatrixQuotientNodes\ exact matrix objects, and \MatrixQuotientEdges\ quotient edges.
1121 
1122 \section{A single-qubit gate orbit}
1123 
1124 The implementation works directly after quotienting global phase. A state is one of the six exact Bloch-axis vectors
1125 \[
1126 (0,0,\pm1),
1127 \qquad
1128 (\pm1,0,0),
1129 \qquad
1130 (0,\pm1,0),
1131 \]
1132 representing the Pauli eigenstates $|0\rangle$, $|1\rangle$, $|+\rangle$, $|-\rangle$, $|+i\rangle$, and $|-i\rangle$. The primitive grammar contains $H,S,X,Z$, acting as signed permutations of the Bloch coordinates:
1133 \[
1134 H(x,y,z)=(z,-y,x),
1135 \qquad
1136 S(x,y,z)=(-y,x,z),
1137 \]
1138 \[
1139 X(x,y,z)=(x,-y,-z),
1140 \qquad
1141 Z(x,y,z)=(-x,-y,z).
1142 \]
1143 These rules are exact and avoid floating-point phase canonicalization.
1144 
1145 Starting from $|0\rangle$, one has
1146 \[
1147 |0\rangle\xrightarrow{H}|+\rangle
1148 \xrightarrow{S}|+i\rangle
1149 \xrightarrow{S}|-\rangle,
1150 \]
1151 while $H^2=X^2=Z^2=I$ on the projective orbit and $S^4=I$. The number of raw gate words grows as $4^n$, but the quotient orbit saturates at six objects.
1152 
1153 \begin{figure}[H]
1154 \centering
1155 \includegraphics[width=0.99\textwidth]{13_qubit_atlas.pdf}
1156 \caption{The left panel realizes the quotient states on the Bloch sphere. The right panel shows the relation-reduced gate graph for the displayed $H$ and $S$ generators.}
1157 \label{fig:qubit-orbit}
1158 \end{figure}
1159 
1160 The depth-five prefix contains \QubitRawPaths\ raw gate words, \QubitQuotientNodes\ quotient states, and \QubitQuotientEdges\ quotient edges.
1161 
1162 \section{Affine geometry with descendant-specific directions}
1163 
1164 Take the rational triangle
1165 \[
1166 K_0=\left\{(0,0),(2,0),\left(1,\frac32\right)\right\}.
1167 \]
1168 The grammar contains quarter turns, reflection in the $x$ axis, horizontal shear, scaling by two, and unit translations in the two coordinate directions. Each descendant is again a finite rational point configuration, so the grammar is queried recursively at every child.
1169 
1170 For example,
1171 \[
1172 R(x,y)=(-y,x),
1173 \qquad
1174 H(x,y)=(x+y,y),
1175 \qquad
1176 S_2(x,y)=(2x,2y).
1177 \]
1178 The relation $R^4=I$ closes one route loop exactly. Reflection is an involution. Rotation preserves signed area, reflection reverses its sign, shear preserves it, and scaling by two multiplies it by four. These invariants separate visually similar but mathematically different descendants.
1179 
1180 \begin{figure}[H]
1181 \centering
1182 \includegraphics[width=0.99\textwidth]{14_affine_shape_atlas.pdf}
1183 \caption{Affine descendants are plotted as concrete shapes. Every child remains a parent for further rotations, reflections, shears, scales, and translations.}
1184 \label{fig:shape-atlas}
1185 \end{figure}
1186 
1187 The explored prefix contains \ShapeRawPaths\ raw paths, \ShapeQuotientNodes\ exact shape states, and \ShapeQuotientEdges\ quotient edges.
1188 
1189 \section{Completion of a summable perturbation branch}
1190 
1191 The executable completion example starts from $x_0=0$ and exposes the two depth-dependent directions
1192 \[
1193 d_n^{\pm}=\pm2^{-(n+1)}.
1194 \]
1195 A finite route with signs $\varepsilon_n\in\{-1,+1\}$ reaches
1196 \[
1197 x_N=\sum_{n=0}^{N-1}\varepsilon_n2^{-(n+1)}.
1198 \]
1199 Every infinite sign route is Cauchy because
1200 \[
1201 |x_m-x_n|
1202 \le
1203 \sum_{k=n}^{m-1}2^{-(k+1)}
1204 <2^{-n}.
1205 \]
1206 The completed object therefore contains a limit for every infinite route. The selected periodic sign pattern used in the figure converges numerically to $5/7$.
1207 
1208 The numerical convergence and the certified tail envelope are displayed in \cref{fig:branch-convergence}.
1209 
1210 The same architecture covers iterated function systems. For the middle-third Cantor maps $L(x)=x/3$ and $R(x)=(x+2)/3$, finite words identify cylinder intervals and infinite words receive points only in the completed boundary. The dyadic example is used in the executable release because its exact tail certificate is especially transparent.
1211 
1212 \section{Associative multi-input products}
1213 
1214 With four ordered leaves and one binary operation, the free operadic layer contains the five Catalan parenthesizations
1215 \[
1216 ((ab)c)d,
1217 \quad
1218 (a(bc))d,
1219 \quad
1220 (ab)(cd),
1221 \quad
1222 a((bc)d),
1223 \quad
1224 a(b(cd)).
1225 \]
1226 They are different syntax trees. If the operation is associative, all five evaluate to one quotient object. If it is also commutative, the $4!$ permutations of the leaves are identified as well. The raw tree and the quotient must therefore remain separate even in elementary algebra. The complete syntax-to-quotient visual appears in \cref{fig:operadic}.
1227 
1228 \section{Congruence lifting and singular branch geometry}
1229 
1230 For a prime $p$, a state at level $m$ is a residue class modulo $p^m$ satisfying a declared congruence. Its direction fiber is the finite set of next base-$p$ digits that lift the current solution to level $m+1$.
1231 
1232 For $x^2=1$ over powers of $3$, the two simple roots lift uniquely:
1233 \[
1234 1\pmod 3\longleftarrow1\pmod 9\longleftarrow1\pmod{27}\longleftarrow\cdots,
1235 \]
1236 \[
1237 -1\pmod 3\longleftarrow-1\pmod 9\longleftarrow-1\pmod{27}\longleftarrow\cdots.
1238 \]
1239 For the singular equation $x^2=0$ over powers of $2$, some nodes have several valid next digits and some branches later die. Thus branching number is an object property detected by the local congruence geometry.
1240 
1241 \begin{figure}[H]
1242 \centering
1243 \includegraphics[width=0.99\textwidth]{17_adic_lift_forest.pdf}
1244 \caption{Simple roots produce unique lift streams, whereas singular roots branch and terminate. The direction fiber at each node is the exact set of valid next digits.}
1245 \label{fig:adic-lifts}
1246 \end{figure}
1247 
1248 The two recorded prefixes contain \AdicSimpleRawPaths\ simple-lift routes and \AdicSingularRawPaths\ singular-lift routes.
1249 
1250 \section{Cross-domain route compression}
1251 
1252 The eleven realizations use different grammars and quotient relations. Their raw-path counts and quotient-object counts should therefore not be compared as one performance benchmark. The comparison is structural: it shows how strongly relations can compress the same finite-depth exploration.
1253 
1254 \begin{figure}[H]
1255 \centering
1256 \includegraphics[width=0.99\textwidth]{19_cross_domain_counts.pdf}
1257 \caption{Raw route growth and quotient-object growth can differ by orders of magnitude. Strong collapse occurs for the integer walk and projective qubit orbit, while the finite sequence and shape grammars retain more distinct descendants.}
1258 \label{fig:cross-domain-counts}
1259 \end{figure}
1260 
1261 \input{tables/example_summary.tex}
1262 
1263 \section{What the examples establish}
1264 
1265 These examples cover several distinct reasons for quotienting, certification, and completion:
1266 \begin{itemize}
1267 \item equal endpoints reached by different paths
1268 \item operator commutation
1269 \item symbolic analytic identities
1270 \item inverse transformation laws
1271 \item graph isomorphism
1272 \item conjugacy invariance
1273 \item global quantum phase
1274 \item associativity of multi-input operations
1275 \end{itemize}
1276 They also show two distinct uses of completion: summable perturbation streams and iterated contraction systems. None of the examples proves that a useful quotient or completion exists in every domain. Each one demonstrates how the required domain theorem enters the common calculus.
1277 
1278 \chapter{Implementation and verification}
1279 
1280 \section{Package architecture}
1281 
1282 The package \texttt{komorebi-generative-deformation} is organized around a small interface.
1283 
1284 \begin{longtable}{L{0.28\textwidth}L{0.62\textwidth}}
1285 \caption{Core software objects.}\label{tab:api}\\
1286 \toprule
1287 Object & Role \\
1288 \midrule
1289 \endfirsthead
1290 \toprule
1291 Object & Role \\
1292 \midrule
1293 \endhead
1294 \texttt{DeformationSystem} & Supplies state-dependent directions, one-step application, canonical keys, invariants, validation, serialization, and optional distance.\\
1295 \texttt{Direction} & Stores a primitive direction label, parameter, cost, family name, resolution, and metadata.\\
1296 \texttt{PerturbationForest} & Stores raw paths, relation-reduced nodes, directed edges, multiplicities, invalid attempts, and export methods.\\
1297 \texttt{ExplorationConfig} & Declares depth, path, node, edge, strategy, and per-path direction limits.\\
1298 \texttt{IntervalDirectionFamily} & Represents a continuous interval of directions through dyadic certified nets.\\
1299 \texttt{InteractionRule} & Declares a multi-input deformation for the operadic hyperforest.\\
1300 \texttt{FamilyCertificate} & Stores parameter assignments, closure directions, relations, invariants, and verification failures.\\
1301 \texttt{metrics} & Implements entropy proxies, growth series, partition functions, cover error, and tail bounds.\\
1302 \bottomrule
1303 \end{longtable}
1304 
1305 \section{Reference experiments}
1306 
1307 The complete experiment driver rebuilds every forest, every table, and all \FigureCount\ vector-PDF and PNG figure pairs from source. The eleven relation-reduced finite prefixes are summarized in \cref{tab:example-summary}. They should not be read as a speed ranking because their grammars, depth limits, state spaces, and quotient relations are different.
1308 
1309 The \BuiltInChecks\ built-in checks verify:
1310 \begin{itemize}
1311 \item exact raw and quotient growth for the integer walk
1312 \item the dyadic completion tail bound
1313 \item monotone decay of the continuous-direction cover error
1314 \item Catalan and Fibonacci identities
1315 \item generation of $\tanh x$ and $\operatorname{sech}x$
1316 \item inverse sequence transforms
1317 \item Gaussian standardization
1318 \item six-state projective qubit closure
1319 \item purity preservation
1320 \item matrix and affine involutions
1321 \item graph relabeling invariance
1322 \item unique simple-root lifting and singular-root branching
1323 \end{itemize}
1324 All \BuiltInChecks\ checks pass in the recorded run. The result file stores every Boolean check separately rather than recording only a final aggregate.
1325 
1326 \section{Independent test suite}
1327 
1328 The independent suite contains \TestCount\ tests. The tests are grouped by mathematical contract rather than by figure:
1329 
1330 \begin{longtable}{L{0.25\textwidth}L{0.64\textwidth}}
1331 \caption{Independent verification groups.}\label{tab:test-groups}\\
1332 \toprule
1333 Group & Representative assertions \\
1334 \midrule
1335 \endfirsthead
1336 \toprule
1337 Group & Representative assertions \\
1338 \midrule
1339 \endhead
1340 Forest accounting & Raw binary path counts, quotient counts, route multiplicity, shortest paths, invalid-attempt audit, and export round trips.\\
1341 Completion and covers & Dyadic Cauchy behavior, interval-net coverage, geometric tail estimates, and propagated finite-cover error.\\
1342 Family certificates & Catalan and Fibonacci values, exact recurrence inference, rejection of false recurrences, invariant clustering, and closure checks.\\
1343 Multi-input structure & Binary addition and multiplication, ternary interaction limits, and preservation of syntax-tree distinctions before quotienting.\\
1344 Symbolic examples & Polynomial route collapse, analytic generation of $\tanh$ and $\operatorname{sech}$, and explicit rejection of invalid reciprocal operations.\\
1345 Extended domains & Sequence inverses, Gaussian parameter actions, six-state qubit closure and purity, matrix adjugation and involutions, affine symmetry, graph isomorphism, and simple versus singular congruence lifting.\\
1346 \bottomrule
1347 \end{longtable}
1348 
1349 The recorded run passes all \TestCount\ tests. The tests do not infer unbounded correctness from finite execution. They verify the exact finite contracts exported by the release and support the manuscript's concrete examples.
1350 
1351 \section{Machine-readable audit trail}
1352 
1353 Every explored example exports:
1354 
1355 \begin{itemize}
1356 \item \texttt{forest.json} with nodes, edges, summaries, paths, and failures
1357 \item \texttt{nodes.csv} with states, invariants, costs, and shortest routes
1358 \item \texttt{edges.csv} with directions, depth, cost, and multiplicity
1359 \item generated figures in vector PDF and PNG formats
1360 \item exact test and experiment summaries
1361 \end{itemize}
1362 
1363 The invalid-attempt log is part of the mathematical record. A missing reciprocal, failed constraint, or complexity cap is not treated as an absent direction without explanation.
1364 
1365 \chapter{What the calculus does and does not claim}
1366 
1367 \section{What has been completed}
1368 
1369 Version 1.1.0 contains \FigureCount\ Python-generated figure pairs and \ExampleCount\ worked domain realizations. The final generative object consists of the following linked structures.
1370 
1371 \begin{enumerate}[label=(\arabic*)]
1372 \item A typed object space with one direction fiber at every object.
1373 \item A raw rooted forest containing every finite admissible route.
1374 \item A relation quotient that separates path multiplicity from distinct object classes.
1375 \item A completion rule for infinite compatible routes.
1376 \item A parameter-cover theorem for continuous direction families.
1377 \item An operadic extension for multi-input interactions.
1378 \item Exact path projectors and refinement identities.
1379 \item Growth entropy and weighted path partition functions.
1380 \item A family certificate with closure, relations, and invariants.
1381 \item A finite explorer that preserves paths, quotient nodes, failures, and residual uncertainty.
1382 \end{enumerate}
1383 
1384 \section{What remains domain dependent}
1385 
1386 The abstract calculus cannot decide which directions are mathematically natural in every subject. It cannot supply a topology when a domain has not chosen one. It cannot infer an exact quotient from an unstable numerical similarity. It cannot guarantee that a family is historically new. It cannot make an infinite uncountable fiber finite.
1387 
1388 These are not defects in the definition. They are the information that a valid realization must provide.
1389 
1390 \section{Meaning of universal}
1391 
1392 Universal means that the same construction accepts many typed realizations. It does not mean that one fixed finite direction list reaches every mathematical object. A universal interface can still require different domain grammars.
1393 
1394 \begin{boundary}[No metaphysical enumeration]
1395 There is no literal operation that tries every conceivable direction in all of mathematics. The formal replacement is exact: declare the admissible direction object, prove that the grammar contains the directions relevant to the domain, and explore finite or parameterized prefixes with coverage certificates.
1396 \end{boundary}
1397 
1398 \begin{boundary}[No finite proof of an infinite search claim]
1399 A finite run proves only the statements certified for that run. Infinite completeness comes from theorems about the grammar, fairness, compactness, convergence, and relations. It does not follow from a large node count.
1400 \end{boundary}
1401 
1402 \begin{boundary}[No novelty from generation alone]
1403 Generating an expression along an unfamiliar route does not make the expression new. A novelty claim requires comparison with known objects, transformations, parameterizations, and identities.
1404 \end{boundary}
1405 
1406 \chapter{Conclusion}
1407 
1408 The proposed object can be summarized in one line:
1409 \[
1410 \boxed{
1411 \begin{gathered}
1412 \text{object}
1413 \longrightarrow
1414 \text{all declared directions at that object}
1415 \longrightarrow
1416 \text{all directions at every child}
1417 \\[0.25em]
1418 \longrightarrow
1419 \text{path forest}
1420 \longrightarrow
1421 \text{relation quotient}
1422 \longrightarrow
1423 \text{completed family universe}.
1424 \end{gathered}
1425 }
1426 \]
1427 
1428 The phrase perturbations of perturbations is therefore not shorthand for a second derivative. It is a recursive closure rule whose local direction space changes with the object. Finite words describe unary routes. Rooted operation trees describe nonlinear interactions. Relations identify different constructions of the same object. Completion gives meaning to infinite routes. Projectors retain local route information. Family certificates identify parameterized structures inside the generated universe.
1429 
1430 This is the final generative framework implemented in the accompanying release. It is broad enough to express algebraic, analytic, geometric, signal, stochastic, quantum, and arithmetic deformation systems, while remaining strict about admissibility, equivalence, convergence, and auditability.
1431 
1432 \appendix
1433 
1434 \chapter{Compact formula registry}
1435 
1436 \begin{longtable}{L{0.31\textwidth}L{0.58\textwidth}}
1437 \caption{Core formula registry.}\label{tab:formula-registry}\\
1438 \toprule
1439 Object & Definition or identity \\
1440 \midrule
1441 \endfirsthead
1442 \toprule
1443 Object & Definition or identity \\
1444 \midrule
1445 \endhead
1446 Direction fiber & $\Dcal_x=\pi^{-1}(x)$ \\
1447 One-step deformation & $x\star d=\Phi(d)$ for $d\in\Dcal_x$ \\
1448 Length-$n$ path layer & $\Tcal_n(x_0)$ \\
1449 Endpoint map & $\ev(d_1,\ldots,d_n)=x_n$ \\
1450 Raw finite forest & $\Tcal_{\mathrm{fin}}(x_0)=\coprod_{n\ge0}\Tcal_n(x_0)$ \\
1451 Path cost & $\cost(\gamma)=\sum_k\kappa(d_k)$ \\
1452 Quotient object graph & $\Ocal(x_0)=\Tcal_{\mathrm{fin}}(x_0)/\sim$ \\
1453 Infinite boundary & $\partial\Tcal(x_0)$ \\
1454 Cover recursion & $e_k\le L_ke_{k-1}+M_k\epsilon_k$ \\
1455 Cover closed bound & $e_n\le (\prod L_k)e_0+\sum_jM_j\epsilon_j\prod_{k>j}L_k$ \\
1456 Summable branch condition & $d(x_{n+1},x_n)\le a_n$, $\sum a_n<\infty$ \\
1457 Cylinder projector & $P^{(n)}_\alpha\lvert\gamma\rangle=1_{\alpha\preceq\gamma}\lvert\gamma\rangle$ \\
1458 Projector refinement & $P_\alpha=\sum_dP_{\alpha d}$ \\
1459 Transfer block & $A_{\beta\alpha}=Q_\beta A P_\alpha$ \\
1460 Path product law & $(BA)_{\gamma\alpha}=\sum_\beta B_{\gamma\beta}A_{\beta\alpha}$ \\
1461 Raw entropy & $h_{\mathrm{raw}}=\limsup n^{-1}\log a_n$ \\
1462 Object entropy & $h_{\mathrm{obj}}=\limsup n^{-1}\log q_n$ \\
1463 Path partition function & $Z_{x_0}(s)=\sum_{\gamma\ne\varnothing}e^{-s\cost(\gamma)}$ \\
1464 Regular-tree threshold & $sc>\log B$ \\
1465 Family closure & $F(\sigma_d(\theta))=d\cdot F(\theta)$ \\
1466 \bottomrule
1467 \end{longtable}
1468 
1469 \chapter{Reproduction commands}
1470 
1471 From PowerShell in the extracted release directory:
1472 
1473 \begin{lstlisting}
1474 python -m pip install -e .
1475 python .\run_all.py
1476 python -m pytest -q
1477 .\Build_Wheel.ps1
1478 .\Run_All.ps1
1479 \end{lstlisting}
1480 
1481 To explore one built-in system directly:
1482 
1483 \begin{lstlisting}
1484 python -m generative_deformation explore --example walk --depth 12 --max-paths 9000 --out .\walk_output
1485 python -m generative_deformation explore --example analytic --depth 3 --max-paths 1800 --out .\analytic_output
1486 \end{lstlisting}
1487 
1488 The paper can be compiled from the \texttt{paper} directory with three passes of \texttt{pdflatex}.
1489 
1490 \chapter{Claim ledger}
1491 
1492 \begin{longtable}{L{0.34\textwidth}L{0.17\textwidth}L{0.39\textwidth}}
1493 \caption{Claim and verification ledger.}\label{tab:claim-ledger}\\
1494 \toprule
1495 Statement & Status & Basis \\
1496 \midrule
1497 \endfirsthead
1498 \toprule
1499 Statement & Status & Basis \\
1500 \midrule
1501 \endhead
1502 Raw forest equals finite admissible routes & proved & induction on path length \\
1503 Quotient extension is well defined & proved & extension-compatible congruence \\
1504 Integer walk has $2^n$ paths and $n+1$ endpoints & proved & exact counting \\
1505 Continuous cover error formula & proved & recursive Lipschitz estimate \\
1506 Summable steps give branch convergence & proved & Cauchy tail bound \\
1507 Multi-input expressions form a free rooted syntax forest & proved & recursive parsing and evaluation \\
1508 Binary interaction shapes have Catalan recurrence & proved & root split \\
1509 Cylinder projectors refine exactly & proved & disjoint descendant cylinders \\
1510 Regular path partition function threshold & proved & geometric series \\
1511 Family closure certificate is inductively sound & proved & induction on parameter words \\
1512 Built-in finite experiments & verified & \BuiltInChecks/\BuiltInChecks~checks passed \\
1513 Independent software suite & verified & \TestCount/\TestCount~tests passed \\
1514 Historical novelty of the combined framework & not claimed & requires external comparison \\
1515 One fixed finite grammar reaches all mathematical objects & not claimed & domain grammars and completeness theorems are required \\
1516 Finite computation exhausts an uncountable direction space & impossible literally & parameterized fibers and covers replace enumeration \\
1517 \bottomrule
1518 \end{longtable}
1519 
1520 \end{document}

Back to source controls

Back to Universal Generative Deformation Calculus