Height Compression Theorem

• Height Compression Theorem is a framework that reduces unbounded invariants into bounded forms across various fields through combinatorial, algebraic, and geometric techniques.
• It provides quasi-polynomial and logarithmic bounds, transforming classical exponential estimates into subexponential or logarithmic growth in contexts like PI-algebras and computation trees.
• The theorem unifies diverse notions of height—including word lengths, evaluation-stack depths, subgroup intersections, and digit expansions—thereby ensuring global structural control and rigidity.

Searching arXiv for recent and foundational uses of “Height Compression Theorem” across domains. In the arXiv literature represented by these works, the expression Height Compression Theorem does not denote a single universal statement. It names a recurrent pattern in which a notion of height—combinatorial, algebraic, geometric, or computational—is shown to admit a stronger-than-expected bound, or to force a rigid structural alternative. The most classical instance in this collection is Shirshov-type height for non-$n$-divided words and PI-algebras, where recursive, double-exponential, and exponential estimates are compressed to quasi-polynomial bounds (Kharitonov, 2014). Closely related usages appear in computation-tree reshaping, where evaluation-stack depth is compressed to $O(\log T)$ (Nye, 20 Aug 2025); in relatively hyperbolic splittings, where finite relative height becomes equivalent to relative quasiconvexity (Pal, 2015); and in arithmetic settings where coefficient heights, subspace heights, or group-element heights are bounded by finite alphabets, explicit isotropic families, or sharp reduction bounds (Akiyama et al., 2012, Akiyama et al., 2014, Chan et al., 2014, Dória et al., 2019, Deitmar et al., 2015).

1. Range of meanings

Across these works, “height” refers to distinct invariants. In Shirshov theory it is the bounded product length over short words. In the 2025 computation-theoretic result it is evaluation-stack depth along DFS paths in a binary evaluation tree. In relatively hyperbolic groups it is relative height, defined through intersections of distinct conjugates containing a loxodromic element. In the height reducing property it is coefficient size in $\mathbb{Z}[\alpha]$-representations. In geometric analysis it is literal distance from a reference hyperplane or slab. In arithmetic geometry it is the absolute multiplicative height of vectors, subspaces, or matrix entries (Kharitonov, 2014, Nye, 20 Aug 2025, Pal, 2015, Akiyama et al., 2012, Monti et al., 2014, Chan et al., 2014, Dória et al., 2019).

Domain	Height notion	Compression statement
PI-algebras and words	Shirshov height over words of degree $<n$	Non $n$-divided words have $h<\Phi(n,l)$, and long words are either $n$-divided or contain a $d$-th power (Kharitonov, 2014)
Computation trees	Evaluation-stack depth $H'_{\mathrm{eval}}$	Canonical left-deep trees are transformed to binary trees with $H'_{\mathrm{eval}}=O(\log T)$ (Nye, 20 Aug 2025)
Relatively hyperbolic groups	Relative height of subgroups	Vertex groups are relatively quasiconvex iff they have finite relative height (Pal, 2015)
$O(\log T)$0 digit systems	Coefficient height / finite digit alphabet	$O(\log T)$1 for finite $O(\log T)$2 in the admissible modulus regimes (Akiyama et al., 2012, Akiyama et al., 2014)
Sub-Riemannian and minimal geometry	Geometric height above a hyperplane or slab	Small excess or sublinear height forces quantitative flatness or Euclidean volume growth (Monti et al., 2014, Colding et al., 14 May 2026)
Arithmetic and adelic groups	Heights of subspaces or group elements	Small-height isotropic families, quadratic reduction bounds, and dominant asymptotic compression are obtained (Chan et al., 2014, Dória et al., 2019, Deitmar et al., 2015)

This multiplicity of meanings is a source of frequent confusion. In particular, the phrase does not refer only to geometric height estimates, nor only to Shirshov’s theorem. Each usage is local to its domain, though the common template is the replacement of a large or poorly controlled object by a bounded-height decomposition, bounded-depth evaluation, or rigidity alternative.

2. Shirshov-type height compression in PI-algebras

The most explicit combinatorial-algebraic formulation appears in “Estimates in Shirshov height theorem” (Kharitonov, 2014). Let $O(\log T)$3 be a finite alphabet with lexicographic order. A word $O(\log T)$4 is called $O(\log T)$5-divided if it can be written as

$O(\log T)$6

For a set $O(\log T)$7, the height $O(\log T)$8 of a set $O(\log T)$9 is the smallest $\mathbb{Z}[\alpha]$0 such that every $\mathbb{Z}[\alpha]$1 either is $\mathbb{Z}[\alpha]$2-divided or admits a factorization $\mathbb{Z}[\alpha]$3 with $\mathbb{Z}[\alpha]$4 and $\mathbb{Z}[\alpha]$5. Shirshov’s Height Theorem, in the form used there, states that if $\mathbb{Z}[\alpha]$6 is a finitely generated associative algebra over a commutative ring satisfying an admissible polynomial identity of degree $\mathbb{Z}[\alpha]$7, then every element of $\mathbb{Z}[\alpha]$8 is a linear combination of words $\mathbb{Z}[\alpha]$9 with $<n$0 and each $<n$1 of length $<n$2. Thus the set of non $<n$3-divided words has bounded height over the set $<n$4 of words of degree $<n$5.

The paper proves a quantitative dichotomy for long words. If $<n$6 are positive integers and $<n$7, then every word over an $<n$8-letter alphabet of length greater than

$<n$9

is either $n$0-divided or contains a contiguous $n$1-th power $n$2. A second explicit estimate is

$n$3

In compressed asymptotic form,

$n$4

for an absolute constant $n$5. The corresponding height bound for non $n$6-divided words over $n$7 is

$n$8

hence

$n$9

The paper also bounds the essential height by

$h<\Phi(n,l)$0

These bounds sharpen a clear historical sequence: Shirshov’s original proof produced only recursive estimates; Kolotov obtained a double-exponential bound in 1982; Belov obtained an exponential bound in 1993; the 2014 result compresses the dependence to subexponential, quasi-polynomial form. Its algebraic consequence is immediate for $h<\Phi(n,l)$1-generated associative algebras satisfying $h<\Phi(n,l)$2: the nilpotency degree is $h<\Phi(n,l)$3. This gives a negative answer to Zelmanov’s 1993 question whether the nilpotency degree of the $h<\Phi(n,l)$4-generated free associative algebra $h<\Phi(n,l)$5 with identity $h<\Phi(n,l)$6 grows exponentially in $h<\Phi(n,l)$7; the obtained upper bound is subexponential.

The proof mechanism is based on Latyshev’s use of Dilworth’s theorem. Suitable posets are built from tails or cyclic shifts of a word, ordered by lexicographic comparison together with left-to-right precedence. Chain decompositions induce a bounded coloring of positions, and one tracks the evolution of colored configurations $h<\Phi(n,l)$8 or $h<\Phi(n,l)$9. A key process inequality is

$n$0

where $n$1 is the chain count and $n$2 measures the persistence of unchanged configurations at scale $n$3. Iteration over geometric schedules $n$4 or $n$5 yields quasi-polynomial bounds and ultimately forces long words either into $n$6-division or into high periodicity.

3. Tree height compression in deterministic time-space simulation

A formally different use of the term appears in “$n$7 via Tree Height Compression” (Nye, 20 Aug 2025). Here the setting is a deterministic multitape Turing machine running for $n$8 steps with a chosen block size $n$9, and $d$0. The run is partitioned into blocks $d$1, each with a summary $d$2 of size $d$3 storing entering and leaving control state, head positions, window endpoints and offsets, a movement log of at most $d$4 micro-ops, and constant-size checksums. Interval summaries $d$5 are composed using an associative merge operator $d$6, with semantic correctness across adjacent intervals certified by exact $d$7 window replay at the unique interface.

The canonical object is a left-deep succinct computation tree $d$8 over $d$9. The Height Compression Theorem states that there is a uniform, logspace-computable transformation of $H'_{\mathrm{eval}}$0 into a binary evaluation tree $H'_{\mathrm{eval}}$1 with an evaluation schedule such that: along any DFS root-to-leaf traversal, the evaluation-stack depth satisfies

$H'_{\mathrm{eval}}$2

workspace at leaves is $H'_{\mathrm{eval}}$3 cells and at internal nodes $H'_{\mathrm{eval}}$4 cells; topology predicates on edges are checkable in $H'_{\mathrm{eval}}$5 space; and the root summary computed by $H'_{\mathrm{eval}}$6 equals the root summary computed by $H'_{\mathrm{eval}}$7. The resulting space tradeoff is

$H'_{\mathrm{eval}}$8

for block sizes $H'_{\mathrm{eval}}$9 with $H'_{\mathrm{eval}}=O(\log T)$0, and the canonical choice $H'_{\mathrm{eval}}=O(\log T)$1 yields

$H'_{\mathrm{eval}}=O(\log T)$2

The compression mechanism is based on midpoint recursion, constant-workspace balanced binary combiners, and a per-path potential function

$H'_{\mathrm{eval}}=O(\log T)$3

where $H'_{\mathrm{eval}}=O(\log T)$4 is the multiset of active interfaces along the current DFS stack. Since active interval length shrinks geometrically under midpoint recursion, $H'_{\mathrm{eval}}=O(\log T)$5 throughout. The evaluator further uses an Algebraic Replay Engine with constant-degree maps over a constant-size field, pointerless DFS, and index-free streaming, so that per-level tokens remain constant-size and no per-level $H'_{\mathrm{eval}}=O(\log T)$6 term accumulates.

The consequences are explicitly algorithmic. A size-$H'_{\mathrm{eval}}=O(\log T)$7 bounded-fan-in circuit can be simulated in $H'_{\mathrm{eval}}=O(\log T)$8, yielding branching-program size $H'_{\mathrm{eval}}=O(\log T)$9. For $O(\log T)$00-complete languages under logspace reductions, the inclusion implies $O(\log T)$01 time lower bounds infinitely often via the deterministic space hierarchy. The same framework also gives $O(\log T)$02-space certifying interpreters and extends, under explicit locality assumptions, to geometric $O(\log T)$03-dimensional models.

4. Height compression in group theory and reduction theory

In geometric group theory, “height compression” appears as a relation between subgroup intersection complexity and ambient quasiconvexity. In a finite graph of relatively hyperbolic groups whose fundamental group is relatively hyperbolic, with edge groups quasi-isometrically embedded and relatively quasiconvex in the vertex groups, the main theorem proves that a vertex group $O(\log T)$04 is relatively quasiconvex in the ambient group $O(\log T)$05 if and only if $O(\log T)$06 has finite relative height in $O(\log T)$07 (Pal, 2015). The relevant invariant is

$O(\log T)$08

The direction from relative quasiconvexity to finite relative height is supplied by Hruska–Wise; the converse is proved through hallways and hyperbolic ladders in the coned-off tree of spaces. If a vertex group were not relatively quasiconvex, one obtains arbitrarily long $O(\log T)$09-boundary thin hallways; gluing arguments then force intersections of arbitrarily many distinct conjugates to contain a loxodromic element, contradicting finite relative height. This theorem compresses a global geometric property into a bounded intersection invariant.

A different reduction-theoretic instance occurs for Bianchi groups. For $O(\log T)$10, $O(\log T)$11, and the Ford fundamental domain $O(\log T)$12, define

$O(\log T)$13

The theorem states that there exists $O(\log T)$14 such that

$O(\log T)$15

the exponent $O(\log T)$16 is sharp, and $O(\log T)$17 for a universal constant $O(\log T)$18 (Dória et al., 2019). The proof combines a quantitative Bézout lemma in $O(\log T)$19, a first step moving a point into the Bianchi–Ford region $O(\log T)$20, and a parabolic translation into the fundamental parallelogram $O(\log T)$21. The same bound yields effective reduction of positive definite binary Hermitian forms $O(\log T)$22: $O(\log T)$23 for a reducing transformation $O(\log T)$24.

In semisimple groups, the phrase is used interpretively for adelic height asymptotics. For a connected semisimple algebraic group $O(\log T)$25 over $O(\log T)$26, local intrinsic heights are defined by Bruhat–Tits building distances at finite places and a Weyl-invariant metric at the archimedean place. The archimedean factor has the form

$O(\log T)$27

where $O(\log T)$28 is the Cartan projection and $O(\log T)$29 is the Harish–Chandra shift. When $O(\log T)$30, the finite-place contribution compresses into the Euler-product constant

$O(\log T)$31

and the rational-point counting function satisfies

$O(\log T)$32

(Deitmar et al., 2015). A plausible interpretation is that the global height asymptotic is controlled by the dominant archimedean slope $O(\log T)$33, while finite places contribute only the multiplicative constant $O(\log T)$34.

5. Arithmetic height reduction and small-height isotropic families

In the arithmetic theory of $O(\log T)$35, the central notion is the height reducing property. For an algebraic number $O(\log T)$36 and a finite subset $O(\log T)$37,

$O(\log T)$38

The statement $O(\log T)$39 means that every element of $O(\log T)$40 has a finite $O(\log T)$41-adic expansion with digits in a fixed finite alphabet. The structural theorem says that if $O(\log T)$42 satisfies this property, then $O(\log T)$43 is algebraic and either all conjugates have modulus $O(\log T)$44 or all have modulus $O(\log T)$45; conversely, roots of unity and algebraic numbers all of whose conjugates have modulus $O(\log T)$46 satisfy the property (Akiyama et al., 2012). A further result gives broad sufficient conditions in the unit-circle case: if $O(\log T)$47 or $O(\log T)$48, then $O(\log T)$49 satisfies the height reducing property. The constructive proof uses a digit map $O(\log T)$50, quantitative Kronecker approximation, and a contraction argument in the embedding

$O(\log T)$51

The companion paper sharpens the arithmetic compression viewpoint (Akiyama et al., 2014). If $O(\log T)$52 with $O(\log T)$53 finite, then there exists a finite $O(\log T)$54 such that $O(\log T)$55, and any minimal digit set must satisfy

$O(\log T)$56

where $O(\log T)$57 is the minimal polynomial. The same paper gives an automaton-theoretic algorithm for the minimal height polynomial of $O(\log T)$58, provided $O(\log T)$59 has no conjugate of modulus one. For fixed $O(\log T)$60, a finite automaton $O(\log T)$61 recognizes words $O(\log T)$62 with

$O(\log T)$63

Increasing $O(\log T)$64 until a nontrivial accepted word appears determines the minimal possible coefficient height of a nonzero polynomial relation for $O(\log T)$65.

A different arithmetic height compression theorem concerns quadratic spaces. Let $O(\log T)$66 be a global field or $O(\log T)$67, let $O(\log T)$68 be a quadratic form on $O(\log T)$69, and let $O(\log T)$70 be an $O(\log T)$71-dimensional subspace such that $O(\log T)$72 has rank $O(\log T)$73 and Witt index $O(\log T)$74. Then there exists an infinite collection of finite families

$O(\log T)$75

of maximal totally isotropic subspaces such that $O(\log T)$76 for every $O(\log T)$77, distinct members have controlled intersections, and

$O(\log T)$78

(Chan et al., 2014). Here

$O(\log T)$79

and for global fields

$O(\log T)$80

The construction uses effective Witt decomposition, Siegel’s lemma on the anisotropic part, and explicit hyperbolic pairs of controlled height. This is a literal compression of ambient heights $O(\log T)$81 and $O(\log T)$82 into many small-height isotropic subspaces that still generate the full quadratic space.

6. Geometric height estimates and rigidity from small height

In geometric analysis, height compression becomes a quantitative flatness statement. For $O(\log T)$83-minima of perimeter in the Heisenberg group $O(\log T)$84, with $O(\log T)$85, let $O(\log T)$86 be the height above the vertical hyperplane $O(\log T)$87, and let $O(\log T)$88 denote cylindrical excess with respect to $O(\log T)$89. The main estimate states that there exist $O(\log T)$90 such that if $O(\log T)$91 is a $O(\log T)$92-minimum in $O(\log T)$93, $O(\log T)$94, $O(\log T)$95, and $O(\log T)$96, then

$O(\log T)$97

The proof uses a new coarea formula for rectifiable sets in $O(\log T)$98, projection identities, slice-wise isoperimetry, and density estimates (Monti et al., 2014). The restriction $O(\log T)$99 is essential; the estimate fails for $\mathbb{Z}[\alpha]$00.

For minimal submanifolds in Euclidean space, the theorem is global rather than local. A complete proper stationary integral varifold whose height grows sublinearly must have Euclidean volume growth (Colding et al., 14 May 2026). In scale-local form, if

$\mathbb{Z}[\alpha]$01

on $\mathbb{Z}[\alpha]$02, then

$\mathbb{Z}[\alpha]$03

In a slab, the asymptotic volume ratio converges to an integer $\mathbb{Z}[\alpha]$04 with rate $\mathbb{Z}[\alpha]$05: $\mathbb{Z}[\alpha]$06 For stable minimal hypersurfaces, sublinear height is decisive: a complete properly immersed two-sided stable minimal hypersurface with sublinearly growing height is a hyperplane. The paper identifies the threshold as sharp, since stable minimal cones such as the Simons cone have exactly linear height growth.

These geometric results clarify a common misconception. In this literature, “compression” need not mean a combinatorial factorization or a computational tree transformation. It may mean that small oscillation of the normal, or sublinear confinement in a slab, compresses the geometry enough to force quantitative flatness, Euclidean growth, or global rigidity.

The broad commonality among these theorems is structural rather than formal. In each setting, a large ambient class is reduced to a bounded-height or bounded-complexity model: words become products of powers of short words; left-deep computation trees become logarithmic-depth evaluation trees; subgroup intersection patterns collapse to finite relative height; $\mathbb{Z}[\alpha]$07 expansions use a fixed finite alphabet; quadratic spaces admit spanning families of small-height isotropic subspaces; and geometric objects with small excess or sublinear height are forced toward flat models. This suggests that “height compression” is best understood as a cross-disciplinary paradigm of rigidity-by-bounded-height, with each field supplying its own invariant, proof technology, and sharpness phenomena.