Kl.Reflection Combinatorics

87 nodes 25 formalized 62 proved

Reflection orders, dihedral combinatorics, and paths

Root geometry supplies an induced dihedral Coxeter system inside each root plane. This topic uses those systems to describe relative reflection graphs, recover the full reflection-edge relation from order data, and interpret ordered edge products as path polynomials.

The inputs are positive roots, their reflection correspondence, and ambient inversion counts. The outputs are the relative dihedral structures, edge reconstruction and transport, and the path identities and plane counts used by the later sheaf comparison.

Geometric orders on positive roots

Choose a linear form \(h\) positive on every simple root. Since a positive root is a nonnegative, nonzero combination of simple roots, \(h(\alpha)>0\) for every positive root \(\alpha\). Choose another functional \(f\) for which the ratios \[ s(\alpha)=\frac{f(\alpha)}{h(\alpha)} \] are all distinct.

This assigns each positive root a real score. Equivalently, divide each root by \(h(\alpha)\), placing it on the affine hyperplane \(h=1\), and then evaluate \(f\).

For two independent roots in a plane, their normalized points and all other normalized roots in that plane lie on an affine line. A linear functional orders points on a line in one of its two directions. This explains the rank-two angular-order condition behind a geometric reflection order.

The generic-functional theorem proves that \(f\) can be chosen to separate the whole root set in finite generator rank. For a fixed interval only finitely many edges need to be listed, even if the group has infinitely many positive roots.

Chambers and the least simple root

A ratio order consists of a positive denominator \(h\) and a numerator \(f\) separating the positive-root ratios. Its chamber condition at \(y\) is \[ f(\rho(y)\alpha_i)>0\quad\text{for every simple root }\alpha_i. \] The chamber and least-descent theorem transports this positivity through \(\rho(y^{-1})\). It identifies negative slopes with the inversions of \(y\). A minimum among the finitely many simple-root slopes then bounds every positive-root slope, by the nonnegative coordinate expansion of a root. For \(y\ne1\), this minimum is negative and its simple reflection is a left descent.

The prescribed-chamber existence theorem begins with \(f_0=h\circ\rho(y^{-1})\), using height for \(h\). It perturbs \(f_0\) by \(f_\tau=f_0+\tau m_\tau\), where the coordinate-polynomial construction makes evaluation at each fixed vector polynomial in \(\tau\). A finite open positivity neighborhood contains zero; the equal-slope parameters form a countable union of finite polynomial zero sets. Choosing a parameter in the neighborhood outside that union gives a generic order in the desired chamber.

Rotation of the chamber order

Let \(s=s_i\) be the least-slope simple descent. The rotated ratio-order construction is \[ f'=f\circ\rho(s),\qquad h'=(f-kh)\circ\rho(s). \] The order-rotation theorem chooses \(k<0\) from finitely many positive simple-root image gaps and verifies \(h'>0\) on every simple root. It puts \(f'\) in the chamber of \(sy\).

For every other positive root with old slope \(u\), the new slope at its positive reflected root is \(u/(u-k)\). Positive-denominator cross multiplication preserves the strict comparisons. Those new slopes lie below one, while the chosen simple root has new slope above one. It therefore moves from least to greatest.

With the running \(A_2\) data \(f(a)=-1,f(b)=3,h=\mathrm{height}\), take \(k=-1/2\) and reflect in \(a\). The old slopes of \(a,a+b,b\) are \(-1,1,3\); the new slopes of \(b,a+b,a\) are \(2/3,6/7,2\). The two orders are the inputs to the path calculation below.

Angular endpoints from finite inversion sets

Fix a root plane \(P\). The induced subgroup will be generated by its two angular endpoint reflections. To construct the endpoints, begin with the finite set of roots in \(P\) inverted by one chosen reflection.

Take its minimum and maximum slopes. If neither were an extreme of the entire plane, there would be positive roots outside both ends. The root defining the chosen reflection would lie in their positive cone. Both outside roots would remain positive under that reflection, forcing their positive combination to remain nonnegative, although the defining root is sent to its negative. This contradiction gives a global extreme.

An extreme root inverts only itself among the positive roots of the plane. Pairing it with another root of nonpositive Tits pairing and repeating the finite-inversion argument produces the opposite extreme. Thus every root plane has a plane basis.

Generation measured by the plane inversion count

Define \[ L_P(w)=\#\{a\in\Phi^+\cap P:w^{-1}a\text{ is negative}\}. \] This count is finite because it selects a subset of the ambient inversions. A plane-simple reflection changes it by \(+1\) or \(-1\), according to the sign of its root. The plane-weight identity proves this by making the extra plane contribution zero.

Every plane reflection inverts at least one endpoint: otherwise it would preserve positivity of the entire cone between them, including its own root. For a reflection different from that endpoint reflection, conjugating by the endpoint lowers \(L_P\) by two.

Strong induction on this nonnegative count proves that the endpoint reflections generate every reflection in \(P\), hence all of \(D_P\). The subgroup generated by two distinct involutions has the dihedral Coxeter presentation with exponent \(\operatorname{ord}(st)\), including the infinite-order case. This constructs the induced system on the actual subgroup.

Signed roots identify the intrinsic inversions

The induced signed relation records three pieces of data: an ambient root representing a subgroup element, membership in \(P\), and its positive or negative sign. A simple endpoint reflection changes the sign exactly at its own positive root, matching the signed Coxeter action.

Applying this compatibility to a reduced induced word identifies negative ambient root images with its intrinsic inversion sequence. It follows that intrinsic reflections are exactly the ambient reflections belonging to \(D_P\), and \[ \ell_P(w)=L_P(w). \] Thus the group presentation and the ambient root count describe the same length.

Least coset representatives and relative intervals

Choose \(a_0\) of least ambient length in \(D_Pa\). Its inverse preserves every positive plane root: a negative image would make the corresponding plane reflection lower its length. It preserves negative roots by linearity as well.

Consequently \(w^{-1}r\) and \((wa_0)^{-1}r\) have the same sign for every plane root \(r\). This identifies the induced reflection graph with the ambient reflection graph restricted to the coset, preserving left labels, and gives \[ L_P(wa_0)=\ell_P(w). \] The count also proves uniqueness of \(a_0\): another minimum has zero plane inversions, so its relative group factor has length zero.

Relative Bruhat order is reachability by these edges inside the coset. Its intervals map monotonically and injectively into ambient intervals. In type \(A_3\), the plane generated by \(\alpha_1+\alpha_2\) and \(\alpha_3\) has an induced simple reflection of ambient length three. Its two-vertex relative interval sits inside a larger ambient interval containing vertices outside the subgroup.

Reflections and odd length gaps

In a two-generator Coxeter system, every odd alternating word is a reflection: the length-one case is simple, and a longer odd alternating word is a conjugate of a shorter odd alternating word by its outer simple generator. Every element has a reduced alternating expression. Together with parity of reflection length, this gives \[ w\text{ a reflection}\Longleftrightarrow\ell(w)\text{ odd}. \]

Since \(\ell(yx^{-1})\equiv\ell(y)+\ell(x)\pmod2\), a directed reflection edge is exactly \[ \ell(x)<\ell(y),\qquad \ell(y)-\ell(x)\text{ odd}. \] The odd-gap edge theorem records both directions. It also makes the product of two elements of opposite parity a reflection after inversion of either one.

For interpolation, an opposite-parity external vertex \(y\le c\), \(y\not\le x\), has length greater than \(x\): a shorter vertex would lie below \(x\) by the dihedral Bruhat comparison. Thus these vertices are precisely the supported outgoing neighbors of \(x\).

Finite subsets inside a principal dihedral interval

To transport a model from an interval to a finite relative lower ideal \(K\), choose a common upper bound for its vertices. If dihedral lengths are bounded, an element of maximal length has both simple generators as descents; induction on generator words shows that every element lies below it. If lengths are unbounded, choose an element longer than every vertex of \(K\), and use the dihedral comparison of strictly different lengths.

The finite-common-upper-bound theorem therefore places \(K\) inside an intrinsic interval \([1,d_0]_D\). The inclusion preserves and reflects the intrinsic order. This gives a finite interval model whose supported edges can be pulled back to the coset graph.

Signed evaluations of alternating words

Write \(s,t\) for the two simple generators and \(r=st\). The signed word values are \[ w_{2n}=r^n,\qquad w_{2n+1}=r^ns\qquad(n\in\mathbb Z). \] Positive indices evaluate alternating words beginning with \(s\); negative indices begin with \(t\). In particular \[ w_{-2}=ts,\quad w_{-1}=t,\quad w_0=1,\quad w_1=s,\quad w_2=st. \] For a negative odd index, the integer quotient matters: \(w_{-1}=r^{-1}s=t\).

Every group element has a representative whose absolute index is its length. To prove that both rays are reduced through a rank \(n\), choose an element of length at least \(n\), take a reduced alternating prefix and exchange the generators by the diagram automorphism. The signed-length theorem gives the full argument.

For an odd reflection parameter \(q\), the signed reflection-action theorem gives \[ w_qw_i=w_{q-i}. \] Consequently the left reflection of an edge \(w_i\to w_j\) has parameter \(i+j\), its doubled midpoint.

Vertex representatives and finite reflection parameters

If a maximum length \(m\) exists, both generators descend at a maximum-length element. Its two alternating expressions give \(w_m=w_{-m}\), and the even-shift identity \(w_{i+2n}=r^nw_i\) then gives \(r^m=1\). Signed values are periodic with period \(2m\).

The admissible signed indices select \(-m<i\le m\). In type \(A_2\), this gives \(-2,-1,0,1,2,3\), with \(w_{-3}=w_3=sts\).

Object Coordinate convention in the bounded case
Vertex Choose \(i\) with \(-m<i\le m\) and \(\ell(w_i)=|i|\).
Reflection label Use the odd endpoint sum \(i+j\), then its residue modulo \(2m\).

When lengths are unbounded, every integer is admissible. The period-and-representative theorem proves both regimes and the repair of the negative longest index.

Two internal vertices and one top

At a positive rank \(n\) below an existing larger length, there are exactly two elements. If the two alternating words of length \(n\) were equal, prefixing one simple generator would identify a reduced word of length \(n+1\) with a word of length \(n-1\), a contradiction. Their distinctness also gives the unique simple descent at an internal rank.

In the lower interval below \(c\), each internal positive rank therefore contributes \(w_n,w_{-n}\), while the top rank contributes only \(c\). For a function with \(f(1)=0\), \[ \sum_{z\le c}f(z)= \sum_{0\le n<\ell(c)}(f(w_n)+f(w_{-n}))+f(c). \] The doubled zero index contributes zero. The lower-interval signed-sum theorem explains this exact reindexing.

The signed comparison and its residue form

The Before relation places positive signed parameters before negative ones, and uses the ordinary increasing order within each sign. The numerical arrow indicator is \[ \operatorname{aft}(i,q,z) =[\,|i|<|z|,\ i+z\text{ odd},\ q\prec i+z\,]. \] Here brackets denote the zero-or-one indicator of the displayed conditions.

For \(|i|<|j|\), the sign of \(i+j\) is the sign of \(j\). The signed-comparison theorem evaluates the same-source and next-source tests using this observation. For example, when \(j>0\), a negative endpoint from \(i\) passes the order test, while a positive endpoint must lie beyond \(j\).

For finite reflection parameters, the residues are ordered increasingly. The finite-residue comparison theorem proves that this residue test agrees with the raw Before test in the two incident-edge configurations, using their admissible endpoint bounds. For \(m=3\), the same-source comparison of parameters \(3\) and \(-1\) becomes \(3<5\) modulo \(6\).

The numerical count with a moving top

Put \(L=|i|\), \(M=|j|\), \(C=|c|\), and assume a positive odd gap \(M-L=2k+1\). The eligible internal ranks below \(M\) are exactly \[ L+1,L+3,\ldots,L+2k-1, \] with bijection \(a\mapsto L+1+2a\) from \(0\le a<k\).

Above a reference rank, an eligible internal level contributes two signed endpoints, and the top contributes one if its parity is eligible. The parity-rank count theorem proves by induction that increasing the top rank adds exactly one: an eligible old top changes from one contribution to two while the next top contributes zero, or an ineligible old top contributes zero and the next top contributes one. The initial boundary step is checked separately.

For the lower source with \(j>0\), the first step above rank \(M\) adds the opposite endpoint at rank \(M\). The source's positiveCount induction then gives \(k+(C-M)\). For \(j<0\), the signed comparison removes every contribution at or above rank \(M\), leaving \(k\).

The signed-count difference theorem obtains the complete table: \[ \begin{array}{c|cc} &h_i&h_j\\ j>0&k+(C-M)&C-M\\ j<0&k&0 \end{array} \] under the top condition \(c=j\) or \(M<C\). Hence \(h_i=h_j+k\).

For the long \(A_2\) edge with indices \(0\to3\) and top \(3\), the counts are \(1,0\). For indices \(1\to2\) with top \(3\), they are \(1,1\); the covering gap has exponent zero.

Actual roots, inversion cuts, and signed labels

The integer labels acquire their geometric order through the actual roots of the ambient Coxeter system. Choose a plane basis \(a_L,a_R\) in increasing geometric slope, and let \(D\) be its induced dihedral subgroup. An intrinsic edge \(x\to_Dy\) has an actual plane edge root: the unique positive ambient root representing \(yx^{-1}\).

The first connection is between signs and order. If \(a_L\) is inverted at an ambient element \(w\), an inverted root \(a\) must precede a noninverted root \(b\) in this plane. Otherwise \(b\) lies strictly between \(a_L\) and \(a\) in slope, and is a positive linear combination of them. Applying \(w^{-1}\) would make \(b\) negative as well. This is the inversion-cut theorem. Its converse follows from the fact that every positive plane root lies in the nonnegative cone of the two endpoints.

On a two-edge path \(x\to y\to z\), the incoming root is inverted at \(y\), whereas the outgoing root is noninverted there. The inversion cut therefore makes increasing root slopes equivalent to inversion of \(a_L\) at \(y\). The inversion-descent bridge identifies that sign with the intrinsic simple descent \(\ell_D(sy)<\ell_D(y)\). This gives a test using the middle vertex alone.

The two prefix-root sequences apply this test to the paths of signed indices \[ 0,1,2,3,\ldots \qquad\text{and}\qquad 0,-1,-2,-3,\ldots. \] The first sequence increases in slope. In the second sequence, each internal middle vertex has the other simple descent, and the unique-descent theorem excludes the first one; the slopes decrease. A witness \(v\) of length \(N\) makes both arguments valid at indices \(n<N\).

There is one further comparison in the unbounded case. For reflection parameters \(2k+1\) and \(-(2l+1)\), the element \(w_{k+1}\) separates their inversion signs: multiplication by the first reflection gives \(w_k\), of smaller length, while multiplication by the second gives \(w_{-(2l+k+2)}\), of greater length. Since \(a_L\) is inverted at \(w_{k+1}\), the inversion cut puts every positive parameter before every negative one.

The angular-order theorem now follows. In the unbounded case, the two monotone sequences and their separation give Before on odd integers. In the maximum-length case, with maximum \(m\), every odd parameter reduces to \(2n+1\) with \(0\le n<m\); a single prefix sequence orders the residues \(1,3,\ldots,2m-1\).

Actual indicator sums and the native count

For a threshold root \(\alpha\), the actual dihedral after-indicator \(\varepsilon_D(x,\alpha,z)\) is one exactly when \(x\to_Dz\) is an edge with larger root slope. The native later-edge count \(h_D(x,\alpha;c)\) sums it over \(z\le_Dc\).

Consider a current edge \(w_i\to w_j\). The angular-indicator theorem identifies the two endpoint indicators with \[ \varepsilon_D(w_i,\alpha,w_z)=\operatorname{aft}(i,i+j,z), \qquad \varepsilon_D(w_j,\alpha,w_z)=\operatorname{aft}(j,i+j,z). \] The same-source formula requires admissible \(j,z\); the next-source formula requires admissible \(z\). Both retain the correct absolute lengths of the three signed words. In the finite case these conditions are exactly what converts residue comparison to the Before tests.

The lower-interval signed-sum theorem then identifies the actual counts with the numerical counts: each positive internal rank contributes its two vertices, the top occurs once, and the identity contributes zero. The native count-difference theorem applies the complete integer calculation to obtain \[ h_D(x,\alpha_{x\to y};c) =h_D(y,\alpha_{x\to y};c) +\frac{\ell_D(y)-\ell_D(x)-1}{2}, \qquad x\to_Dy\le_Dc. \]

The full relative interval in the target

Start with a relative interval \(J=[g,k]_D\) in a plane subgroup and a least coset representative \(a_0\). Ambient endpoint bounds make the coset interval embedding \[ z\longmapsto za_0\in[u,b]_W \] well-defined. Composing it with an ambient interval order isomorphism gives \(f(z)=\varphi(za_0)\). Reflection-graph invariance makes this an injective map preserving the directed reflection edges.

For relative rank at least two, the first internal rank has exactly two vertices. Their image edges from the bottom give two independent root directions and hence a plane \(P'\). The two-edge plane theorem explains how this plane propagates. In a diamond \(p\to q,s\to z\), the two reflection products are equal because both paths have the same endpoints. The product determines the common root plane. Once \(f(p),f(q),f(s)\) lie in the plane coset, the last edge puts \(f(z)\) there too. The image-plane theorem carries this out by induction, with all vertex choices inside \(J\).

Choose a least representative \(a'_0\) of that target coset. The intrinsic coordinate of each image is \[ F(z)=f(z)(a'_0)^{-1}\in D',\qquad F(z)a'_0=f(z). \] This equality is later used to identify actual reflection labels and root thresholds.

The next step determines relative ranks. A source pair two ranks apart has exactly two intermediate vertices for two-edge paths, including ambient paths. Coset confinement proves this ambient assertion. Any target intermediate lies between the image endpoints in the full ambient interval, so the interval isomorphism pulls it back. The exact two-element set is therefore preserved.

In a dihedral group, a two-edge path has a positive even total length gap. If that gap were at least four, two vertices one rank above the bottom and a third three ranks above it would all be intermediates. Their distinctness contradicts the two-element bound. This is the rank-two detection theorem used by Lean.

Now place a consecutive source pair in a path of total gap two. If its upper vertex is below the interval top, extend above; at the top, extend below. The two positive target gaps sum to two, so each is one. The affine-rank theorem then inducts upward to obtain \[ \ell_{D'}(F(z))+\ell_D(g) =\ell_{D'}(F(g))+\ell_D(z). \]

For example, source lengths \(2,3,4,5\) and target bottom length \(4\) force target lengths \(4,5,6,7\). The relative rank fibers have sizes \(1,2,2,1\) on both sides. This illustrates the constant shift in absolute intrinsic lengths.

To prove the full range, take an interior target vertex of length \(n'\). The affine formula determines a strict internal source rank. Its two vertices have distinct images of length \(n'\). The target dihedral group has exactly two elements of that length below the image top, so the chosen target vertex is already one of these images. Endpoints are images by construction. The full-range theorem thus gives \[ \operatorname{range}(F)=[F(g),F(k)]_{D'}. \] The same rank identity and injectivity preserve and reflect order; the positive odd-gap criterion preserves and reflects reflection edges. The full relative-interval transport theorem assembles these facts with the exact coordinate equation.

Reindexing the transported root counts

For the current edge \(x\to_Dy\), let \(\alpha'\) be the positive root of its ambient target image. The ambient after-indicator counts an edge when its target root slope is strictly greater than this threshold.

Multiplication by the target representative preserves the left reflection: \[ (F(z)a'_0)(F(p)a'_0)^{-1}=F(z)F(p)^{-1}. \] Thus the actual root is unchanged. The coset-count transport theorem identifies each ambient after-indicator with its intrinsic counterpart, checking both the edge and nonedge cases.

Injectivity and full range give a bijection between the source relative interval and the target relative interval. Reindexing either endpoint count by this bijection is therefore exact. The sum may then be enlarged to the full lower ideal of the target top: a removed vertex is not above the interval bottom, so it cannot be the target of an outgoing edge from either current source.

The native count-difference theorem now applies in the target dihedral group. The affine rank identity converts its correction back to \[ \frac{\ell_D(y)-\ell_D(x)-1}{2}. \] For a relative interval of rank one, there are only the current endpoints. The two zero properties of the ambient after-indicator show that every term vanishes, and the correction is also zero. This completes the transported-count theorem in every rank.

Reflection-edge reconstruction and transported schedules

An interval order isomorphism preserves covers directly. Reflection-edge reconstruction extends this transport to longer edges, such as \(1\to w_0\).

Diamond closure starts from covers and repeatedly applies rank-two configurations visible in the order. One theorem shows that the procedure adds only genuine reflection edges; the other shows that it obtains every reflection edge. Together they identify the recovered relation with the actual reflection graph.

Now an isomorphism \(\phi\) transports each source reflection edge to a target reflection edge. We can assign it the score of that target edge and list the source edges in that order. This is the transported schedule.

The scores come from the target realization. Relative-dihedral transport identifies the relevant rank-two configurations, and its counting theorems provide the schedule properties used in the source comparison.

The local model comparison specializes the transported schedule to two outgoing selections, one at each endpoint of the current edge. Both use the target root of that edge as a common slope threshold. The coset later-edge selections entry defines these sets, and the supported-edge counting theorem reindexes them by target vertices in the full relative interval. This identifies the two finite indicator sums whose numerical difference is computed by the dihedral count theorem.

Chronological suffixes and strict later scores

For the actual transported schedule, each edge receives the target geometric slope of its image root. Lean stores the score as the lexicographic pair \((s(e),s(e))\), and merge-sorts the full finite edge set in weakly decreasing order. Reversal gives a complete duplicate-free chronological list in weakly increasing order.

Consider a suffix whose first edge is \(e:x\to y\), followed by \(q\). If an outgoing edge at \(x\) has the same score as \(e\), injectivity of geometric slope gives the same target reflection. The common source then gives the same target, so it is \(e\) itself. The no-repetition property excludes it from \(q\). An outgoing edge at \(y\) cannot have that score: its target reflection would be the current reflection again, sending the path back to \(x\).

For either endpoint, a remaining outgoing edge therefore occurs in \(q\) exactly when its score is strictly greater than that of \(e\). An earlier position has score at most the current one; a later position has score at least it, and the incident-root arguments make it strict. The schedule-suffix theorem gives this proof and identifies both remaining selections with the pullback of one ambient later-edge set.

This is the condition-set identity used when the global ambient image inclusion is applied to a step of the chronological schedule.

Edge factors and their degrees

Fix a source edge \(e:x\to y\) with root label \(\alpha_e\). Work in the edge coefficient ring \(A/(\alpha_e)\). Other root labels are reduced modulo \(\alpha_e\). The edge factor \(f_e\) is a product of these reduced linear factors with multiplicities determined by root-plane counts.

The nonzero reduced linear factors make this product nonzero in the quotient domain. The geometric plane counts determine its homogeneous degree: \[ \deg f_e=\frac{\ell(y)-\ell(x)-1}{2}. \] For a cover this is zero. For the long edge \(1\to w_0\) in the example it is one, so the correction has degree one.

The relative edge-degree theorem identifies each individual weight with the intrinsic dihedral formula on the least-representative coset: \[ 2k_P(za_0,a)=\ell_D(w)-\ell_D(z)-1 \] for the intrinsic edge \(z\to_Dw\) with ambient root \(a\). This follows by applying the plane inversion-change formula and replacing the two plane inversion counts by intrinsic lengths. It is the bridge from half-root counting to the exponent used in the local product comparison.

Localization and edge-image comparison use this factor to compare embedded edge images. Hilbert comparison uses its degree to normalize the comparison, and uses the path matrix to recognize the accumulated result as a target R-matrix.

The topic supplies the connections used later: order data can transport the edge schedule, and the schedule has both an algebraic path interpretation and controlled local correction factors.

Transfer order and the contribution of a two-edge path

For a directed edge \(e:x\to y\), the elementary transfer changes coordinate \(x\) to \(v_x+T v_y\), leaving the other coordinates unchanged. Dispatch on \([e_1,\ldots,e_m]\) means \(U_{e_1}\cdots U_{e_m}\), so the chronological processing list appears in reverse order in that expression.

For \(e:x\to y\), \(f:y\to z\), processing \(f\) before \(e\) gives \[ (a,b,c)\longmapsto(a,b+Tc,c) \longmapsto(a+Tb+T^2c,b+Tc,c). \] Thus the product associated with the path \(x\to y\to z\) contributes \(T^2c\) at \(x\). Reversing the processing order gives \((a+Tb,b+Tc,c)\). The ordered path expansion generalizes this calculation by choosing which edge contributions occur in the full product.

For real \(T\ge0\), each transfer preserves the coordinatewise order. For distinct endpoints it is injective, since the unchanged upper coordinate can be subtracted back out of the lower one. The transfer-properties theorem proves that both properties pass to finite products. Order preservation propagates the sheaf Hilbert bounds, and injectivity recovers exactness at earlier stages when the final comparison is an equality.

From position choices to directed paths

A directed path is an edge list with matching successive endpoints. Its path term multiplies by \(T\) at each edge and reads the vector at the final vertex. With delta initial data at \(y\), that term is \(T^m\) for an \(m\)-edge path from \(x\) to \(y\), and zero for a failed incidence condition.

Expanding each transfer into its identity and edge contribution chooses a sublist of positions. The path-sum expansion proves this by separating the head-position choices into omitted and retained cases. When the schedule lists every edge once, different position choices give different edge lists, so the sum becomes a finite sum over the selected path set. For a repeated schedule \([e,e]\), the two singleton position choices would both contribute the list \([e]\); this accounts for the no-repetition condition in the finite-set formula.

The sorted-path selection theorem converts the list ordering into a strict path condition. Successive path edges are composable, while equal reflection labels cannot be composable. Weak sorting therefore becomes strict along the path. Conversely, a strictly sorted path whose edges occur in a weakly sorted schedule is automatically a sublist of that schedule; the proof recovers it by matching or discarding the containing head.

The actual increasing Bruhat path polynomial

For a ratio order \(O\), score each reflection edge by the slope of its positive left root. The finite edge set below \(y\) consists exactly of edges with target at most \(y\). Its increasing sorted list defines the finite selected paths from \(x\) to \(y\), and \[ F_O(x,y;X)=\sum_q X^{|q|}. \] The Bruhat path-selection theorem proves that these are precisely the increasing directed paths. The endpoint-bound argument keeps every such path in the finite lower region. It also gives vanishing when \(x\not\le y\); strict increase of length leaves only the empty path when \(x=y\).

The path-sum expansion identifies this polynomial with the appropriate transfer-product entry. Its later identification with the Hecke-defined normalized R-polynomial uses the path recursion for an adapted chamber and the invariance of geometric entries under changes of functional.

Avoiding a simple label and splitting off its prefix

Fix \(s=s_i\). The edge rotation sends an edge \(x\to y\) to \(sx\to sy\) when its label differs from \(\alpha_i\), and fixes edges labelled by \(\alpha_i\). On lists avoiding the chosen label, all endpoints follow the same vertex map \(x\mapsto sx\). The simple-root path transport theorem preserves the directed-path condition and, for the two rotated orders, every strict label comparison.

The extremal-label theorem locates the possible chosen label. A least label can occur only first, and a descent at the initial vertex excludes that occurrence. A greatest label can occur only last, and an ascent at the final vertex excludes it.

Suppose \(sy\) is shorter than \(y\). In the rotated order the chosen label is greatest, and \(s(sy)=y\) is longer than \(sy\). Thus every increasing path ending at \(sy\) avoids that label. The rotated path-sum theorem uses edge rotation as a length-preserving bijection from the original avoiding paths to the full rotated path set.

When \(sx\) is longer than \(x\), the remaining original paths start with the unique edge \(x\to sx\). Deleting that edge gives an avoiding path from \(sx\) to \(y\). The simple-prefix decomposition proves both directions of this bijection, and prefixing one edge multiplies its weight by \(X\).

The path recursion and the normalized R-family

Combining the two bijections gives \[ F_O(x,y;X)= \begin{cases} F_{O'}(sx,sy;X),&\ell(sx)<\ell(x),\\ F_{O'}(sx,sy;X)+X F_{O'}(x,sy;X),&\ell(x)<\ell(sx). \end{cases} \] The path-polynomial recursion states the least-label, greatest-label and reflected-comparison hypotheses on \(O,O'\) explicitly.

For a chamber order at \(y\), the order-rotation theorem supplies those hypotheses and a new chamber order at the shorter element \(sy\). The adapted path identification uses strong induction on \(\ell(y)\), quantifying over the order and both endpoints. It substitutes the two smaller path polynomials at \((sx,sy)\) and \((x,sy)\), then applies the normalized R-recursion. At top \(1\), the diagonal and outside-order cases provide the base. Thus the increasing path polynomial of a chamber order equals the independently defined R-polynomial.

An A2 path calculation

Return to \(W=\langle s,t\rangle\) with \(sts=tst=w_0\). Put \(r=sts\), the third reflection, and take the reflection order \[ s<r<t. \] For decreasing labels, the paths from \(1\) to \(w_0\) are:

Path Left-reflection labels Weight
\(1\to w_0\) \(r\) \(T\)
\(1\to t\to ts\to w_0\) \(t,r,s\) \(T^3\)

For example, \(rt=ts\) and \(s(ts)=w_0\), which checks the labels in the second row. The other saturated chains have labels \(s,r,t\) along \(1,s,st,w_0\), labels \(s,t,s\) along \(1,s,ts,w_0\), and labels \(t,s,t\) along \(1,t,st,w_0\). None is decreasing. Every edge increases length by at least one, and every reflection reverses length parity. A path from length zero to length three therefore has one or three edges, so the two displayed paths exhaust the possibilities.

The decreasing path sum is \(T+T^3\). The geometric Dyer theorem identifies this path construction with the already-defined normalized R-family, giving \(\widetilde R_{1,w_0}(T)=T+T^3\).

The direct edge jumps three ranks but contributes \(T\), because path weight counts edges, not rank difference. This is a useful distinction when reading the general path formula.

The root-plane path involution

In an intrinsic two-generator interval, the interval-flip construction fixes its endpoints and switches the internal-rank vertices. The map preserves lengths, Bruhat order and reflection edges. Its edge involution therefore maps directed paths to directed paths with the same fixed endpoints and length.

For a composable pair through a middle vertex, the middle-vertex slope test expresses its label comparison as a simple-descent condition. The middle vertex has exactly one simple descent, and the internal flip exchanges it. Thus the comparison reverses after the edge map.

Sorted path transport turns this local comparison into a map from decreasing paths in a full schedule to paths in the reverse schedule. The same involution supplies the inverse map. The path-involution theorem reindexes the weighted finite sums, proving the plane-path reversal theorem. This full intrinsic-interval identity is the local block calculation used by the geometric order-independence argument.

From the adapted order to the fixed geometric order

The chamber existence theorem can keep height as denominator. Negating its numerator therefore converts increasing ratio scores into decreasing functional scores on normalized roots. This relates its path polynomial to a decreasing edge-product entry.

The geometric order-independence argument compares that entry with the fixed geometric functional. A change of functional is reduced to complete root-plane blocks; the intrinsic plane-path reversal above supplies the local identity after passage through the appropriate coset graph. The geometric path theorem develops this comparison and then applies the adapted path identification.

Definitions and proofs by construction

Geometric orders on roots

Dihedral systems inside a root plane

Dihedral interval combinatorics

Reconstruction of reflection edges

Schedules and their transport

Paths and transfer products

Inversion changes by root plane

The partner involution pairs the positive root lines exchanged by one reflection. The inversion-sign calculation gives contribution zero or two for each pair. Choosing the smaller slope records one root from each contributing pair, and grouping those roots by their planes defines the plane weights. Their sum measures the excess length of an edge beyond one.