diff b1f9bc4..56bc728 quadratic_operator_paving_v2.txt
diff --git a/quadratic_operator_paving_v2.txt b/quadratic_operator_paving_v2.txt
index a7a3cf7..8d677e6 100644
--- a/quadratic_operator_paving_v2.txt
+++ b/quadratic_operator_paving_v2.txt
@@ -33,53 +33,53 @@ hypersetup{pdftitle={Quadratic Strong-Operator Paving over Arbitrary Maximal Abe
 maketitle
 Abstract: 
 We prove the quadratic strong-operator paving conjecture of Popa and Vaes.
-For every $0<eps<1$, every self-adjoint element of a von Neumann algebra
+For every 0<eps<1, every self-adjoint element of a von Neumann algebra
 admits strong-operator paving over each maximal abelian subalgebra with at most
-$5times10^8eps^{-2}$ projections. The bound is uniform over representations
+5times10^8eps^{-2} projections. The bound is uniform over representations
 and requires no separability or conditional-expectation assumption.
 
 
 # Introduction
 
-Throughout, $M⊆mathcal B(H)$calB(H)$ is a von Neumann algebra on a complex
-Hilbert space, and $A⊆ M$ is a maximal abelian unital $*$-subalgebra,
-or masa. A finite partition $P=(p_1,...,p_r)$ in $A$ consists of
-mutually orthogonal projections with sum $1$. Its pinching is
+Throughout, M⊆mathcal B(H)calB(H) is a von Neumann algebra on a complex
+Hilbert space, and A⊆ M is a maximal abelian unital *-subalgebra,
+or masa. A finite partition P=(p_1,...,p_r) in A consists of
+mutually orthogonal projections with sum 1. Its pinching is
 [
- S_P(x)=sum_{i=1}^r p_i x p_i,qquad xin M.
+ S_P(x)=sum_{i=1}^r p_i x p_i, xin M.
 ]
 Paving asks how small this can be made after subtracting a diagonal element of
-$A$. Norm paving, which requires $S_P(x)-a$ itself to be small, fails for
+A. Norm paving, which requires S_P(x)-a itself to be small, fails for
 general masas cite[Theorem~3.3]{PV15}. The strong-operator formulation instead
 allows a compression whose complement is arbitrarily small in the strong
 operator topology.
 
 [Strong-operator paving]
-Let $x=x^*in M$, $eps>0$, and $rge1$. We say $x$ is $(eps,r)$
-so-pavable over $A$ if for every finite $Fsubset H$ and every $delta>0$
-there are a partition $P=(p_1,...,p_r)$ in $A$, an element $a=a^*in A$
-with $|a|le|x|$, and a projection $qin M$ such that
+Let x=x^*in M, eps>0, and rge1. We say x is (eps,r)
+so-pavable over A if for every finite Fsubset H and every delta>0
+there are a partition P=(p_1,...,p_r) in A, an element a=a^*in A
+with |a|le|x|, and a projection qin M such that
 [
  |(1-q)xi|<deltaquad(xiin F),
- qquad |q(S_P(x)-a)q|leeps|x|.
+  |q(S_P(x)-a)q|leeps|x|.
 ]
-The integer $r$ is fixed before $F$ and $delta$ are chosen.
+The integer r is fixed before F and delta are chosen.
 
 
 Thus the whole game is to produce, with a number of projections depending only
-on $eps$, a compression that captures all but an arbitrarily small piece of
+on eps, a compression that captures all but an arbitrarily small piece of
 the operator while making the visible piece small. The main result is that a
 quadratic number of projections always suffices.
 
 
-For every $0<eps<1$ set
+For every 0<eps<1 set
 [
  r_eps=leftlceil400000000eps^{-2}rightrceil
               +leftlceil4/epsrightrceil+1
           le500000000eps^{-2}.
 ]
-For every masa $A⊆ M⊆mathcal B(H)$calB(H)$ and every $x=x^*in M$,
-the element $x$ is $(eps,r_eps)$ so-pavable over $A$. The number of
+For every masa A⊆ M⊆mathcal B(H)calB(H) and every x=x^*in M,
+the element x is (eps,r_eps) so-pavable over A. The number of
 projections is thus fixed before the inclusion, the operator, and the strong
 neighborhood. No separability or conditional-expectation assumption is imposed.
 
@@ -88,7 +88,7 @@ Theorem~thm:main proves the quadratic so-paving conjecture of Popa and
 Vaes cite[Conjecture~2.8(2)]{PV15}. Their definition of so-paving uses
 arbitrary strong neighborhoods of zero; Definition~def:so-paving gives the
 equivalent finite-vector formulation cite[Definition~2.2]{PV15}. For
-self-adjoint $x$, allowing a general diagonal element is the same as allowing a
+self-adjoint x, allowing a general diagonal element is the same as allowing a
 self-adjoint one, since taking the real part preserves the norm bound and the
 compression estimate.
 
@@ -97,43 +97,43 @@ associated ultrapower.
 
 [Quadratic norm paving in Ocneanu ultrapowers]
 
-Let $A⊆ M$ be a masa with $A$ countably decomposable and the range of a
-normal conditional expectation from $M$. Let $omega$ be a free ultrafilter on
-$mathbb N$bbN$ and $A^omega⊆ M^omega$ the inclusion in the Ocneanu
-ultrapower. For every $0<eps<1$ and every $x=x^*in M^omega$ there are a
-partition $p_1,...,p_{r_eps}in A^omega$ and an $a=a^*in A^omega$ with
+Let A⊆ M be a masa with A countably decomposable and the range of a
+normal conditional expectation from M. Let omega be a free ultrafilter on
+mathbb NbbN and A^omega⊆ M^omega the inclusion in the Ocneanu
+ultrapower. For every 0<eps<1 and every x=x^*in M^omega there are a
+partition p_1,...,p_{r_eps}in A^omega and an a=a^*in A^omega with
 [
  |a|le|x|,
- qquad
+ 
  left|sum_{i=1}^{r_eps}p_i x p_i-aright|leeps|x|.
 ]
 
 
 Proof. 
-Theorem~thm:main makes every self-adjoint element of $M$
-$(eps,r_eps)$ so-pavable. The forward transfer in cite[Theorem~2.7(3)]{PV15}
-gives the stated norm paving in $M^omega$ with the same error and count.
+Theorem~thm:main makes every self-adjoint element of M
+(eps,r_eps) so-pavable. The forward transfer in cite[Theorem~2.7(3)]{PV15}
+gives the stated norm paving in M^omega with the same error and count.
 Taking the real part of the diagonal element preserves both estimates.
 
 
 ## History and optimality
 
 The classical Kadison--Singer paving problem concerns the diagonal masa in
-$mathcal B(calB(ell^2mathbb N)$,bbN)$, resolved by Marcus, Spielman, and Srivastava with
+mathcal B(calB(ell^2mathbb N),bbN), resolved by Marcus, Spielman, and Srivastava with
 mixed characteristic polynomials [cite]. For general masas norm paving is
-much more restrictive: when $M$ has separable predual it holds exactly when $M$
-is type~I and $A$ is the range of a normal conditional expectation
+much more restrictive: when M has separable predual it holds exactly when M
+is type~I and A is the range of a normal conditional expectation
 cite[Theorem~3.3]{PV15}. Strong-operator paving keeps a uniform count while
 allowing a compression approaching the identity strongly.
 
-Popa's $II_1$-factor approach gave paving for singular-masa
+Popa's II_1-factor approach gave paving for singular-masa
 ultraproducts [cite]. Popa and Vaes obtained so-paving bounds of order
-$eps^{-4}$ for arbitrary masas in type~I algebras with separable predual, for
+eps^{-4} for arbitrary masas in type~I algebras with separable predual, for
 Cartan masas in amenable algebras, and for the Cartan inclusions associated
 with essentially free ergodic probability-measure-preserving profinite actions
 cite[Proposition~3.2 and Theorem~4.1]{PV15}. For tracial ultraproducts of
 singular masas in finite algebras they obtained a common norm-paving partition
-of order $eps^{-2}$ for each finite family cite[Theorem~5.1]{PV15}, and later
+of order eps^{-2} for each finite family cite[Theorem~5.1]{PV15}, and later
 the optimal error for singular-masa ultraproducts cite[Theorem~4.1]{PV16}.
 They explicitly noted that the Cartan and singular arguments had to be united
 cite[Remark~2.9(ii)]{PV15}. On the finite-dimensional side, Ravichandran and
@@ -141,11 +141,11 @@ Srivastava obtained asymptotically optimal multipaving bounds from mixed
 determinantal polynomials [cite]; these are the finite input used here.
 
 A naive combination shows the quantitative difficulty. For an intermediate
-inclusion $A⊆ N⊆ M$ with a faithful normal tracial state, Popa
-and Vaes assume $L^2(Mominus N)$ has no nonzero $A$--$N$ subbimodule finitely
-generated on the right; their bound uses at most $400eps^{-2}$ times the
-so-paving count for $A⊆ N$ at error $eps/2$ cite[Remark~5.3]{PV15}.
-Inserting their quartic Cartan bound gives order $eps^{-6}$. The present
+inclusion A⊆ N⊆ M with a faithful normal tracial state, Popa
+and Vaes assume L^2(Mominus N) has no nonzero A--N subbimodule finitely
+generated on the right; their bound uses at most 400eps^{-2} times the
+so-paving count for A⊆ N at error eps/2 cite[Remark~5.3]{PV15}.
+Inserting their quartic Cartan bound gives order eps^{-6}. The present
 argument handles the normalizer algebra through a nonsingular measured
 equivalence relation and uses a single partition for both terms in the
 normalizer decomposition of one operator, which retains quadratic order. (The
@@ -153,54 +153,50 @@ finite-family obstruction in cite[Remark~5.2]{PV15} concerns the stronger
 requirement of a common partition for arbitrarily many separately prescribed
 operators.)
 
-The exponent two is already forced by the $L^2$ obstruction in a $II_1$
-factor from cite[Proposition~5.4]{PV15}. Fix $1le n<eps^{-2}$, choose
-$eps<eta<n^{-1/2}$, and use the centered self-adjoint unitary $x$ of that
-proposition at threshold $eta$. Write $tau$ for the normalized trace and
-$|b|_2=tau(b^*b)^{1/2}$. For any partition $P$ with $n$ pieces and any
-$ain A$, trace orthogonality gives
+The exponent two is already forced by the L^2 obstruction in a II_1
+factor from cite[Proposition~5.4]{PV15}. Fix 1le n<eps^{-2}, choose
+eps<eta<n^{-1/2}, and use the centered self-adjoint unitary x of that
+proposition at threshold eta. Write tau for the normalized trace and
+|b|_2=tau(b^*b)^{1/2}. For any partition P with n pieces and any
+ain A, trace orthogonality gives
 [
  |S_P(x)-a|_2^2=|S_P(x)|_2^2+|a|_2^2.
 ]
-If $|a|le1$ set $z=S_P(x)-a$, so $|z|le2$. Whenever $|qzq|leeps$, the
-two discarded corners give $|S_P(x)|_2le|z|_2leeps+4tau(1-q)^{1/2}$. In
+If |a|le1 set z=S_P(x)-a, so |z|le2. Whenever |qzq|leeps, the
+two discarded corners give |S_P(x)|_2le|z|_2leeps+4tau(1-q)^{1/2}. In
 the trace representation a strong neighborhood can force
-$tau(1-q)<((eta-eps)/4)^2$, which contradicts the lower bound
-$|S_P(x)|_2>eta$ valid for every such $P$. So the obstruction survives the
+tau(1-q)<((eta-eps)/4)^2, which contradicts the lower bound
+|S_P(x)|_2>eta valid for every such P. So the obstruction survives the
 varying diagonals and compressions allowed here, and the exponent in
 Theorem~thm:main is optimal (the constant is not optimized).
 
 Theorem~thm:main controls a compression of the pinching of the original
-$x$. A separate expectation-free approximation-paving theorem instead produces
-self-adjoint approximants $y$ close to $x$ strongly, with $|y|le3|x|$,
-and norm-paves them with error measured against $|y|$
+x. A separate expectation-free approximation-paving theorem instead produces
+self-adjoint approximants y close to x strongly, with |y|le3|x|,
+and norm-paves them with error measured against |y|
 cite[Theorem~1.1]{ApproxCompanion26}; that conclusion has a finite
-$eps$-dependent count but no quadratic rate, and the two proofs are
+eps-dependent count but no quadratic rate, and the two proofs are
 independent.
 
 ## Map of the proof
 
 We give a short map before the formal work; each step is a tool that is later
 used exactly once or twice.
-
-
-item Reduce to one faithful central state (Section~sec:reductions).
-The masa need not be the range of a normal conditional expectation. Split $M$
-into a corner where a faithful $A$-central state exists and a corner where no
-nonzero $A$-central functional exists. The first corner is where all the
+ Reduce to one faithful central state (Section~sec:reductions).
+The masa need not be the range of a normal conditional expectation. Split M
+into a corner where a faithful A-central state exists and a corner where no
+nonzero A-central functional exists. The first corner is where all the
 measured-space work happens; the second is handled by a symmetric Haagerup
-$L^2$ estimate and two randomizations with $O(eps^{-1})$ colors. Disjoint
+L^2 estimate and two randomizations with O(eps^{-1}) colors. Disjoint
 palettes on the corners make the pinching kill cross terms and the counts add.
-This is the only place $L^p$ theory enters.
-
-item Atomless kernels off the normalizer algebra
-(Section~sec:kernels). Let $N$ be generated by the groupoid normalizers
-of $A$; then $A⊆ N$ is Cartan and represented by a nonsingular measured
-equivalence relation. Decompose $x-E_Ax=(E_Nx-E_Ax)+(x-E_Nx)$. For the second
-term $yinker E_N$, the positive map $fmapsto E_A(y^*fy)$ has an atomless
+This is the only place L^p theory enters.
+ Atomless kernels off the normalizer algebra
+(Section~sec:kernels). Let N be generated by the groupoid normalizers
+of A; then A⊆ N is Cartan and represented by a nonsingular measured
+equivalence relation. Decompose x-E_Ax=(E_Nx-E_Ax)+(x-E_Nx). For the second
+term yinker E_N, the positive map fmapsto E_A(y^*fy) has an atomless
 kernel. This atomlessness is what later makes collision terms vanish.
-
-item Finite polynomial estimate (Section~sec:polynomials). The
+ Finite polynomial estimate (Section~sec:polynomials). The
 finite engine is the mixed determinantal bound of cite[Theorem~25]{RS17}. A
 partial coloring of a Hermitian matrix defines a real-rooted expected
 polynomial; the sum of its positive roots after a negative diagonal shift is a
@@ -208,33 +204,29 @@ cost. The upper mixed bound makes the all-unassigned cost zero; the
 two-sign lower bound makes excessive norm on a fully assigned block force
 positive cost; and a single-site assignment inequality bounds the cost increase
 of fixing one color. Rooted spectral measures spread the cost over vertices.
-
-item Measured graphs and the integral-change identity
+ Measured graphs and the integral-change identity
 (Section~sec:invariant). For an invariant measured graph, the total cost
 changes by the integral of single-site deletion increments, in an ordered
 sequence of comparison environments. A separate transform comparison bounds the
 effect of deleting far-away large weights on any fixed weight band, uniformly
 in the largest weight. This is what lets weight cutoffs be removed later.
-
-item Coloring theorem (Section~sec:coloring). On the relative
+ Coloring theorem (Section~sec:coloring). On the relative
 Bernoulli extension of a nonsingular relation, an invariant lift with an
 extra height coordinate makes mass transport available. Joint cutoff limits
 produce homogeneous costs, and random assignment times give a coloring of small
 total cost in which any prescribed ball is fully assigned with high probability,
 with uniform conditional color marginals. Completion of the remaining sites
 preserves the estimate on those balls.
-
-item Transfer (Section~sec:transfer). Put $M$ and the Bernoulli
-extension of $N$ in their amalgamated free product over $N$. Uniform marginals
-give a free-compression bound on $ker E_N$, while the same color projections
+ Transfer (Section~sec:transfer). Put M and the Bernoulli
+extension of N in their amalgamated free product over N. Uniform marginals
+give a free-compression bound on ker E_N, while the same color projections
 give the Cartan estimate. A finite-symbol evaluation plus a shortest-pair
 argument in cyclic words transfers each fixed mixed scalar moment from the model
-to actual projections of $A$; the atomless kernels kill the collision terms.
-
-item Completion (Section~sec:completion). Turn the local Cartan
+to actual projections of A; the atomless kernels kill the collision terms.
+ Completion (Section~sec:completion). Turn the local Cartan
 estimate into bounded approximants by clipping, run continuous calculus for the
 resulting varying sequences, and transfer one sufficiently high fixed moment. A
-spectral cut produces the projection $q$. All approximation parameters are
+spectral cut produces the projection q. All approximation parameters are
 chosen after the number of colors, and the corner assembly gives
 Theorem~thm:main.
 
@@ -242,15 +234,15 @@ Theorem~thm:main.
 ## Conventions
 
 
-We allow zero projections in a partition and write $1$ also for the identity of
+We allow zero projections in a partition and write 1 also for the identity of
 a corner regarded as an algebra. Except when restoring scale at the very end,
 operators to be paved are self-adjoint contractions. Maximal abelianness makes
-$A$ weakly closed with $A'cap M=A$: the weak closure is abelian, and any
+A weakly closed with A'cap M=A: the weak closure is abelian, and any
 self-adjoint relative-commutant element could otherwise be adjoined. Corners by
-projections in $A$ are again masas in the corresponding corners of $M$.
+projections in A are again masas in the corresponding corners of M.
 
-For a normal positive functional $varphi$ write
-$|z|_varphi=varphi(z^*z)^{1/2}$. For a faithful normal state, the
+For a normal positive functional varphi write
+|z|_varphi=varphi(z^*z)^{1/2}. For a faithful normal state, the
 centralizer is the fixed algebra of its modular group. Measurable statements
 are understood modulo null sets; in the presence of a countable nonsingular
 relation, a null exceptional set is enlarged to its null saturation.
@@ -258,47 +250,47 @@ relation, a null exceptional set is enlarged to its null saturation.
 # Reduction to a faithful central state
 
 We first remove the obstruction to using a normal conditional expectation: the
-original masa need not be the range of one. A normal positive functional $rho$
-on $M$ is $A$-central if $rho(bz)=rho(zb)$ for $bin A$, $zin M$. The
-idea is to split $M$ by the support of all such functionals.
+original masa need not be the range of one. A normal positive functional rho
+on M is A-central if rho(bz)=rho(zb) for bin A, zin M. The
+idea is to split M by the support of all such functionals.
 
-Let $pin A$ be the supremum of the supports of the $A$-central normal positive
-functionals; it lies in $A$ because conjugation by any unitary of $A$ preserves
+Let pin A be the supremum of the supports of the A-central normal positive
+functionals; it lies in A because conjugation by any unitary of A preserves
 such a functional and hence its support. The supports form a directed family
-(the support of a sum is the join), so $p$ is the join. The corner
-$eMe$ for $ele p$ carries a faithful normal state central over $eAe$, and on
-$(1-p)M(1-p)$ no nonzero central functional survives. Given finite vector tests
-we pick $ele p$ so that $p-e$ acts arbitrarily small on those vectors and
-discard the corner $p-e$. Because $p$ and $e$ need not be central, we use
+(the support of a sum is the join), so p is the join. The corner
+eMe for ele p carries a faithful normal state central over eAe, and on
+(1-p)M(1-p) no nonzero central functional survives. Given finite vector tests
+we pick ele p so that p-e acts arbitrarily small on those vectors and
+discard the corner p-e. Because p and e need not be central, we use
 disjoint palettes on the two retained corners so that a pinching removes all
 cross-corner terms and the projection counts add. This single assembly reduces
-Theorem~thm:main to the case of a faithful $A$-central state, proved in
+Theorem~thm:main to the case of a faithful A-central state, proved in
 Proposition~prop:expected. All steps keep a projection count independent of
 the prescribed strong neighborhood.
 
 ## The corner without central normal functionals
 
-For this corner we use Haagerup's $L^p$ spaces for an arbitrary von Neumann
+For this corner we use Haagerup's L^p spaces for an arbitrary von Neumann
 algebra, built from a normal semifinite faithful weight; see
-cite[Section~1]{Haa79}. A normal positive functional $varphi$ has a density
-$h_varphiin L^1(M)_+$ under the isometric identification $L^1(M)=M_*$, paired
-by $Tr$. The space $L^2(M)$ with its left and right multiplications and its
-positive cone is a standard form of $M$, and the cone vector of $varphi$ is
-$h_varphi^{1/2}$. We use H"older's inequality, cyclicity of the
-$L^p$--$L^{p'}$ pairing, and polar decomposition, whose partial isometry lies in
-$M$.
+cite[Section~1]{Haa79}. A normal positive functional varphi has a density
+h_varphiin L^1(M)_+ under the isometric identification L^1(M)=M_*, paired
+by Tr. The space L^2(M) with its left and right multiplications and its
+positive cone is a standard form of M, and the cone vector of varphi is
+h_varphi^{1/2}. We use H"older's inequality, cyclicity of the
+L^p--L^{p'} pairing, and polar decomposition, whose partial isometry lies in
+M.
 
 The positive-cone estimate below is Haagerup's standard-form inequality
 cite[Lemma~2.10]{Haa75}; see also cite[Lemma~4.1]{PS70} and
 cite[Theorem~4(8) and Section~8, Remark~2]{Ara74}. We include a proof so that
 no faithfulness of the two functionals is hidden. For normal positive
-$varphi,psi$,
+varphi,psi,
 
  |h_varphi^{1/2}-h_psi^{1/2}|_2^2le|varphi-psi|.
 
-Indeed, put $xi_1=h_varphi^{1/2}$, $xi_2=h_psi^{1/2}$, $D=xi_1-xi_2$, and
-let $sin M$ be the sign of $D$, so $|s|le1$ and $sD=Ds=|D|$. Writing
-$D=D_+-D_-$, cyclicity gives
+Indeed, put xi_1=h_varphi^{1/2}, xi_2=h_psi^{1/2}, D=xi_1-xi_2, and
+let sin M be the sign of D, so |s|le1 and sD=Ds=|D|. Writing
+D=D_+-D_-, cyclicity gives
 
  Tr(s(h_varphi-h_psi))
  &=Tr(s(Dxi_1+xi_2D))
@@ -307,315 +299,315 @@ $D=D_+-D_-$, cyclicity gives
  ge|D|_2^2.
 
 The last two pairings are nonnegative by self-duality of the positive cone, and
-the first expression is at most $|varphi-psi|$.
+the first expression is at most |varphi-psi|.
 
 
-Let $A⊆ M$ be a masa such that $M$ has no nonzero $A$-central normal
-positive functional. Let $x=x^*in M$ with $|x|le1$ and $0<eps<1$, and put
-$r_{rm p}=lceil4/epsrceil$. For every normal state $varphi$ on $M$ and
-every $rho>0$ there are a partition $p_1,...,p_{r_{rm p}}$ in $A$, an
-$a=a^*in A$ with $|a|le1$, and a projection $qin M$ such that
+Let A⊆ M be a masa such that M has no nonzero A-central normal
+positive functional. Let x=x^*in M with |x|le1 and 0<eps<1, and put
+r_{rm p}=lceil4/epsrceil. For every normal state varphi on M and
+every rho>0 there are a partition p_1,...,p_{r_{rm p}} in A, an
+a=a^*in A with |a|le1, and a projection qin M such that
 [
  varphi(1-q)<rho,
- qquad
+ 
  Bigl|q(sum_{i=1}^{r_{rm p}}p_i x p_i-a)qBigr|leeps.
 ]
 
 
 Proof. 
-Write $xi=h_varphi^{1/2}$ and let $T(z)=h_varphi^{1/4}zh_varphi^{1/4}colon
-Mto L^2(M)$. It is bounded and takes weak-star to weak convergence: testing
-against $zetain L^2(M)$ uses the $L^1$ element
-$h_varphi^{1/4}zeta^*h_varphi^{1/4}$. For a finite partition $P$ in $A$ the
-pinching $S_P=sum_{ein P}e(cdot)e$ acts on $L^2(M)$ as an orthogonal
+Write xi=h_varphi^{1/2} and let T(z)=h_varphi^{1/4}zh_varphi^{1/4}colon
+Mto L^2(M). It is bounded and takes weak-star to weak convergence: testing
+against zetain L^2(M) uses the L^1 element
+h_varphi^{1/4}zeta^*h_varphi^{1/4}. For a finite partition P in A the
+pinching S_P=sum_{ein P}e(cdot)e acts on L^2(M) as an orthogonal
 projection. These projections decrease under refinement to the projection onto
-$A$-central vectors. That projection is zero: a nonzero vector in its range
-would give a nonzero $A$-central normal positive functional (move a unitary of
-$A$ from left to right, where it commutes with the left action of $M$). Hence
-$|S_P(xi)|_2to0$ as $P$ refines.
+A-central vectors. That projection is zero: a nonzero vector in its range
+would give a nonzero A-central normal positive functional (move a unitary of
+A from left to right, where it commutes with the left action of M). Hence
+|S_P(xi)|_2to0 as P refines.
 
-For $zin M$, cyclicity and H"older give
+For zin M, cyclicity and H"older give
 
  sum_{e,fin P}|T(ezf)|_2^2
  =Tr(z^*S_P(xi)zS_P(xi))le|z|^2|S_P(xi)|_2^2.
 
-In particular $|T(eze)|_2^2le|exi e|_2^2$ when $|z|le1$.
-
-Fix $kappa>0$ and a coarse $P_0$ with $|S_{P_0}(xi)|_2<kappa$. As $P$
-refines $P_0$, $S_P(x)$ has a weak-star cluster point $bin A$ (it commutes
-with every partition of $A$). Since the weak and norm closures of a convex set
-in a Hilbert space agree, there are refinements $P^1,...,P^d$ of $P_0$ and
-probabilities $t_1,...,t_d$ with
-$|sum_j t_jT(S_{P^j}(x))-T(b)|_2<kappa$. Within each $ein P_0$ pick
-independently a refinement $P^j$ with this distribution; the resulting partition
-$R$ refines $P_0$. By independence and (eq:symmetric-block-sum) the
-variance of $T(S_R(x))$ is $sum_{ein P_0}|exi e|_2^2=|S_{P_0}(xi)|_2^2<
-kappa^2$, so one realization has $|T(S_R(x)-b)|_2<sqrt2kappa$.
-
-Now color the pieces of $R$ independently and uniformly with $r_{rm p}$ colors
-and let $S_{rm col}$ be the color pinching. Its mean is
-$mathbb EbbE S_{rm col}(x)=r_{rm p}^{-1}x+(1-r_{rm p}^{-1})S_R(x)$. Equal-color
+In particular |T(eze)|_2^2le|exi e|_2^2 when |z|le1.
+
+Fix kappa>0 and a coarse P_0 with |S_{P_0}(xi)|_2<kappa. As P
+refines P_0, S_P(x) has a weak-star cluster point bin A (it commutes
+with every partition of A). Since the weak and norm closures of a convex set
+in a Hilbert space agree, there are refinements P^1,...,P^d of P_0 and
+probabilities t_1,...,t_d with
+|sum_j t_jT(S_{P^j}(x))-T(b)|_2<kappa. Within each ein P_0 pick
+independently a refinement P^j with this distribution; the resulting partition
+R refines P_0. By independence and (eq:symmetric-block-sum) the
+variance of T(S_R(x)) is sum_{ein P_0}|exi e|_2^2=|S_{P_0}(xi)|_2^2<
+kappa^2, so one realization has |T(S_R(x)-b)|_2<sqrt2kappa.
+
+Now color the pieces of R independently and uniformly with r_{rm p} colors
+and let S_{rm col} be the color pinching. Its mean is
+mathbb EbbE S_{rm col}(x)=r_{rm p}^{-1}x+(1-r_{rm p}^{-1})S_R(x). Equal-color
 indicators for distinct pairs of pieces are uncorrelated even when pairs share a
-piece; with $|u+v|_2^2le2|u|_2^2+2|v|_2^2$ and
-(eq:symmetric-block-sum) the variance after applying $T$ is at most
-$2|S_R(xi)|_2^2<2kappa^2$. Thus some coloring makes
+piece; with |u+v|_2^2le2|u|_2^2+2|v|_2^2 and
+(eq:symmetric-block-sum) the variance after applying T is at most
+2|S_R(xi)|_2^2<2kappa^2. Thus some coloring makes
 [
  y=S_{rm col}(x)-(1-r_{rm p}^{-1})b-r_{rm p}^{-1}x
 ]
-satisfy $|T(y)|_2<2sqrt2kappa$, with $|y|le2$. Since $kappa$ is
+satisfy |T(y)|_2<2sqrt2kappa, with |y|le2. Since kappa is
 arbitrary the symmetric seminorm can be made as small as we like with the count
 already fixed.
 
 It remains to turn a small symmetric seminorm into a large compression. Set
-$t=|T(y)|_2$ and $psi(z)=varphi(y^*zy)$. Since small $t$ need not control
-$|y|_varphi$, we first find $q_0$ with both $varphi(1-q_0)$ and
-$psi(q_0)$ small. Polar decomposition gives $h_psi=yh_varphi y^*$ and
-$h_psi^{1/2}=yxi u^*$ for a partial isometry $uin M$, so
+t=|T(y)|_2 and psi(z)=varphi(y^*zy). Since small t need not control
+|y|_varphi, we first find q_0 with both varphi(1-q_0) and
+psi(q_0) small. Polar decomposition gives h_psi=yh_varphi y^* and
+h_psi^{1/2}=yxi u^* for a partial isometry uin M, so
 [
  0leTr(xi h_psi^{1/2})le|xi yxi|_1
  =|h_varphi^{1/4}T(y)h_varphi^{1/4}|_1le t,
 ]
-the last step using $|h_varphi^{1/4}|_4=1$. Let $q_0$ support the positive
-part in the Jordan decomposition of $varphi-psi$. Then
+the last step using |h_varphi^{1/4}|_4=1. Let q_0 support the positive
+part in the Jordan decomposition of varphi-psi. Then
 (eq:positive-cone-estimate) gives
 
  varphi(1-q_0)+psi(q_0)
  =tfrac12(varphi(1)+psi(1)-|varphi-psi|)le t,
 
 and
-$|q_0yq_0xi|_2le|q_0yxi|_2+|q_0y(1-q_0)xi|_2lepsi(q_0)^{1/2}+
-2varphi(1-q_0)^{1/2}le3sqrt t$.
-Inside $q_0Mq_0$ let $q$ be the spectral projection of $q_0yq_0$ for
-$[-eps/2,eps/2]$. The spectral Markov inequality gives
+|q_0yq_0xi|_2le|q_0yxi|_2+|q_0y(1-q_0)xi|_2lepsi(q_0)^{1/2}+
+2varphi(1-q_0)^{1/2}le3sqrt t.
+Inside q_0Mq_0 let q be the spectral projection of q_0yq_0 for
+[-eps/2,eps/2]. The spectral Markov inequality gives
 
  varphi(1-q)levarphi(1-q_0)+frac4{eps^2}varphi((q_0yq_0)^2)
  le(1+(36)/(eps^2))t.
 
-Taking $kappa$ small makes this less than $rho$. With
-$a=(1-r_{rm p}^{-1})b$ we have $|a|le1$ and
-$|q(S_{rm col}(x)-a)q|leeps/2+r_{rm p}^{-1}le3eps/4$.
+Taking kappa small makes this less than rho. With
+a=(1-r_{rm p}^{-1})b we have |a|le1 and
+|q(S_{rm col}(x)-a)q|leeps/2+r_{rm p}^{-1}le3eps/4.
 
 
 ## Assembling the corners
 
 
-Fix $0<eps<1$ and an integer $Rge1$. Suppose that for every masa $A⊆ M$
-admitting a faithful $A$-central normal state $varphi$, every self-adjoint
-contraction $xin M$, and every $rho>0$, there exist a partition of $1$ into
-$R$ projections of $A$, a self-adjoint $ain A$ of norm at most one, and a
-projection $qin M$ with $varphi(1-q)<rho$ and
+Fix 0<eps<1 and an integer Rge1. Suppose that for every masa A⊆ M
+admitting a faithful A-central normal state varphi, every self-adjoint
+contraction xin M, and every rho>0, there exist a partition of 1 into
+R projections of A, a self-adjoint ain A of norm at most one, and a
+projection qin M with varphi(1-q)<rho and
 [
  Bigl|q(sum_{i=1}^{R}p_i x p_i-a)qBigr|leeps.
 ]
 Then every masa in every represented von Neumann algebra has the strong-operator
-paving property of Theorem~thm:main at error $eps$, with
-$r=R+lceil4/epsrceil+1$. This number is chosen before the finite vector set
+paving property of Theorem~thm:main at error eps, with
+r=R+lceil4/epsrceil+1. This number is chosen before the finite vector set
 and the strong-neighborhood tolerance.
 
 
 Proof. 
-Let $p$ be the supremum of the supports of the $A$-central normal positive
-functionals on $M$; as noted it lies in $A$, and the supports form a directed
-family. The corner $(1-p)M(1-p)$ has no nonzero $A(1-p)$-central normal
-functional, since compression would extend one to $M$.
-
-Given finite $Fsubset H$ and $delta>0$, set
-$chi(z)=sum_{xiin F}< zxi,xi>$. By normality there is a
-support $ele p$ in the family with $chi(p-e)$ arbitrarily small. On the corner
-$eMe$ the corresponding functional restricts, after normalization, to a
-faithful normal state central on $Ae$. (Small $omega$-mass on projections gives
+Let p be the supremum of the supports of the A-central normal positive
+functionals on M; as noted it lies in A, and the supports form a directed
+family. The corner (1-p)M(1-p) has no nonzero A(1-p)-central normal
+functional, since compression would extend one to M.
+
+Given finite Fsubset H and delta>0, set
+chi(z)=sum_{xiin F}< zxi,xi>. By normality there is a
+support ele p in the family with chi(p-e) arbitrarily small. On the corner
+eMe the corresponding functional restricts, after normalization, to a
+faithful normal state central on Ae. (Small omega-mass on projections gives
 uniformly small mass for any fixed normal positive functional: a contrary
-sequence in the positive unit ball has a weak-star limit of zero $omega$-mass
+sequence in the positive unit ball has a weak-star limit of zero omega-mass
 but positive mass, contradicting faithfulness.) Apply the assumed result on
-$eMe$, using this observation to keep the lost $chi$-mass small. On
-$(1-p)M(1-p)$ apply Lemma~lem:no-central to the normalized restriction of
-$chi$ when it is nonzero, and take $q=0$ otherwise.
+eMe, using this observation to keep the lost chi-mass small. On
+(1-p)M(1-p) apply Lemma~lem:no-central to the normalized restriction of
+chi when it is nonzero, and take q=0 otherwise.
 
-Use disjoint palettes on the two corners and the projection $p-e$ as a last
+Use disjoint palettes on the two corners and the projection p-e as a last
 piece, taking corner sums of the two diagonal terms and the two large
-projections (zero on $p-e$). Each pinching piece is supported in one corner, so
-all off-corner terms of $x$ vanish, the compressed norm is at most $eps$, and
-the diagonal has norm at most one. We arrange $chi(1-q)<delta^2$, so
-$|(1-q)xi|^2lechi(1-q)<delta^2$ for $xiin F$. Zero projections pad when
-needed. The total count is $r$, independent of $F$ and $delta$. Scaling handles
-a general nonzero self-adjoint $x$; $x=0$ is immediate.
+projections (zero on p-e). Each pinching piece is supported in one corner, so
+all off-corner terms of x vanish, the compressed norm is at most eps, and
+the diagonal has norm at most one. We arrange chi(1-q)<delta^2, so
+|(1-q)xi|^2lechi(1-q)<delta^2 for xiin F. Zero projections pad when
+needed. The total count is r, independent of F and delta. Scaling handles
+a general nonzero self-adjoint x; x=0 is immediate.
 
 
 ## State norm and separable reduction
 
-Now fix a faithful $A$-central normal state $varphi$. Write
-$|z|_varphi=varphi(z^*z)^{1/2}$ and let
-$M_varphi={bin M:varphi(bz)=varphi(zb) ∀ z}$ be the centralizer,
-the fixed algebra of the modular group; in particular $A⊆ M_varphi$. We
+Now fix a faithful A-central normal state varphi. Write
+|z|_varphi=varphi(z^*z)^{1/2} and let
+M_varphi={bin M:varphi(bz)=varphi(zb) ∀ z} be the centralizer,
+the fixed algebra of the modular group; in particular A⊆ M_varphi. We
 first approximate by diagonal pinchings in the state norm, then build a
 separable inclusion containing the operator.
 
 
-Let $A⊆ M$ be a masa and $varphi$ a faithful $A$-central normal state.
-There is a faithful normal $varphi$-preserving conditional expectation
-$E_Acolon Mto A$. On bounded sets, convergence to zero in $|cdot|_varphi$
+Let A⊆ M be a masa and varphi a faithful A-central normal state.
+There is a faithful normal varphi-preserving conditional expectation
+E_Acolon Mto A. On bounded sets, convergence to zero in |cdot|_varphi
 is equivalent to strong convergence in the faithful GNS representation. For
-every $zin M$,
+every zin M,
 
  |S_P(z)-E_A(z)|_varphi→0
 
-as finite partitions $P$ of $A$ refine; the same holds along any increasing
-sequence of finite partitions generating $A$.
+as finite partitions P of A refine; the same holds along any increasing
+sequence of finite partitions generating A.
 
 
 Proof. 
-Takesaki's theorem [cite] applies because the modular group fixes $A$
-pointwise. Let $Ω$ be the GNS vector; it is separating, hence cyclic for
-$M'$. If a bounded net $z_j$ has $z_jΩto0$, then
-$z_jc'Ω=c'z_jΩto0$ for every $c'in M'$, so $z_jto0$ strongly; the
-converse is testing on $Ω$.
+Takesaki's theorem [cite] applies because the modular group fixes A
+pointwise. Let Ω be the GNS vector; it is separating, hence cyclic for
+M'. If a bounded net z_j has z_jΩto0, then
+z_jc'Ω=c'z_jΩto0 for every c'in M', so z_jto0 strongly; the
+converse is testing on Ω.
 
-Since the partition projections lie in the centralizer, $S_P$ induces an
+Since the partition projections lie in the centralizer, S_P induces an
 orthogonal projection on the GNS space; it is the state-preserving expectation
-onto the commutant in $M$ of the projections of $P$. These projections decrease
-under refinement. On $zΩ$ let the limit vector be $eta$. A bounded
-weak-star cluster point $b$ of $S_P(z)$ commutes with every partition of $A$,
-hence lies in $A$; its state vector is $eta$, and $E_AS_P(z)=E_A(z)$ forces
-$b=E_A(z)$. If an increasing sequence generates $A$, its limit commutes with
+onto the commutant in M of the projections of P. These projections decrease
+under refinement. On zΩ let the limit vector be eta. A bounded
+weak-star cluster point b of S_P(z) commutes with every partition of A,
+hence lies in A; its state vector is eta, and E_AS_P(z)=E_A(z) forces
+b=E_A(z). If an increasing sequence generates A, its limit commutes with
 the generators and the same argument applies.
 
 
 
-Let $A⊆ M$ be a masa with faithful $A$-central normal state $varphi$,
-and $x=x^*in M$. There are unital von Neumann subalgebras $A_0⊆ A$ and
-$M_0⊆ M$ with $xin M_0$, $M_0$ of separable predual, $A_0$ a masa in
-$M_0$, and $E_A(M_0)⊆ A_0$. The restricted $varphi$ is faithful,
-normal, and $A_0$-central. Thus a uniform faithful-state paving result in the
+Let A⊆ M be a masa with faithful A-central normal state varphi,
+and x=x^*in M. There are unital von Neumann subalgebras A_0⊆ A and
+M_0⊆ M with xin M_0, M_0 of separable predual, A_0 a masa in
+M_0, and E_A(M_0)⊆ A_0. The restricted varphi is faithful,
+normal, and A_0-central. Thus a uniform faithful-state paving result in the
 separable-predual case implies the general faithful-state case.
 
 
 Proof. 
-Start with $D_1=W^*(1,x)$. Inductively choose a countable
-$|cdot|_varphi$-dense subset $mathcal D_j$calD_j$ of the unit ball of $D_j$ (every
-countably generated $D_j$ has one, by Kaplansky density applied to its separable
+Start with D_1=W^*(1,x). Inductively choose a countable
+|cdot|_varphi-dense subset mathcal D_jcalD_j of the unit ball of D_j (every
+countably generated D_j has one, by Kaplansky density applied to its separable
 generator C*-algebra; its faithful normal GNS representation is separable, so
-$D_j$ has separable predual). Let $D_{j+1}$ be generated by $D_j$, all
-$E_A(z)$ for $zinmathcal D_j$,calD_j$, and the projections of finite partitions
-$P(z,k)subset A$ with $|S_{P(z,k)}(z)-E_A(z)|_varphi<1/k$ for
-$zinmathcal D_j$,calD_j$, $kge1$, which exist by Lemma~lem:pinching. Only
+D_j has separable predual). Let D_{j+1} be generated by D_j, all
+E_A(z) for zinmathcal D_j,calD_j, and the projections of finite partitions
+P(z,k)subset A with |S_{P(z,k)}(z)-E_A(z)|_varphi<1/k for
+zinmathcal D_j,calD_j, kge1, which exist by Lemma~lem:pinching. Only
 countably many generators are added.
 
-Set $M_0=W^*(∪_jD_j)$ and let $A_0$ be generated unitally by the adjoined
-elements of $A$. Bounded Kaplansky approximation, then approximation from the
-$mathcal D_j$,calD_j$, shows the union is state-norm dense in the unit ball of $M_0$.
-Since $E_A$ is state-norm contractive, every $E_A(z)$ for $zin M_0$ is a
-bounded state-norm limit from $A_0$, so $E_A(M_0)⊆ A_0$ by
+Set M_0=W^*(∪_jD_j) and let A_0 be generated unitally by the adjoined
+elements of A. Bounded Kaplansky approximation, then approximation from the
+mathcal D_j,calD_j, shows the union is state-norm dense in the unit ball of M_0.
+Since E_A is state-norm contractive, every E_A(z) for zin M_0 is a
+bounded state-norm limit from A_0, so E_A(M_0)⊆ A_0 by
 Lemma~lem:pinching and strong closedness.
 
-If a unit-ball $zin M_0$ commutes with $A_0$, each selected partition has
-$S_P(z)=z$, and for $z'inmathcal D_j$calD_j$ and $P=P(z',k)$,
-$|z-E_A(z)|_varphile2|z-z'|_varphi+|S_P(z')-E_A(z')|_varphi$, which
-can be made small; faithfulness gives $z=E_A(z)in A_0$. Thus $A_0$ is maximal
-abelian in $M_0$. Also $M_0$ is countably generated with separable predual by
-the same GNS argument. A paving in $M_0$ has the same operator-norm bound and
-lost $varphi$-mass in $M$, giving the estimate needed by
+If a unit-ball zin M_0 commutes with A_0, each selected partition has
+S_P(z)=z, and for z'inmathcal D_jcalD_j and P=P(z',k),
+|z-E_A(z)|_varphile2|z-z'|_varphi+|S_P(z')-E_A(z')|_varphi, which
+can be made small; faithfulness gives z=E_A(z)in A_0. Thus A_0 is maximal
+abelian in M_0. Also M_0 is countably generated with separable predual by
+the same GNS argument. A paving in M_0 has the same operator-norm bound and
+lost varphi-mass in M, giving the estimate needed by
 Proposition~prop:state-reduction.
 
 
 # The normalizing algebra and atomless kernels
 
-For the structural part we assume $M$ has separable predual, $A⊆ M$ is
-a masa, and $varphi$ is a faithful $A$-central normal state;
+For the structural part we assume M has separable predual, A⊆ M is
+a masa, and varphi is a faithful A-central normal state;
 Lemma~lem:separable-reduction permits these assumptions. We separate the
 algebra generated by normalizers from its orthogonal complement under a
 state-preserving expectation. The former has a measured-relation description;
-on the latter we prove the maps $fmapsto E_A(y^*fy)$ have atomless kernels,
+on the latter we prove the maps fmapsto E_A(y^*fy) have atomless kernels,
 which later makes fine-partition collision terms vanish.
 
 ## Normalizers and the measured relation
 
-A groupoid normalizer of $A$ is a partial isometry $vin M$ with
-$v^*v,vv^*in A$ and $vAv^*=Avv^*$, $v^*Av=Av^*v$. Let $N$ be the von Neumann
-algebra they generate. For such $v$, $sigma_t^varphi(v)$ induces the same
-corner isomorphism as $v$ (the modular group fixes $A$ pointwise), so
-$v^*sigma_t^varphi(v)in Av^*v$ is unitary. These unitaries form a strongly
+A groupoid normalizer of A is a partial isometry vin M with
+v^*v,vv^*in A and vAv^*=Avv^*, v^*Av=Av^*v. Let N be the von Neumann
+algebra they generate. For such v, sigma_t^varphi(v) induces the same
+corner isomorphism as v (the modular group fixes A pointwise), so
+v^*sigma_t^varphi(v)in Av^*v is unitary. These unitaries form a strongly
 continuous one-parameter group (the cocycle identity becomes the group identity
-because the modular group fixes $Av^*v$), hence there is a positive nonsingular
-$h_v$ affiliated with $Av^*v$ with
+because the modular group fixes Av^*v), hence there is a positive nonsingular
+h_v affiliated with Av^*v with
 
- sigma_t^varphi(v)=v h_v^{it}qquad(tinmathbb R).bbR).
+ sigma_t^varphi(v)=v h_v^{it}(tinmathbb R).bbR).
 
-So $N$ is $sigma^varphi$-invariant, and Takesaki's theorem [cite] gives
-a faithful normal $varphi$-preserving conditional expectation
-$E_Ncolon Mto N$; write $M^circ=ker E_N$. The cuts
-$v1_{[-k,k]}(log h_v)$ are entire analytic for $sigma^varphi$
-((eq:normalizer-modular) extends to $v h_v^{iz}1_{[-k,k]}(log h_v)$ for
-$zinmathbb C$)bbC$) and converge strongly-* to $v$.
+So N is sigma^varphi-invariant, and Takesaki's theorem [cite] gives
+a faithful normal varphi-preserving conditional expectation
+E_Ncolon Mto N; write M^circ=ker E_N. The cuts
+v1_{[-k,k]}(log h_v) are entire analytic for sigma^varphi
+((eq:normalizer-modular) extends to v h_v^{iz}1_{[-k,k]}(log h_v) for
+zinmathbb C)bbC) and converge strongly-* to v.
 
-Identify $A=L^∞(X,mu)$ on a standard probability space with $varphi|_A$
+Identify A=L^∞(X,mu) on a standard probability space with varphi|_A
 integration. A groupoid normalizer induces a measure-class isomorphism between
-measurable subsets of $X$; on its fixed points it lies in $A$. Partition its
+measurable subsets of X; on its fixed points it lies in A. Partition its
 remaining domain into countably many measurable pieces each disjoint from its
 image (use a countable separating family, then disjointize). On one such piece
-the cut normalizer $v'$ has orthogonal initial and final projections, and
-$v'+{v'}^*+(1-{v'}^*v'-v'{v'}^*)$ is a unitary normalizer; recovering $v'$ and
-summing strongly shows full unitary normalizers also generate $N$. Thus
-$A⊆ N$ is Cartan: $A$ is maximal abelian, regular, and the range of the
-faithful normal expectation $E_A|_N$.
+the cut normalizer v' has orthogonal initial and final projections, and
+v'+{v'}^*+(1-{v'}^*v'-v'{v'}^*) is a unitary normalizer; recovering v' and
+summing strongly shows full unitary normalizers also generate N. Thus
+A⊆ N is Cartan: A is maximal abelian, regular, and the range of the
+faithful normal expectation E_A|_N.
 
 The Feldman--Moore theorem cite[Theorem~1]{FM77} realizes this pair as the
-algebra of a countable nonsingular measured equivalence relation $mathcal R$calR$ on
-$(X,mu)$, possibly twisted by a scalar $2$-cocycle; separability holds in the
+algebra of a countable nonsingular measured equivalence relation mathcal RcalR on
+(X,mu), possibly twisted by a scalar 2-cocycle; separability holds in the
 faithful GNS representation. We use the right counting representation
 
  L^2(mathcal R,calR,nu_{rm r})
  =int_X^ ⊕ ell^2([v]_{mathcal R}calR),dmu(v),
- qquad
+ 
  int f,dnu_{rm r}
  =int_Xsum_{usim v}f(u,v),dmu(v).
 
-Here $A$ acts diagonally by the left site. With a normalized cocycle $s$, a
-kernel $a$ acts in the fiber rooted at $v$ by the matrix
-$(a(u,w)s(u,w,v))_{u,win[v]_{mathcal R}calR}$. The identity
-$s(u,w,v)s(u,v,v')=s(u,w,v')s(w,v,v')$ shows that rerooting conjugates these
+Here A acts diagonally by the left site. With a normalized cocycle s, a
+kernel a acts in the fiber rooted at v by the matrix
+(a(u,w)s(u,w,v))_{u,win[v]_{mathcal R}calR}. The identity
+s(u,w,v)s(u,v,v')=s(u,w,v')s(w,v,v') shows that rerooting conjugates these
 matrices by a diagonal unitary, so their norms and principal matrix data are
 root-independent. Finite sums of bounded base functions times lifted kernel
-normalizers form a weakly dense $*$-algebra $N_{rm alg}$
-cite[Section~2]{FM77}, on which the canonical expectation to $A$ is the
-diagonal; it agrees with $E_A|_N$ (same value on the fixed part of a normalizer,
+normalizers form a weakly dense *-algebra N_{rm alg}
+cite[Section~2]{FM77}, on which the canonical expectation to A is the
+diagonal; it agrees with E_A|_N (same value on the fixed part of a normalizer,
 both zero on pieces with disjoint supports, extended by normality). Hence
-$varphi|_N$ is diagonal integration, represented in
-(eq:right-counting-space) by the unit vectors $delta_v$. All data are
+varphi|_N is diagonal integration, represented in
+(eq:right-counting-space) by the unit vectors delta_v. All data are
 taken modulo null sets, and countably many null exceptions are discarded with
 their null saturations.
 
 ## A type I observation
 
 The kernel argument produces an intertwiner whose initial projection is not yet
-known to be in $A$; the next observation turns it into a groupoid normalizer.
+known to be in A; the next observation turns it into a groupoid normalizer.
 Its state hypothesis excludes the diffuse masas that can occur in a type I
 algebra without a normal conditional expectation (compare the type I criterion
 of cite[Theorem~4.1]{AS12}).
 
 
-Let $D$ be a type I von Neumann algebra with separable predual, $C⊆ D$ a
-masa, and suppose $D$ admits a faithful $C$-central normal state. Then the
-groupoid normalizers of $C$ generate $D$, and every abelian projection of $D$
-is Murray--von Neumann equivalent in $D$ to a projection of $C$.
+Let D be a type I von Neumann algebra with separable predual, C⊆ D a
+masa, and suppose D admits a faithful C-central normal state. Then the
+groupoid normalizers of C generate D, and every abelian projection of D
+is Murray--von Neumann equivalent in D to a projection of C.
 
 
 Proof. 
 Use the direct integral decomposition of a type I algebra into full operator
 algebras over its center cite[Sections~6.6 and 14.1--14.2]{KR86}. Integrating
 fiber traces against a central probability measure gives a faithful normal
-semifinite trace $tau$. The state has a positive trace density $h$ of full
-support with $tau(h)=1$ (the semifinite predual identification, e.g.
-cite[Proposition~1.5 and Theorem~2.1]{Haa79}). Unitaries of $C$ preserve the
-state and hence $h$, so $h$ commutes with $C$ and its spectral projections lie
-in $C$. The projections $f_n=1_{[1/n,∞)}(h)$ exhaust $1$ with
-$tau(f_n)le n$; successive differences give a countable partition of $1$ in
-$C$ into trace-finite projections, each of finite rank a.e. in the type I
+semifinite trace tau. The state has a positive trace density h of full
+support with tau(h)=1 (the semifinite predual identification, e.g.
+cite[Proposition~1.5 and Theorem~2.1]{Haa79}). Unitaries of C preserve the
+state and hence h, so h commutes with C and its spectral projections lie
+in C. The projections f_n=1_{[1/n,∞)}(h) exhaust 1 with
+tau(f_n)le n; successive differences give a countable partition of 1 in
+C into trace-finite projections, each of finite rank a.e. in the type I
 fibers. Split further by central projections according to rank: each corner is
-a measurable field of fixed finite matrix size with compressed $C$ a masa.
+a measurable field of fixed finite matrix size with compressed C a masa.
 
 Diagonalize measurably in these finite corners. Trivialize the finite-rank
 field and pick countably many commuting self-adjoint matrix fields generating
-the compressed $C$. Fiberwise the algebras generated by the first $j$ fields
+the compressed C. Fiberwise the algebras generated by the first j fields
 eventually stabilize in dimension (nondecreasing bounded integers); the first
 index attaining the limit is measurable (finite dimensions are ranks of finite
 polynomial lists, the limit a countable supremum), so split the center by it.
@@ -623,118 +615,118 @@ On each piece, successive eigenspace decompositions measurably diagonalize the
 finite commuting list; ordered eigenvalues give measurable spectral
 projections and Gram--Schmidt gives measurable orthonormal bases. The resulting
 rank-one fields commute with all generators, hence lie in the masa. We obtain
-rank-one fields $e_jin C$ (zero off their central domains) with $sum_je_j=1$.
+rank-one fields e_jin C (zero off their central domains) with sum_je_j=1.
 
-On $e_jDe_j$ the algebra $Ce_j$ is the center scalars over that domain (the
-center of $D$ lies in $C$), so measurable matrix units between $e_j,e_k$ on a
-common central domain partially normalize $C$. These units and center
-coefficients generate $D$: compress a bounded field by finite sums of the
-$e_j$, express matrix entries in the units, and take a strong limit.
+On e_jDe_j the algebra Ce_j is the center scalars over that domain (the
+center of D lies in C), so measurable matrix units between e_j,e_k on a
+common central domain partially normalize C. These units and center
+coefficients generate D: compress a bounded field by finite sums of the
+e_j, express matrix entries in the units, and take a strong limit.
 
-An abelian projection $ein D$ has rank one a.e. on its central support $z$.
-Choose fiberwise on $z$ the first nonzero $e_j$; this is $e'in C$ with the same
+An abelian projection ein D has rank one a.e. on its central support z.
+Choose fiberwise on z the first nonzero e_j; this is e'in C with the same
 central support and rank. Measurable unit vectors spanning the ranges give a
-partial isometry $win D$ with $ww^*=e$, $w^*w=e'$. The case $e=0$ is trivial.
+partial isometry win D with ww^*=e, w^*w=e'. The case e=0 is trivial.
 
 
 ## Atomless kernels off the normalizing algebra
 
-A positive kernel on $X$ is a measurable family of finite positive Borel
-measures $K(v,du)$. It represents a positive normal map $Kcolon Ato A$ if
-$(Kf)(v)=int f(u)K(v,du)$ for $fin L^∞(X,mu)$, where for each fixed
-$mu$-null Borel $S$ one has $K(v,S)=0$ for a.e. $v$. This makes the formula
+A positive kernel on X is a measurable family of finite positive Borel
+measures K(v,du). It represents a positive normal map Kcolon Ato A if
+(Kf)(v)=int f(u)K(v,du) for fin L^∞(X,mu), where for each fixed
+mu-null Borel S one has K(v,S)=0 for a.e. v. This makes the formula
 well defined on classes; it does not assert absolute continuity of rows.
 
 
-In the setting of this section, for every $yin M^circ$ the map
-$K_y(f)=E_A(y^*fy)$ has a positive kernel of mass at most $|y|^2$ that is
-atomless for a.e. output $v$. For $kge1$ and $y_1,...,y_kin M^circ$ the
-composition $K_{y_1}circ...circ K_{y_k}$ has a positive kernel of mass at
-most $prod_j|y_j|^2$ that is atomless a.e. If $p_{rm at}in A$ is the
-projection of the atomic part of $A$, then $p_{rm at}y=yp_{rm at}=0$ for
-$yin M^circ$.
+In the setting of this section, for every yin M^circ the map
+K_y(f)=E_A(y^*fy) has a positive kernel of mass at most |y|^2 that is
+atomless for a.e. output v. For kge1 and y_1,...,y_kin M^circ the
+composition K_{y_1}circ...circ K_{y_k} has a positive kernel of mass at
+most prod_j|y_j|^2 that is atomless a.e. If p_{rm at}in A is the
+projection of the atomic part of A, then p_{rm at}y=yp_{rm at}=0 for
+yin M^circ.
 
 
 Proof. 
-Let $Ω$ be the standard vector for $varphi$. The commuting left and right
-actions of $A$ have a joint spectral measure $lambda$ on $Xtimes X$ on
-$yΩ$. Because $A⊆ M_varphi$, the right action of $f$ gives
-$yfΩ$, so
+Let Ω be the standard vector for varphi. The commuting left and right
+actions of A have a joint spectral measure lambda on Xtimes X on
+yΩ. Because A⊆ M_varphi, the right action of f gives
+yfΩ, so
 
  lambda(Stimes T)=varphi(y^*1_Sy1_T)=int_T E_A(y^*1_Sy),dmu
 
-for Borel $S,T$. Disintegrating in the right coordinate gives a kernel for
-$K_y$ of mass $E_A(y^*y)(v)le|y|^2$; choose a Borel version, zeroing rows on
+for Borel S,T. Disintegrating in the right coordinate gives a kernel for
+K_y of mass E_A(y^*y)(v)le|y|^2; choose a Borel version, zeroing rows on
 a null exceptional set so the mass bound holds for every row. For fixed null
-$S$, (eq:joint-kernel-measure) gives $K_y(v,S)=0$ a.e. No product-measure
+S, (eq:joint-kernel-measure) gives K_y(v,S)=0 a.e. No product-measure
 absolute continuity is needed.
 
 Suppose the kernel is not a.e. atomless. The map
-$(v,u)mapsto K_y(v,{u})$ is measurable (a decreasing limit of masses of pieces
-separating $u$). For some $k$, the set where it is at least $1/k$ has finite
+(v,u)mapsto K_y(v,{u}) is measurable (a decreasing limit of masses of pieces
+separating u). For some k, the set where it is at least 1/k has finite
 sections and positive-measure projection, so the Lusin--Novikov theorem
-cite[Theorem~18.10]{Kec95} yields a measurable selection $scolon Tto X$ on a
-positive-measure Borel $T$ with
+cite[Theorem~18.10]{Kec95} yields a measurable selection scolon Tto X on a
+positive-measure Borel T with
 
  K_y(v,{s(v)})ge1/kqquad(vin T).
 
-For null Borel $S$ this gives $mu({vin T:s(v)in S})le
-kint_TK_y(v,S),dmu=0$. Hence $theta(f)=1_T(fcirc s)$ is a well-defined
-unital $*$-homomorphism $Ato A1_T$; it is normal, since the pushforward under
-$s$ of a finite measure with $L^1$ density on $T$ is absolutely continuous with
+For null Borel S this gives mu({vin T:s(v)in S})le
+kint_TK_y(v,S),dmu=0. Hence theta(f)=1_T(fcirc s) is a well-defined
+unital *-homomorphism Ato A1_T; it is normal, since the pushforward under
+s of a finite measure with L^1 density on T is absolutely continuous with
 density giving the preadjoint.
 
-Take increasing finite Borel partitions $P_j$ generating $A$ and separating
+Take increasing finite Borel partitions P_j generating A and separating
 points, and put
 
  b_j=sum_{ein P_j}e,y,theta(e).
 
 The left and right support projections are orthogonal families, so
-$|b_j|le|y|$, and $E_N(b_j)=0$ by $N$-bimodularity. Their state vectors are
-the decreasing joint cuts of $yΩ$ to the sets where the left coordinate and
-$s$ of the right coordinate lie in the same piece of $P_j$, with the right
-coordinate in $T$; the limit $eta$ has
-$|eta|^2=int_TK_y(v,{s(v)}),dmu>0$. A weak-star convergent subnet of the
-bounded $b_j$ has limit $bin M^circ$ with $bΩ=eta$ (test against
-$zΩ$, $zin M$). So $bne0$, $b=b1_T$, and
+|b_j|le|y|, and E_N(b_j)=0 by N-bimodularity. Their state vectors are
+the decreasing joint cuts of yΩ to the sets where the left coordinate and
+s of the right coordinate lie in the same piece of P_j, with the right
+coordinate in T; the limit eta has
+|eta|^2=int_TK_y(v,{s(v)}),dmu>0. A weak-star convergent subnet of the
+bounded b_j has limit bin M^circ with bΩ=eta (test against
+zΩ, zin M). So bne0, b=b1_T, and
 
- fb=btheta(f)qquad(fin A),
+ fb=btheta(f)(fin A),
 
 first for step functions by refinement in (eq:graph-intertwiner-cuts),
-then for all $f$ by bounded strong approximation and normality of $theta$.
-
-We show (eq:graph-intertwiner) forces $bin N$. In the polar
-decomposition $b=u|b|$, the relations give $|b^*|in A$,
-$fu=utheta(f)$ for $fin A$, and $e:=u^*uin D:=theta(A)'cap1_TM1_T$ (since
-$bb^*$ commutes with $A$ and $b^*b$ with $theta(A)$). For $din eDe$, the
-element $udu^*$ commutes with $A$ and lies in $M$, hence in $A$, so $e$ is
-abelian in $D$. Let $z$ be its central support in $D$. By the
+then for all f by bounded strong approximation and normality of theta.
+
+We show (eq:graph-intertwiner) forces bin N. In the polar
+decomposition b=u|b|, the relations give |b^*|in A,
+fu=utheta(f) for fin A, and e:=u^*uin D:=theta(A)'cap1_TM1_T (since
+bb^* commutes with A and b^*b with theta(A)). For din eDe, the
+element udu^* commutes with A and lies in M, hence in A, so e is
+abelian in D. Let z be its central support in D. By the
 abelian-projection characterization of type I algebras
-cite[Sections~6.4--6.6]{KR86}, $Dz$ is type I (every nonzero central part
-meets $e$), and since $A1_T$ is a masa in $D$, its center, in particular $z$,
-is in $A1_T$.
-
-The restricted state on $Dz$ is faithful, normal, and central on the masa $Az$,
-so by Lemma~lem:type-i-normalizers $Dz$ is generated by groupoid
-normalizers of $Az$. These partially normalize $A$ (supports in $Az$, and $A$
-restricts onto $Az$), so $Dz⊆ N$. The same lemma gives $win Dz$,
-$e'in Az$ with $ww^*=e$, $w^*w=e'$. Set $v=uw$. Its range projection
-$uu^*=supp|b^*|$ and initial projection $e'$ lie in $A$, and since $w$ commutes
-with $theta(A)$, $v^*Av=w^*u^*Auw=theta(A)e'⊆ Ae'$. The left side is a
-masa in $e'Me'$ (conjugating the masa on the range corner), so equality holds
-and $v$ is a groupoid normalizer. Hence $u=vw^*in N$ and $b=|b^*|uin N$,
-contradicting $0ne binker E_N$. The kernel is atomless a.e.
-
-For compositions, let $K_1,K_2$ be two such kernels and fix a null $S_0$
-outside which every row of $K_2$ is atomless. For a.e. $v$,
-$K_1(v,S_0)=0$, and for such $v$ the measure
-$(K_1circ K_2)(v,E)=int K_2(u,E)K_1(v,du)$ vanishes on every singleton $E$
-simultaneously, because the integrand rows are atomless off $S_0$. The mass
+cite[Sections~6.4--6.6]{KR86}, Dz is type I (every nonzero central part
+meets e), and since A1_T is a masa in D, its center, in particular z,
+is in A1_T.
+
+The restricted state on Dz is faithful, normal, and central on the masa Az,
+so by Lemma~lem:type-i-normalizers Dz is generated by groupoid
+normalizers of Az. These partially normalize A (supports in Az, and A
+restricts onto Az), so Dz⊆ N. The same lemma gives win Dz,
+e'in Az with ww^*=e, w^*w=e'. Set v=uw. Its range projection
+uu^*=supp|b^*| and initial projection e' lie in A, and since w commutes
+with theta(A), v^*Av=w^*u^*Auw=theta(A)e'⊆ Ae'. The left side is a
+masa in e'Me' (conjugating the masa on the range corner), so equality holds
+and v is a groupoid normalizer. Hence u=vw^*in N and b=|b^*|uin N,
+contradicting 0ne binker E_N. The kernel is atomless a.e.
+
+For compositions, let K_1,K_2 be two such kernels and fix a null S_0
+outside which every row of K_2 is atomless. For a.e. v,
+K_1(v,S_0)=0, and for such v the measure
+(K_1circ K_2)(v,E)=int K_2(u,E)K_1(v,du) vanishes on every singleton E
+simultaneously, because the integrand rows are atomless off S_0. The mass
 bound and fixed-null-set property pass to the composition; iterate.
 
 Finally the atomic part of a standard probability space is a countable set of
-positive masses. Atomlessness gives $E_A(y^*p_{rm at}y)=0$, and faithfulness
-gives $p_{rm at}y=0$; applying the same to $y^*$ gives $yp_{rm at}=0$.
+positive masses. Atomlessness gives E_A(y^*p_{rm at}y)=0, and faithfulness
+gives p_{rm at}y=0; applying the same to y^* gives yp_{rm at}=0.
 
 
 # Finite polynomials and the cost of assigning a color
@@ -749,73 +741,73 @@ bounds proved here let this be used on measured graphs.
 
 ## Finite data and mixed determinantal bounds
 
-Fix a finite set $I$, a zero-diagonal Hermitian $T=(T_{uv})_{u,vin I}$ with
-$|T|le1$, an integer $rge1$, and $t>0$. Throughout the coloring argument
+Fix a finite set I, a zero-diagonal Hermitian T=(T_{uv})_{u,vin I} with
+|T|le1, an integer rge1, and t>0. Throughout the coloring argument
 
  beta=80.
 
-A partial coloring is $c:Ito{0,1,...,r}$, with $0$ meaning unassigned.
-Give each site a weight $W_v>0$ and put
+A partial coloring is c:Ito{0,1,...,r}, with 0 meaning unassigned.
+Give each site a weight W_v>0 and put
 [
  tau_v=t,&c(v)=0, beta t,&c(v)>0,
- qquad a_v=tau_vW_v.
+  a_v=tau_vW_v.
 ]
-At site $v$ the allowed color--sign pairs are
+At site v the allowed color--sign pairs are
 [
  mathcal A_v=calA_v=
   {1,...,r}times{+1,-1},&c(v)=0,
   {c(v)}times{+1,-1},&c(v)>0.
  
 ]
-Choose $omega_v$ uniformly from $mathcal A_v$,calA_v$, independently over sites, and
-write $k_v=|mathcal A_v|$.calA_v|$. For $j=(i,sigma)$ set
-$U_j={v:jinmathcal A_vcalA_v}$ and
-$B_j=sigma D_WTD_W-diag(a_v)_{vin I}$ with $D_W=diag(√(W_v))$. For a
-realization $omega$ let $I_j(omega)={v:omega_v=j}$ and let $B_omega$ be the
-direct sum of the principal matrices $B_j|_{I_j(omega)}$ placed on these
-subsets (zero between subsets). With $Z=diag(z_v)_{vin I}$ define
+Choose omega_v uniformly from mathcal A_v,calA_v, independently over sites, and
+write k_v=|mathcal A_v|.calA_v|. For j=(i,sigma) set
+U_j={v:jinmathcal A_vcalA_v} and
+B_j=sigma D_WTD_W-diag(a_v)_{vin I} with D_W=diag(√(W_v)). For a
+realization omega let I_j(omega)={v:omega_v=j} and let B_omega be the
+direct sum of the principal matrices B_j|_{I_j(omega)} placed on these
+subsets (zero between subsets). With Z=diag(z_v)_{vin I} define
 
- P_I(Z)=mathbb E_bbE_omegadet(Z-B_omega),qquad P_I(z)=P_I(z,...,z).
+ P_I(Z)=mathbb E_bbE_omegadet(Z-B_omega), P_I(z)=P_I(z,...,z).
 
-Write $P=P_I$ when $I$ is fixed; restriction is always principal with the same
-laws at remaining sites. Thus $P_{-v}=P_{I∖{v}}$ and
-$P_∅=1$.
+Write P=P_I when I is fixed; restriction is always principal with the same
+laws at remaining sites. Thus P_{-v}=P_{I∖{v}} and
+P_∅=1.
 
 [Mixed determinantal bounds]
-For Hermitian $A_1,...,A_k$ on a common nonempty finite set,
+For Hermitian A_1,...,A_k on a common nonempty finite set,
 [
  chi[A_1,...,A_k](z)=mathbb EbbEprod_{j=1}^kdet(zI-A_j|_{S_j}),
 ]
-one uniform assignment per site, empty determinant $1$, is real-rooted. If
-$kge2$ and the $A_j$ are zero-diagonal contractions, then
+one uniform assignment per site, empty determinant 1, is real-rooted. If
+kge2 and the A_j are zero-diagonal contractions, then
 
  lambda_{max}chi[A_1,...,A_k]<(3sqrt2)/(sqrt k).
 
-For any zero-diagonal Hermitian $L$ (no norm restriction),
+For any zero-diagonal Hermitian L (no norm restriction),
 
  lambda_{max}chi[L,-L]ge(|L|)/(2).
 
 Consequently, with all sites unassigned the unit-weight unshifted polynomial has
-largest root $<3/sqrt r$; with all sites assigned it equals
-$prod_{i=1}^rchi[T|_{c^{-1}(i)},-T|_{c^{-1}(i)}]$ (empty factors $1$).
+largest root <3/sqrt r; with all sites assigned it equals
+prod_{i=1}^rchi[T|_{c^{-1}(i)},-T|_{c^{-1}(i)}] (empty factors 1).
 
 
 Proof. 
 The normalization matches cite[Definition~5]{RS17}; real-rootedness is part of
 the mixed determinantal theory and also follows from
 Lemma~lem:polynomial-quotients. Equation~(eq:poly-mixed-upper) is
-cite[Theorem~25]{RS17} (zero-diagonal Hermitian contractions, $kge2$). By
+cite[Theorem~25]{RS17} (zero-diagonal Hermitian contractions, kge2). By
 cite[Proposition~16 and Theorem~15]{RS17}, for zero-diagonal Hermitian matrices
 replacing one member by zero cannot increase the largest root; expansion in
-principal minors gives $chi[L,0](z)=det(zI-L/2)$ (a minor on $S$ occurs with
-probability $2^{-|S|}$). Keeping either $L$ or $-L$ and using monotonicity proves
-(eq:poly-mixed-lower). All-unassigned uses $r$ copies each of $T,-T$, so
-$k=2r$; fully assigned factors over independent color classes.
+principal minors gives chi[L,0](z)=det(zI-L/2) (a minor on S occurs with
+probability 2^{-|S|}). Keeping either L or -L and using monotonicity proves
+(eq:poly-mixed-lower). All-unassigned uses r copies each of T,-T, so
+k=2r; fully assigned factors over independent color classes.
 
 
 ## Stable quotients and rooted spectral measures
 
-For a matrix $B$ write $ImB=(B-B^*)/(2i)$. The next lemma gives
+For a matrix B write ImB=(B-B^*)/(2i). The next lemma gives
 the interlacing that measures the effect of one site; its half-plane part also
 allows non-Hermitian matrices, for the analytic perturbations later.
 
@@ -824,226 +816,226 @@ For the finite data,
 
  P_I(Z)=(prod_{vin I}tfrac1{k_v!}∂_{z_v}^{,k_v-1})
             prod_jdet((Z-B_j)|_{U_j}),
- qquad ∂_{z_v}P_I=P_{-v}.
+  ∂_{z_v}P_I=P_{-v}.
 
-More generally use arbitrary $B_j$ on $U_j$ in the same independent-choice
-construction. If $ImB_jle hI$ then $P_I(Z)$ and every deletion
-polynomial are nonzero whenever $Imz_v>h$ at every remaining site,
+More generally use arbitrary B_j on U_j in the same independent-choice
+construction. If ImB_jle hI then P_I(Z) and every deletion
+polynomial are nonzero whenever Imz_v>h at every remaining site,
 and there
 
  Imfrac{P_I(Z)}{P_{-v}(Z|_{I∖{v}})}
  geImz_v-h.
 
-Back in the Hermitian data, set $Q_v=P_{-v}$, $F_v=P_I/Q_v$, $G_v=Q_v/P_I$. Then
-$F_v,G_v$ are analytic in the upper half-plane and
+Back in the Hermitian data, set Q_v=P_{-v}, F_v=P_I/Q_v, G_v=Q_v/P_I. Then
+F_v,G_v are analytic in the upper half-plane and
 
- F_v(z)=z+a_v-sum_lambdafrac{b_{v,lambda}}{z-lambda},qquad b_{v,lambda}ge0,
+ F_v(z)=z+a_v-sum_lambdafrac{b_{v,lambda}}{z-lambda}, b_{v,lambda}ge0,
 
-over distinct roots $lambda$ of $Q_v$. Letting $Λ_v$ list those roots with
-multiplicity, there is $g_v$ with squared coordinate $b_{v,lambda}$ on one copy
-of each $lambda$ (zero on others) such that
+over distinct roots lambda of Q_v. Letting Λ_v list those roots with
+multiplicity, there is g_v with squared coordinate b_{v,lambda} on one copy
+of each lambda (zero on others) such that
 
  H_v=-a_v&g_v^* g_v&Λ_v
 
-has characteristic polynomial $P_I$. Its first-coordinate spectral measure
-$rho_v$ is a probability measure with
+has characteristic polynomial P_I. Its first-coordinate spectral measure
+rho_v is a probability measure with
 
- G_v(z)=int(rho_v(dx))/(z-x),qquad
+ G_v(z)=int(rho_v(dx))/(z-x),
  sum_{vin I}rho_v=sum_{P_I(x)=0}delta_x,
 
-roots counted with multiplicity, and $Q_v$ interlaces $P_I$.
+roots counted with multiplicity, and Q_v interlaces P_I.
 
 
 Proof. 
 Each determinant factor is affine in any diagonal variable. Differentiating
-$k_v-1$ times in $z_v$, every surviving Leibniz term leaves $v$ in exactly one
-of its $k_v$ factors and deletes it from the others; the factor
-$(k_v-1)!/k_v!=1/k_v$ gives the outcome average, and one more derivative in
-$z_v$ deletes the site, proving (eq:poly-differential).
+k_v-1 times in z_v, every surviving Leibniz term leaves v in exactly one
+of its k_v factors and deletes it from the others; the factor
+(k_v-1)!/k_v!=1/k_v gives the outcome average, and one more derivative in
+z_v deletes the site, proving (eq:poly-differential).
 
-If $Imz_v>h$, each $Z-B_j$ on $U_j$ has positive imaginary part
+If Imz_v>h, each Z-B_j on U_j has positive imaginary part
 and is invertible. Differentiation preserves nonvanishing in a half-plane by
 Gauss--Lucas; the degree cannot drop because the leading coefficient in an
-unprocessed $z_v$ is the corresponding less-processed expression with $v$
+unprocessed z_v is the corresponding less-processed expression with v
 removed, handled by simultaneous induction over processed vertices and principal
 restrictions. So every deletion polynomial is nonzero there too. For fixed other
-coordinates, $P_I/P_{-v}=z_v+zeta_0$ (derivative identity and multiaffineness)
-has zero in $Imle h$, giving (eq:poly-half-plane).
+coordinates, P_I/P_{-v}=z_v+zeta_0 (derivative identity and multiaffineness)
+has zero in Imle h, giving (eq:poly-half-plane).
 
 For Hermitian data the uniform-variable polynomials are real-rooted (real
-coefficients, half-plane nonvanishing with conjugation). With $h=0$,
-$F_v(z)-z$ has nonnegative imaginary part and constant $a_v$ at infinity. A
+coefficients, half-plane nonvanishing with conjugation). With h=0,
+F_v(z)-z has nonnegative imaginary part and constant a_v at infinity. A
 nonremovable real pole of a real rational function with nonnegative imaginary
 part is simple with nonpositive residue (approach the pole from different angles
 to exclude higher order, then vertically for the sign), proving
-(eq:poly-arrow-quotient). The Schur complement of $z-Λ_v$ in $z-H_v$
-is $F_v(z)$, so $det(z-H_v)=Q_vF_v=P_I$ and its first resolvent entry is
-$G_v$. The spectral theorem gives the first identity in
+(eq:poly-arrow-quotient). The Schur complement of z-Λ_v in z-H_v
+is F_v(z), so det(z-H_v)=Q_vF_v=P_I and its first resolvent entry is
+G_v. The spectral theorem gives the first identity in
 (eq:poly-root-measures), principal-matrix interlacing the interlacing.
 Uniform specialization of (eq:poly-differential) gives
-$sum_vG_v=P_I'/P_I$, the Cauchy transform of the root-counting measure; unique
+sum_vG_v=P_I'/P_I, the Cauchy transform of the root-counting measure; unique
 partial fractions give the second identity.
 
 
-At an unassigned $v$, first condition its pair on color $i$ keeping both signs
-uniform and shift $t$; call these $P^0_{v,i}$ and $F^0_{v,i}=P^0_{v,i}/Q_v$. Then
-raise its shift to $beta t$ as a permanent assignment: $P_{v,i}$, $F_{v,i}$.
-Deletion of $v$ gives the same $Q_v$, so
+At an unassigned v, first condition its pair on color i keeping both signs
+uniform and shift t; call these P^0_{v,i} and F^0_{v,i}=P^0_{v,i}/Q_v. Then
+raise its shift to beta t as a permanent assignment: P_{v,i}, F_{v,i}.
+Deletion of v gives the same Q_v, so
 
- frac1rsum_{i=1}^r F^0_{v,i}=F_v,qquad
+ frac1rsum_{i=1}^r F^0_{v,i}=F_v,
  F_{v,i}=F^0_{v,i}+(beta-1)tW_v.
 
 The arrow representation and rooted measures apply to these too; the residues of
-$F^0_{v,i}$ are nonnegative and average to those of $F_v$.
+F^0_{v,i} are nonnegative and average to those of F_v.
 
 ## Positive-root costs and locality
 
-For a monic real-rooted $P$ define $s(P)=sum_{P(x)=0}x_+$ with
-$x_+=max{x,0}$ and $s(1)=0$, roots with multiplicity. For the finite data set
+For a monic real-rooted P define s(P)=sum_{P(x)=0}x_+ with
+x_+=max{x,0} and s(1)=0, roots with multiplicity. For the finite data set
 
- m_v=int x_+,rho_v(dx),qquad d_v=s(P_I)-s(P_{-v}),
+ m_v=int x_+,rho_v(dx), d_v=s(P_I)-s(P_{-v}),
 
-so $sum_vm_v=s(P_I)$. Superscripts $0,i$ and $i$ denote the conditioned and
-permanently assigned options; e.g. $d_v^{,i}=s(P_{v,i})-s(Q_v)$.
+so sum_vm_v=s(P_I). Superscripts 0,i and i denote the conditioned and
+permanently assigned options; e.g. d_v^{,i}=s(P_{v,i})-s(Q_v).
 
 [Initial cost and detection of excessive norm]
 
-(i) If all sites are unassigned and $t>3/sqrt r$ then $P_I(x)>0$ for $xge0$
-and $s(P_I)=0$ (including $I=∅$). (ii) If $Ine∅$ and all
-sites are assigned, put $T_c=sum_i1_{{c=i}}T1_{{c=i}}$. For every
-$alpha>0$, if $|T_c|>2beta t+alpha$ then
+(i) If all sites are unassigned and t>3/sqrt r then P_I(x)>0 for xge0
+and s(P_I)=0 (including I=∅). (ii) If Ine∅ and all
+sites are assigned, put T_c=sum_i1_{{c=i}}T1_{{c=i}}. For every
+alpha>0, if |T_c|>2beta t+alpha then
 [
  lambda_{max}P_Igefracalpha2min_{vin I}W_v,
- qquad s(P_I)gefracalpha2min_{vin I}W_v.
+  s(P_I)gefracalpha2min_{vin I}W_v.
 ]
 
 
 Proof. 
-For $J⊆ I$ let $P_J^{un}$ be the expected polynomial with same
+For J⊆ I let P_J^{un} be the expected polynomial with same
 pair laws, unit weights, no shifts. Factoring weights from each outcome
 determinant,
 
  P_J(x)=(prod_{vin J}W_v)
  P_J^{un}((tau_v+x/W_v)_{vin J}),
 
-both sides $1$ for $J=∅$. Multiaffineness and
-(eq:poly-differential) also give, for real $b$ and increments $h_vge0$,
+both sides 1 for J=∅. Multiaffineness and
+(eq:poly-differential) also give, for real b and increments h_vge0,
 [
  P_J^{un}((b+h)|_J)
  =sum_{S⊆ J}P_{J∖ S}^{un}(b|_{J∖ S})prod_{vin S}h_v,
 ]
-so positivity of all principal evaluations at $b$ survives raising any
+so positivity of all principal evaluations at b survives raising any
 coordinates.
 
-(i) If all unassigned and $t>3/sqrt r$, Lemma~lem:mixed-bounds puts $t$
-above the largest root of every nonempty $P_J^{un}$, so all principal
-evaluations at $t$ are positive; for $xge0$ the coordinates $t+x/W_v$ are $ge
-t$, so $P_I(x)>0$ and $s(P_I)=0$.
+(i) If all unassigned and t>3/sqrt r, Lemma~lem:mixed-bounds puts t
+above the largest root of every nonempty P_J^{un}, so all principal
+evaluations at t are positive; for xge0 the coordinates t+x/W_v are ge
+t, so P_I(x)>0 and s(P_I)=0.
 
-(ii) Suppose all assigned and $|T_c|>2beta t+alpha$. The fully assigned
-factorization and (eq:poly-mixed-lower) give $P_I^{un}$ a root
-above $beta t+alpha/2$. If $lambda_{max}P_I<x_0:=(alpha/2)min_vW_v$, then
-repeated interlacing puts $x_0$ above every root of every nonempty principal
-restriction, so $P_J(x_0)>0$; by (eq:poly-remove-weights) the corresponding
+(ii) Suppose all assigned and |T_c|>2beta t+alpha. The fully assigned
+factorization and (eq:poly-mixed-lower) give P_I^{un} a root
+above beta t+alpha/2. If lambda_{max}P_I<x_0:=(alpha/2)min_vW_v, then
+repeated interlacing puts x_0 above every root of every nonempty principal
+restriction, so P_J(x_0)>0; by (eq:poly-remove-weights) the corresponding
 unweighted evaluations are positive at
-$b_v=beta t+x_0/W_vlebeta t+alpha/2$, and raising to a common
-$qgebeta t+alpha/2$ keeps them positive, so $P_I^{un}(q)>0$ for all
-such $q$, contradicting the root above $beta t+alpha/2$. Hence
-$lambda_{max}P_Ige x_0$, giving the $s(P_I)$ bound.
+b_v=beta t+x_0/W_vlebeta t+alpha/2, and raising to a common
+qgebeta t+alpha/2 keeps them positive, so P_I^{un}(q)>0 for all
+such q, contradicting the root above beta t+alpha/2. Hence
+lambda_{max}P_Ige x_0, giving the s(P_I) bound.
 
 
 [Deletion costs and tails]
-For the finite Hermitian data let $n_P(x)$ count roots of $P$ above $x$ with
-multiplicity. For every $vin I$,
+For the finite Hermitian data let n_P(x) count roots of P above x with
+multiplicity. For every vin I,
 
  0&le n_{P_I}(x)-n_{Q_v}(x)le1, 
  d_v&=int_0^∞(n_{P_I}-n_{Q_v})(x),dx, 
  int_B^∞(n_{P_I}-n_{Q_v})(x),dx
- &le2int(x-B)_+,rho_v(dx)+(a_v+B)rho_v((B,∞)),quad Bge0.
+ &le2int(x-B)_+,rho_v(dx)+(a_v+B)rho_v((B,∞)), Bge0.
  
 
-In particular $0le d_vle2m_v+a_v$. All hold for either modification at $v$,
+In particular 0le d_vle2m_v+a_v. All hold for either modification at v,
 with its own shift and rooted measure.
 
 
 Proof. 
 (eq:poly-interlacing-count) is principal interlacing from
 Lemma~lem:polynomial-quotients; integration gives
-(eq:poly-cost-integral). For the tail write $H=H_v$,
-$Λ=Λ_v$, $a=a_v$, $L=H-BI$, and let $E$ be the first-coordinate
-projection, $D=1-E$, $R=1_{(0,∞)}(L)$. The compression $DRD$ is a positive
+(eq:poly-cost-integral). For the tail write H=H_v,
+Λ=Λ_v, a=a_v, L=H-BI, and let E be the first-coordinate
+projection, D=1-E, R=1_{(0,∞)}(L). The compression DRD is a positive
 contraction on deletion coordinates, so by the variational formula for the
-positive trace $Tr(Λ-BI)_+geTr((Λ-BI)DRD)$. Since
-$L-DLD=EL+LE-ELE$,
+positive trace Tr(Λ-BI)_+geTr((Λ-BI)DRD). Since
+L-DLD=EL+LE-ELE,
 
  Tr(H-BI)_+-Tr(Λ-BI)_+
  &leTr((EL+LE-ELE)R)
  &=2< e_0,LRe_0>+(a+B)< e_0,Re_0>
  &=2int(x-B)_+,rho_v(dx)+(a+B)rho_v((B,∞)),
 
-with $e_0$ the first unit vector of the arrow matrix. The left side is the tail
-integral. Taking $B=0$ and $rho_v$ a probability gives the last bound.
+with e_0 the first unit vector of the arrow matrix. The left side is the tail
+integral. Taking B=0 and rho_v a probability gives the last bound.
 
 
-A graph supports $T$ if it is undirected, loopless, and $T_{uv}=0$ for
-nonadjacent $une v$.
+A graph supports T if it is undirected, loopless, and T_{uv}=0 for
+nonadjacent une v.
 
 [Rooted locality and bounds]
-If a graph supports the finite zero-diagonal Hermitian contraction $T$, the
-coefficient of $z^{-1-k}$ in $G_v(z)$ is $int x^krho_v(dx)$ and is determined
-by the data on the radius-$k$ ball about $v$. The Laurent coefficients of $F_v$
-and of $F^0_{v,i},F_{v,i}$ at an unassigned site depend only on a radius bounded
+If a graph supports the finite zero-diagonal Hermitian contraction T, the
+coefficient of z^{-1-k} in G_v(z) is int x^krho_v(dx) and is determined
+by the data on the radius-k ball about v. The Laurent coefficients of F_v
+and of F^0_{v,i},F_{v,i} at an unassigned site depend only on a radius bounded
 by their order. Moreover
 
  int x^2,rho_v(dx)
  =a_v^2+sum_{une v}W_vW_u|T_{vu}|^2mathbb P(bbP(omega_u=omega_v),
- quad
+ 
  sum_lambda b_{v,lambda}
  =sum_{une v}W_vW_u|T_{vu}|^2mathbb P(bbP(omega_u=omega_v),
 
 and the same holds for either site option with its own pair law and shift. If
-neighboring weight ratios lie in $[K^{-1},K]$, $Kge1$, then
+neighboring weight ratios lie in [K^{-1},K], Kge1, then
 
- int x^2rho_vle((beta t)^2+K)W_v^2,quad
- m_vle C_0W_v,quad d_vle(2C_0+beta t)W_v,
- quad C_0=√((beta t)^2+K).
+ int x^2rho_vle((beta t)^2+K)W_v^2,
+ m_vle C_0W_v, d_vle(2C_0+beta t)W_v,
+  C_0=√((beta t)^2+K).
 
-If all weights are at most $H_0$, every polynomial and principal restriction has
+If all weights are at most H_0, every polynomial and principal restriction has
 roots in
 
- [-L,L],qquad L=(1+beta t)H_0,
+ [-L,L], L=(1+beta t)H_0,
 
 and the rooted measures share this support bound.
 
 
 Proof. 
-Put $u_v=z_v^{-1}$ and $mathcal H_I(u)=P_I(Z)calH_I(u)=P_I(Z)prod_vu_v=sum_{Ssubset I}c_Sprod_{vin S}u_v$,
-where $c_S$ is the expected determinant of the negative outcome matrix restricted
-to $S$. If $S$ splits into graph components the outcome splits and choices are
-independent, so $c_S$ factors over components. Since $mathcal H_I(0)=1$calH_I(0)=1$ its formal
+Put u_v=z_v^{-1} and mathcal H_I(u)=P_I(Z)calH_I(u)=P_I(Z)prod_vu_v=sum_{Ssubset I}c_Sprod_{vin S}u_v,
+where c_S is the expected determinant of the negative outcome matrix restricted
+to S. If S splits into graph components the outcome splits and choices are
+independent, so c_S factors over components. Since mathcal H_I(0)=1calH_I(0)=1 its formal
 logarithm is defined, and on a disconnected subset it is a sum over components,
-so every nonconstant monomial in $logmathcal H_I$calH_I$ has connected support. As
-$G_v=∂_{z_v}log P_I$, apart from $1/z_v$ its monomials use connected
-supports containing $v$; the coefficient of order $z^{-1-k}$ uses only the
-radius-$k$ ball, matching the $k$-th rooted moment. Formal inversion gives the
-same for $F_v$, and conditioning/changing the shift preserves this.
+so every nonconstant monomial in logmathcal H_IcalH_I has connected support. As
+G_v=∂_{z_v}log P_I, apart from 1/z_v its monomials use connected
+supports containing v; the coefficient of order z^{-1-k} uses only the
+radius-k ball, matching the k-th rooted moment. Formal inversion gives the
+same for F_v, and conditioning/changing the shift preserves this.
 
 Singleton and two-point coefficients are
-$c_{{v}}=a_v$ and
-$c_{{u,v}}=a_ua_v-W_uW_v|T_{uv}|^2mathbb P(bbP(omega_u=omega_v)$. Expanding the
-logarithmic derivative to order $z^{-3}$ gives the first identity of
-(eq:poly-second-moment); the arrow matrix has first diagonal $-a_v$ and
-off-diagonal squared norm $sum_lambda b_{v,lambda}$, giving the second.
+c_{{v}}=a_v and
+c_{{u,v}}=a_ua_v-W_uW_v|T_{uv}|^2mathbb P(bbP(omega_u=omega_v). Expanding the
+logarithmic derivative to order z^{-3} gives the first identity of
+(eq:poly-second-moment); the arrow matrix has first diagonal -a_v and
+off-diagonal squared norm sum_lambda b_{v,lambda}, giving the second.
 
 Under the ratio hypothesis the sum in (eq:poly-second-moment) is at most
-$KW_v^2sum_{une v}|T_{vu}|^2le KW_v^2$ (also for each conditioned law, as
-collision probabilities are $le1$), and $a_vlebeta tW_v$.
-Cauchy--Schwarz for $rho_v$ plus Lemma~lem:positive-cost gives
+KW_v^2sum_{une v}|T_{vu}|^2le KW_v^2 (also for each conditioned law, as
+collision probabilities are le1), and a_vlebeta tW_v.
+Cauchy--Schwarz for rho_v plus Lemma~lem:positive-cost gives
 (eq:poly-weight-bounds).
 
-Each outcome has norm at most $(1+beta t)H_0$, so for real $z>L$ all
-characteristic determinants are positive and for $z<-L$ all have sign
-$(-1)^{|I|}$; their average has no root outside $[-L,L]$, and real-rootedness
+Each outcome has norm at most (1+beta t)H_0, so for real z>L all
+characteristic determinants are positive and for z<-L all have sign
+(-1)^{|I|}; their average has no root outside [-L,L], and real-rootedness
 proves (eq:poly-root-support). Restriction, site options, and the spectral
 representation give the rest.
 
@@ -1051,9 +1043,9 @@ representation give the rest.
 ## The single-site assignment estimate
 
 [Assigning one color]
-For a finite zero-diagonal Hermitian contraction $T$, arbitrary positive weights,
-$rge1$, $t>0$, a partial coloring as above, and $v$ unassigned, with
-$beta=80$ there is $iin{1,...,r}$ with
+For a finite zero-diagonal Hermitian contraction T, arbitrary positive weights,
+rge1, t>0, a partial coloring as above, and v unassigned, with
+beta=80 there is iin{1,...,r} with
 
  d_v^{,i}-d_vle24m_v.
 
@@ -1061,84 +1053,84 @@ The bound is independent of the finite set and its weights.
 
 
 Proof. 
-Write $a=a_v=tW_v>0$, $m=m_v$, $d=d_v$, and
-$H=-a&g^* g&Λ$ for the old arrow matrix. By
-(eq:poly-options) the color-conditioned residues $b_{i,lambda}$ are
-nonnegative with average $b_lambda$; a permanent assignment replaces them by
-$b_{i,lambda}$ and the first diagonal by $-beta a$, with $Λ$ fixed.
+Write a=a_v=tW_v>0, m=m_v, d=d_v, and
+H=-a&g^* g&Λ for the old arrow matrix. By
+(eq:poly-options) the color-conditioned residues b_{i,lambda} are
+nonnegative with average b_lambda; a permanent assignment replaces them by
+b_{i,lambda} and the first diagonal by -beta a, with Λ fixed.
 
-Split the deletion coordinates into $lambda>0$ and $lambdale0$ (subscripts
-$+$,$-$), write $H=H_+-H_-$, and set
-$e_+=Tr(H_-)_{++}$, $e_-=Tr(H_+)_{--}$. Comparing diagonal blocks of
-$H=H_+-H_-$ and using that the first diagonal of $H_+$ is $m$,
+Split the deletion coordinates into lambda>0 and lambdale0 (subscripts
++,-), write H=H_+-H_-, and set
+e_+=Tr(H_-)_{++}, e_-=Tr(H_+)_{--}. Comparing diagonal blocks of
+H=H_+-H_- and using that the first diagonal of H_+ is m,
 
- d=m+e_++e_-,qquad (H_-)_{00}=m+a.
+ d=m+e_++e_-, (H_-)_{00}=m+a.
 
-We treat $mge a$ and $m<a$.
+We treat mge a and m<a.
 
-Case $mge a$. Consider the two side arrows keeping the first coordinate
-and one deletion block, first diagonal $-a$. The positive block has exactly one
-negative eigenvalue $-L_0$ with
-$L_0=a+sum_{lambda>0}b_lambda/(L_0+lambda)$ and increment $d^+=L_0-a$
-(valid empty, $L_0=a$). The nonpositive block has at most one positive
-eigenvalue $s_0$ (else $0$); its increment is $d^-=s_0$ and
+Case mge a. Consider the two side arrows keeping the first coordinate
+and one deletion block, first diagonal -a. The positive block has exactly one
+negative eigenvalue -L_0 with
+L_0=a+sum_{lambda>0}b_lambda/(L_0+lambda) and increment d^+=L_0-a
+(valid empty, L_0=a). The nonpositive block has at most one positive
+eigenvalue s_0 (else 0); its increment is d^-=s_0 and
 
  sum_{lambdale0}(b_lambda)/(s_0+|lambda|)le s_0+a.
 
-If $s_0=0$ all residues at $lambda=0$ vanish (otherwise a positive zero
-appears); omit those terms, and if the remaining sum at zero exceeds $a$,
+If s_0=0 all residues at lambda=0 vanish (otherwise a positive zero
+appears); omit those terms, and if the remaining sum at zero exceeds a,
 continuity at infinity again yields a positive zero, so
 (eq:poly-nonpositive-test) holds. Pinching off the other coupling cannot
-increase the positive trace, so $d^+le d$ and $d^-le d$. By averaging, at least
+increase the positive trace, so d^+le d and d^-le d. By averaging, at least
 three quarters of the colors satisfy
-$sum_{lambda>0}b_{i,lambda}/(L_0+lambda)le4d^+$ and at least three quarters
-satisfy $sum_{lambdale0}b_{i,lambda}/(s_0+|lambda|)le4(s_0+a)$, so some
+sum_{lambda>0}b_{i,lambda}/(L_0+lambda)le4d^+ and at least three quarters
+satisfy sum_{lambdale0}b_{i,lambda}/(s_0+|lambda|)le4(s_0+a), so some
 color satisfies both (zero old residues stay zero).
 
-For this color the positive-block arrow with diagonal $-a$ has increment at most
-$4d^+$ (immediate if the eigenvalue magnitude is $le L_0$, else bounded by the
-test sum), and the nonpositive-block arrow with diagonal $-4a$ has increment at
-most $4s_0$ (a positive root $s>4s_0$ would need a residue sum above
-$4(s_0+a)$). Extend both by zero and add: the full arrow has diagonal $-5a$ and,
-by subadditivity of the positive trace, increment at most $4d^++4d^-$;
-decreasing the diagonal to $-beta a$ only lowers positive trace. Hence
+For this color the positive-block arrow with diagonal -a has increment at most
+4d^+ (immediate if the eigenvalue magnitude is le L_0, else bounded by the
+test sum), and the nonpositive-block arrow with diagonal -4a has increment at
+most 4s_0 (a positive root s>4s_0 would need a residue sum above
+4(s_0+a)). Extend both by zero and add: the full arrow has diagonal -5a and,
+by subadditivity of the positive trace, increment at most 4d^++4d^-;
+decreasing the diagonal to -beta a only lowers positive trace. Hence
 [
- d_v^{,i}le4(d^++d^-)le8d,qquad
+ d_v^{,i}le4(d^++d^-)le8d,
  d_v^{,i}-dle7dle7(2m+a)le21m,
 ]
 using Lemma~lem:positive-cost.
 
-Case $m<a$. For a positive block matrix the outer product of its first
+Case m<a. For a positive block matrix the outer product of its first
 coupling column is bounded by its first diagonal times its remaining block. Apply
-to $H_+$ and $H_-$ and use $(p-q)(p-q)^*le2pp^*+2qq^*$; since
-$(H_+)_{++}=Λ_++(H_-)_{++}$ and
-$(H_-)_{--}=|Λ_-|+(H_+)_{--}$,
+to H_+ and H_- and use (p-q)(p-q)^*le2pp^*+2qq^*; since
+(H_+)_{++}=Λ_++(H_-)_{++} and
+(H_-)_{--}=|Λ_-|+(H_+)_{--},
 
  g_+g_+^*&le 2mΛ_++(4m+2a)(H_-)_{++}, 
  g_-g_-^*&le 2(m+a)|Λ_-|+(4m+2a)(H_+)_{--}. 
 
-Conjugating (eq:poly-positive-block) by $(32a+Λ_+)^{-1/2}$ and taking
+Conjugating (eq:poly-positive-block) by (32a+Λ_+)^{-1/2} and taking
 the operator norm (left side rank one, right error bounded by its trace, and
-$4m+2ale6a$) gives
+4m+2ale6a) gives
 
  sum_{lambda>0}(b_lambda)/(32a+lambda)le2m+frac6{32}e_+.
 
-Put $L_1=e_-$. If $L_1>0$, conjugating (eq:poly-negative-block) by
-$(L_1+|Λ_-|)^{-1/2}$ gives
+Put L_1=e_-. If L_1>0, conjugating (eq:poly-negative-block) by
+(L_1+|Λ_-|)^{-1/2} gives
 
  sum_{lambdale0}(b_lambda)/(L_1+|lambda|)le2(m+a)+(4m+2a)le10a;
 
-if $L_1=0$, positivity gives $(H_+)_{--}=0$ and
+if L_1=0, positivity gives (H_+)_{--}=0 and
 (eq:poly-negative-block) forces zero coupling on the kernel of
-$|Λ_-|$, so invert there and omit zero residues at $lambda=0$, keeping
+|Λ_-|, so invert there and omit zero residues at lambda=0, keeping
 (eq:poly-negative-test). Choose a color with both residue sums at most four
-times their bounds. For it, the positive-block arrow with diagonal $-32a$ has
-increment $le8m+(24)/(32)e_+$; the nonpositive-block arrow with diagonal
-$-40a$ has increment $le L_1$ (a root $s>L_1$ would force residue sum above
-$40a$, contradicting $s+40a>40a$; covers $L_1=0$ with zero residues omitted).
-Adding gives diagonal $-72a$; subadditivity and lowering to $-80a$ give
+times their bounds. For it, the positive-block arrow with diagonal -32a has
+increment le8m+(24)/(32)e_+; the nonpositive-block arrow with diagonal
+-40a has increment le L_1 (a root s>L_1 would force residue sum above
+40a, contradicting s+40a>40a; covers L_1=0 with zero residues omitted).
+Adding gives diagonal -72a; subadditivity and lowering to -80a give
 [
- d_v^{,i}le8m+tfrac34e_++e_-,qquad
+ d_v^{,i}le8m+tfrac34e_++e_-,
  d_v^{,i}-dle7m-tfrac14e_+le7m
 ]
 with (eq:poly-error-decomposition). Both cases give
@@ -1155,32 +1147,32 @@ in ordered comparison environments; and deleting vertices outside a fixed weight
 band has a small effect within the band, uniformly in the largest weight. The
 uniformity removes weight cutoffs in the next section.
 
-Let $(X,nu)$ be a standard finite measure space and $mathcal R$calR$ a countable
-measured equivalence relation preserving $nu$, so each measurable partial
-isomorphism with graph in $mathcal R$calR$ preserves measure. Let $Gsubsetmathcal
-R$calR$ be a measurable undirected loopless graph of degree at most $D_0<∞$.
-On orbit spaces let $T$ be measurable Hermitian matrices of norm at most one,
-zero diagonal, supported on $G$; changing the root conjugates them by diagonal
+Let (X,nu) be a standard finite measure space and mathcal RcalR a countable
+measured equivalence relation preserving nu, so each measurable partial
+isomorphism with graph in mathcal RcalR preserves measure. Let Gsubsetmathcal
+RcalR be a measurable undirected loopless graph of degree at most D_0<∞.
+On orbit spaces let T be measurable Hermitian matrices of norm at most one,
+zero diagonal, supported on G; changing the root conjugates them by diagonal
 unitaries, and all constructions below are invariant under this. Null
-exceptions are discarded along their $mathcal R$-saturations.calR$-saturations.
+exceptions are discarded along their mathcal R-saturations.calR-saturations.
 
-Fix $rge1$, $t>0$, $beta=80$. A partial coloring is $c:Xto{0,1,...,r}$
-($0$ unassigned). Use the shifts and polynomial choices of
-Section~sec:polynomials and a weight $W:Xto(0,H_0]$, $H_0<∞$. After a
+Fix rge1, t>0, beta=80. A partial coloring is c:Xto{0,1,...,r}
+(0 unassigned). Use the shifts and polynomial choices of
+Section~sec:polynomials and a weight W:Xto(0,H_0], H_0<∞. After a
 measurable set is deleted take principal restrictions, keeping the original
 graph for distances. At absent vertices every rooted cost or transform is zero
-and the rooted measure is zero. On a finite principal $I$ use
+and the rooted measure is zero. On a finite principal I use
 [
- P_I,quad Q_{I,v}=P_{I∖{v}},quad
- F_{I,v}=tfrac{P_I}{Q_{I,v}},quad G_{I,v}=tfrac1{F_{I,v}},quad
- rho_{I,v},quad m_{I,v},quad d_{I,v},
+ P_I, Q_{I,v}=P_{I∖{v}},
+ F_{I,v}=tfrac{P_I}{Q_{I,v}}, G_{I,v}=tfrac1{F_{I,v}},
+ rho_{I,v}, m_{I,v}, d_{I,v},
 ]
-with $m_{I,v}=int x_+rho_{I,v}(dx)$,
-$d_{I,v}=s(P_I)-s(Q_{I,v})$, $s(P)=sum_{P(x)=0}x_+$. Write $a_v=tau_vW_v$ and
+with m_{I,v}=int x_+rho_{I,v}(dx),
+d_{I,v}=s(P_I)-s(Q_{I,v}), s(P)=sum_{P(x)=0}x_+. Write a_v=tau_vW_v and
 
  L=(1+beta t)H_0.
 
-All roots and rooted measures of these finite data lie in $[-L,L]$, including
+All roots and rooted measures of these finite data lie in [-L,L], including
 virtual options at an unassigned vertex.
 
 Measure invariance gives mass transport
@@ -1188,67 +1180,67 @@ Measure invariance gives mass transport
  int_Xsum_{uin[v]_{mathcal R}calR}h(v,u),dnu(v)
  =int_Xsum_{vin[u]_{mathcal R}calR}h(v,u),dnu(u)
 
-for nonnegative (and for integrable complex) measurable $h$: partition
-$mathcal R$calR$ into countably many partial one-to-one maps and change variables.
+for nonnegative (and for integrable complex) measurable h: partition
+mathcal RcalR into countably many partial one-to-one maps and change variables.
 
 ## Limits on bounded weighted graphs
 
 
-For the above data on the finite measure space $(X,nu)$, the rooted measures,
+For the above data on the finite measure space (X,nu), the rooted measures,
 transforms, and costs converge as finite principal sets exhaust the graph
-component of the root, to limits $rho_v,G_v,F_v,m_v,d_v$ independent of the
-exhaustion and measurable in $v$. At each present vertex $rho_v$ is a
-probability measure on $[-L,L]$ and, for $z$ in the open upper half-plane,
+component of the root, to limits rho_v,G_v,F_v,m_v,d_v independent of the
+exhaustion and measurable in v. At each present vertex rho_v is a
+probability measure on [-L,L] and, for z in the open upper half-plane,
 [
- G_v(z)=int(rho_v(dx))/(z-x),qquad F_v(z)=frac1{G_v(z)}.
+ G_v(z)=int(rho_v(dx))/(z-x), F_v(z)=frac1{G_v(z)}.
 ]
 The same holds after conditioning the color at one unassigned vertex with either
-shift. For fixed $D_0,H_0,r,t$ the costs $m_v,d_v,d_v^{,i}$ ($1le ile r$) are
+shift. For fixed D_0,H_0,r,t the costs m_v,d_v,d_v^{,i} (1le ile r) are
 uniformly approximable by a radius depending only on the parameters, not on the
 graph, weights, coloring, or deleted set. These statements apply to a
 finite-measure restriction of an invariant relation.
 
 
 Proof. 
-By Lemma~lem:rooted-locality each moment $int x^krho_{I,v}$ is fixed by a
-bounded neighborhood and unchanged once $I$ contains it. Compactness of
-probabilities on $[-L,L]$ and moment uniqueness give an exhaustion-independent
-weak limit $rho_v$ with Cauchy transform $G_{I,v}to G_v$ locally uniformly on
-the upper half-plane. Since $-ImG_v(z)=(Imz)int
-|z-x|^{-2}rho_v(dx)>0$, this limit inverts, so $F_{I,v}to F_v$ there; the same
+By Lemma~lem:rooted-locality each moment int x^krho_{I,v} is fixed by a
+bounded neighborhood and unchanged once I contains it. Compactness of
+probabilities on [-L,L] and moment uniqueness give an exhaustion-independent
+weak limit rho_v with Cauchy transform G_{I,v}to G_v locally uniformly on
+the upper half-plane. Since -ImG_v(z)=(Imz)int
+|z-x|^{-2}rho_v(dx)>0, this limit inverts, so F_{I,v}to F_v there; the same
 holds for altered color laws or shift at the root. The finite quotients have the
-form $F_{I,v}(z)-z-a_v=-intkappa_{I,v}(dx)/(z-x)$ with a positive residue
-measure supported on $[-L,L]$ of mass at most $H_0^2$
+form F_{I,v}(z)-z-a_v=-intkappa_{I,v}(dx)/(z-x) with a positive residue
+measure supported on [-L,L] of mass at most H_0^2
 (Lemma~lem:rooted-locality); compactness and Cauchy uniqueness give a
-limiting residue measure, justifying Laurent-coefficient limits. Since $x_+$ is
-continuous on $[-L,L]$, $m_{I,v}$ converges, uniformly approximable by finite
+limiting residue measure, justifying Laurent-coefficient limits. Since x_+ is
+continuous on [-L,L], m_{I,v} converges, uniformly approximable by finite
 neighborhoods.
 
 Deletion increments need separate treatment because they involve roots at every
-vertex of a finite set. For real $f$ put
-$d_{I,v}[f]=sum_{P_I(x)=0}f(x)-sum_{Q_{I,v}(x)=0}f(x)$; the rooted-measure
+vertex of a finite set. For real f put
+d_{I,v}[f]=sum_{P_I(x)=0}f(x)-sum_{Q_{I,v}(x)=0}f(x); the rooted-measure
 sum identity gives
 
  d_{I,v}[f]=sum_{uin I}(int f,drho_{I,u}
    -1_{{une v}}int f,drho_{I∖{v},u}).
 
-Each summand is local (vanishes unless $u$ is within a radius depending only on
-$deg f$ of $v$), so the whole increment is a finite-neighborhood calculation
-independent of the ambient $I$.
+Each summand is local (vanishes unless u is within a radius depending only on
+deg f of v), so the whole increment is a finite-neighborhood calculation
+independent of the ambient I.
 
-For absolutely continuous real $f$ on $[-L,L]$, interlacing gives
+For absolutely continuous real f on [-L,L], interlacing gives
 
  d_{I,v}[f]=f(-L)+int_{-L}^{L}f'(x)(n_{P_I}-n_{Q_{I,v}})(x),dx,
- qquad 0le n_{P_I}-n_{Q_{I,v}}le1,
+  0le n_{P_I}-n_{Q_{I,v}}le1,
 
-$n_P$ counting roots above $x$. Hence
+n_P counting roots above x. Hence
 
  |d_{I,v}[f]-d_{I,v}[g]|le|f(-L)-g(-L)|+|f'-g'|_{L^1([-L,L])}.
 
-Approximate $f'$ in $L^1$ by polynomials with prescribed initial value, giving
+Approximate f' in L^1 by polynomials with prescribed initial value, giving
 polynomial approximants with both function and increment error tending to zero.
-Apply to $f=x_+$: the local polynomial increments are eventually constant with
-uniform approximation error in every finite set, so $d_{I,v}$ has the stated
+Apply to f=x_+: the local polynomial increments are eventually constant with
+uniform approximation error in every finite set, so d_{I,v} has the stated
 limit and uniform locality; similarly for each virtual option.
 
 Finite balls of a measurable bounded-degree graph, their matrix data, and the
@@ -1262,50 +1254,50 @@ calculation.
 
 
 Under the finite-measure hypotheses of Lemma~lem:bounded-costs, let old and
-new data differ only by coloring changes or deletion at a measurable set $E_0$,
-and fix a measurable real function injective on $E_0$ within each class,
-ordering $E_0$ by its values. At $uin E_0$ use new data at earlier vertices and
-old data at later vertices; let $delta d_u$ be the new-minus-old difference of
-the single-vertex deletion cost at $u$ in this environment (zero if $u$ absent).
+new data differ only by coloring changes or deletion at a measurable set E_0,
+and fix a measurable real function injective on E_0 within each class,
+ordering E_0 by its values. At uin E_0 use new data at earlier vertices and
+old data at later vertices; let delta d_u be the new-minus-old difference of
+the single-vertex deletion cost at u in this environment (zero if u absent).
 Then
 
  int_X(m_v^{new}-m_v^{old}),dnu(v)=int_{E_0}delta d_u,dnu(u).
 
-In particular deletion alone cannot increase the integral of $m_v$.
+In particular deletion alone cannot increase the integral of m_v.
 
 
 Proof. 
 An injective measurable point code supplies such an ordering if none is given.
-Start with a real polynomial $f$ for $x_+$, writing $m_v[f]=int f,drho_v$ at
-present vertices and zero at absent ones. For each root $v$ this depends only on
-a fixed finite-radius neighborhood; listing the finitely many $E_0$-vertices in
+Start with a real polynomial f for x_+, writing m_v[f]=int f,drho_v at
+present vertices and zero at absent ones. For each root v this depends only on
+a fixed finite-radius neighborhood; listing the finitely many E_0-vertices in
 that neighborhood in order and telescoping, each term agrees with the change
 produced at that vertex in the global environment (changes outside the
 neighborhood do not affect the polynomial rooted datum). No well-ordering or
 infinite telescoping at a single root is assumed. Integrate the finite local sum
 and apply (inv:mass-transport) to interchange root and changed vertex; for
-fixed $u$ the sum over affected roots equals the change in $d_u[f]$
+fixed u the sum over affected roots equals the change in d_u[f]
 (checkable on a large finite principal neighborhood via
 (inv:polynomial-increment), since the old and new deletion polynomials at
-$u$ agree; a deleted vertex has both new terms zero). Bounded degree, bounded
-support, and finite measure justify the integrals, giving the identity for $f$.
-Choose the polynomial approximants to $x_+$ from Lemma~lem:bounded-costs;
-uniform function approximation controls the left terms, the $L^1$ derivative
+u agree; a deleted vertex has both new terms zero). Bounded degree, bounded
+support, and finite measure justify the integrals, giving the identity for f.
+Choose the polynomial approximants to x_+ from Lemma~lem:bounded-costs;
+uniform function approximation controls the left terms, the L^1 derivative
 bound (inv:increment-approximation) the right increments, and finite
 measure passes to the limit, proving (inv:integral-change). A deletion has
-comparison increment $-d_ule0$ by interlacing, giving the last assertion.
+comparison increment -d_ule0 by interlacing, giving the last assertion.
 
 
 ## Deletion estimates independent of the largest weight
 
-Earlier arguments allow constants depending on $H_0$; removing weight cutoffs
-needs a fixed-band comparison uniform as $H_0$ grows. Cauchy transforms supply
+Earlier arguments allow constants depending on H_0; removing weight cutoffs
+needs a fixed-band comparison uniform as H_0 grows. Cauchy transforms supply
 it, and we must also control the conditional color choices of the single-site
-inequality. At an unassigned present $v$, let $F^0_{v,i}$ condition on color $i$
-with shift $t$, so the permanent option is
-$F_{v,i}(z)=F^0_{v,i}(z)+(beta-1)tW_v$. Define the auxiliary analytic field
+inequality. At an unassigned present v, let F^0_{v,i} condition on color i
+with shift t, so the permanent option is
+F_{v,i}(z)=F^0_{v,i}(z)+(beta-1)tW_v. Define the auxiliary analytic field
 
- J_i(v,z)=(F^0_{v,i}(z)-z-a_v)G_v(z),qquad c(v)=0, v present,
+ J_i(v,z)=(F^0_{v,i}(z)-z-a_v)G_v(z), c(v)=0, v present,
 
 zero elsewhere. Reciprocals are nonzero by Lemma~lem:bounded-costs, and
 the residue bound of Lemma~lem:rooted-locality makes these fields bounded
@@ -1313,79 +1305,79 @@ on upper-half-plane compacts when the root weight and neighbor ratios are
 bounded.
 
 
-On the above data, additionally assume neighbor weight ratios in $[K^{-1},K]$,
-$Kge1$. Fix $0<b_1<b_2<∞$ and the band $mathcal U=calU={b_1le W_vle b_2}$.
-Delete a measurable set $E_{hi}$ disjoint from $mathcal U$calU$ without
-changing colors at retained vertices, and let $delta_{tot}$ be the
-difference in either orientation. For each compact $Ω$ in the open upper
-half-plane there is $C$ such that for all $zinΩ$ and $1le ile r$,
+On the above data, additionally assume neighbor weight ratios in [K^{-1},K],
+Kge1. Fix 0<b_1<b_2<∞ and the band mathcal U=calU={b_1le W_vle b_2}.
+Delete a measurable set E_{hi} disjoint from mathcal UcalU without
+changing colors at retained vertices, and let delta_{tot} be the
+difference in either orientation. For each compact Ω in the open upper
+half-plane there is C such that for all zinΩ and 1le ile r,
 
  int_{mathcal U}calU|delta_{tot}G_v(z)|,dnu(v)
  +int_{mathcal UcalUcap{c=0}}|delta_{tot}J_i(v,z)|,dnu(v)
  le Cnu(E_{hi}),
 
 including only band vertices present before deletion. The constant may depend on
-$Ω,r,t,beta,D_0,K,b_1,b_2$ but not on $H_0$, total mass, coloring, graph,
+Ω,r,t,beta,D_0,K,b_1,b_2 but not on H_0, total mass, coloring, graph,
 or deleted set.
 
 
 Proof. 
-Discard initially absent vertices from $E_{hi}$ (configurations
+Discard initially absent vertices from E_{hi} (configurations
 unchanged, measure only smaller), so every deletion vertex is present first. We
 bound one deletion in a finite principal set, pass to infinite bounded graphs,
 then integrate using an analytic ordered-change identity.
 
-A single deletion and $G$. Fix finite $I$, $uin I$, $z$ with
-$y=Imz>0$. For $|lambda_v|le1$ perturb the diagonal variables to
-$z+zetalambda_v$; on $|zeta|<y/2$ they stay in the upper half-plane. By
-Lemma~lem:polynomial-quotients $P_I/P_{I∖{u}}$ has positive
+A single deletion and G. Fix finite I, uin I, z with
+y=Imz>0. For |lambda_v|le1 perturb the diagonal variables to
+z+zetalambda_v; on |zeta|<y/2 they stay in the upper half-plane. By
+Lemma~lem:polynomial-quotients P_I/P_{I∖{u}} has positive
 imaginary part, so its logarithm has a holomorphic branch with imaginary part in
-$(0,pi)$. A holomorphic function on a disk of radius $R$ with bounded imaginary
+(0,pi). A holomorphic function on a disk of radius R with bounded imaginary
 part has center derivative bounded by a numerical constant times the oscillation
-of that imaginary part over $R$ (recover the first Taylor coefficient from the
+of that imaginary part over R (recover the first Taylor coefficient from the
 first Fourier coefficient of the imaginary part on a concentric circle of radius
-$R/2$). Applying to the logarithm bounds its derivative at $0$ by $C_0/y$.
-Logarithmic differentiation of $P_I$ in a diagonal variable is
-$sum_vlambda_v(G_{I,v}-G_{I∖{u},v})$ (second transform zero at
-$v=u$); choosing phases to make summands nonnegative gives
+R/2). Applying to the logarithm bounds its derivative at 0 by C_0/y.
+Logarithmic differentiation of P_I in a diagonal variable is
+sum_vlambda_v(G_{I,v}-G_{I∖{u},v}) (second transform zero at
+v=u); choosing phases to make summands nonnegative gives
 
  sum_{vin I}|G_{I,v}(z)-G_{I∖{u},v}(z)|le C_0/y,
 
 with no weight bound.
 
 A single deletion and the conditional color field. Assume
-$unotinmathcal U$calU$ and take $lambda_v=0$ except at unassigned band vertices.
-Perturb the off-diagonal parts of both pair matrices of color $i$ by left and
-right diagonal scalings with entries $1+zetalambda_v$ (no conjugation on the
-right; the shift is unchanged), analytic but not Hermitian for complex $zeta$.
-With $D_lambda=diag(lambda_v)$ and $D_W=diag(√(W_v))$ on the relevant
+unotinmathcal UcalU and take lambda_v=0 except at unassigned band vertices.
+Perturb the off-diagonal parts of both pair matrices of color i by left and
+right diagonal scalings with entries 1+zetalambda_v (no conjugation on the
+right; the shift is unchanged), analytic but not Hermitian for complex zeta.
+With D_lambda=diag(lambda_v) and D_W=diag(√(W_v)) on the relevant
 allowed-site set, the off-diagonal matrix (sign suppressed) is
-$C_i=D_WTD_W$ and its perturbation is
-$zeta(D_lambda C_i+C_iD_lambda)+zeta^2D_lambda C_iD_lambda$. In the first
-term only band rows and their graph neighbors occur: row weights at most $b_2$,
-neighbor column weights at most $Kb_2$, so factoring diagonal weights and using
-$|T|le1$ gives $|D_lambda C_i|lesqrt Kb_2$ (same for $C_iD_lambda$); the
-quadratic term has both indices in the band, norm at most $b_2$. Hence the full
+C_i=D_WTD_W and its perturbation is
+zeta(D_lambda C_i+C_iD_lambda)+zeta^2D_lambda C_iD_lambda. In the first
+term only band rows and their graph neighbors occur: row weights at most b_2,
+neighbor column weights at most Kb_2, so factoring diagonal weights and using
+|T|le1 gives |D_lambda C_i|lesqrt Kb_2 (same for C_iD_lambda); the
+quadratic term has both indices in the band, norm at most b_2. Hence the full
 perturbation has norm at most
 
  2|zeta|sqrt Kb_2+|zeta|^2b_2.
 
-Choose a disk of radius depending only on $y,b_2,K$ making this less than $y/2$;
-then the perturbed pair matrices have imaginary part $le(y/2)I$, and with
-diagonal variable $z$ the non-Hermitian Lemma~lem:polynomial-quotients
-gives $P_I/P_{I∖{u}}$ positive imaginary part, its logarithm with
-imaginary part in $(0,pi)$, uniformly in $H_0$. The derivative of $log P_I$
+Choose a disk of radius depending only on y,b_2,K making this less than y/2;
+then the perturbed pair matrices have imaginary part le(y/2)I, and with
+diagonal variable z the non-Hermitian Lemma~lem:polynomial-quotients
+gives P_I/P_{I∖{u}} positive imaginary part, its logarithm with
+imaginary part in (0,pi), uniformly in H_0. The derivative of log P_I
 here is
 
  frac2rsum_{vin I}lambda_vJ_i(v,z).
 
 Indeed, in the permutation expansion of an outcome determinant, a nontrivial
-cycle through a color-$i$ target $v$ uses one row and one column factor, giving
-multiplier $2lambda_v$ (a fixed point contributes $z+a_v$, unchanged since $T$
-has zero diagonal); conditional on color $i$ at $v$ the sum of nontrivial terms
-through $v$ is $P_{I∖{v}}(z)(F^0_{v,i}(z)-z-a_v)$, with color
-probability $1/r$, so dividing by $P_I$ gives (inv:color-derivative),
-including both signs; also after deleting $u$. Applying the derivative bound and
+cycle through a color-i target v uses one row and one column factor, giving
+multiplier 2lambda_v (a fixed point contributes z+a_v, unchanged since T
+has zero diagonal); conditional on color i at v the sum of nontrivial terms
+through v is P_{I∖{v}}(z)(F^0_{v,i}(z)-z-a_v), with color
+probability 1/r, so dividing by P_I gives (inv:color-derivative),
+including both signs; also after deleting u. Applying the derivative bound and
 choosing phases as before,
 
  sum_{vin Icapmathcal UcalUcap{c=0}}|J_i^{I}(v,z)-J_i^{I∖{u}}(v,z)|
@@ -1394,7 +1386,7 @@ choosing phases as before,
 Constants in (inv:single-G),(inv:single-J) are uniform on
 upper-half-plane compacts.
 
-Passage to bounded infinite graphs. Exhaust the component of $u$ by
+Passage to bounded infinite graphs. Exhaust the component of u by
 finite principal sets; transforms converge by Lemma~lem:bounded-costs
 (with conditionals in (inv:J-definition)), and Fatou passes both
 single-deletion estimates to the infinite graph, over the whole class (no effect
@@ -1402,12 +1394,12 @@ on other components), valid in every comparison environment. These estimates are
 uniform in the largest weight.
 
 The analytic integral change identity. Fix measurable complex directions
-$lambda_v$, $|lambda_v|le1$, supported on the band targets ($mathcal U$calU$ for
-$G$, $mathcal UcalUcap{c=0}$ for $J_i$), and let $Z_v(z)$ be the respective
-field, extended by zero off them. Order $E_{hi}$ measurably and
-injectively along classes; at a deletion vertex $u$ let $delta_uZ_v$ be
+lambda_v, |lambda_v|le1, supported on the band targets (mathcal UcalU for
+G, mathcal UcalUcap{c=0} for J_i), and let Z_v(z) be the respective
+field, extended by zero off them. Order E_{hi} measurably and
+injectively along classes; at a deletion vertex u let delta_uZ_v be
 presence-minus-absence with earlier vertices deleted and later present. Orient
-$delta_{tot}$ before-minus-after. We claim
+delta_{tot} before-minus-after. We claim
 
  int_Xlambda_vdelta_{tot}Z_v(z),dnu(v)
  =int_{E_{hi}}(sum_{vin[u]_{mathcal R}calR}
@@ -1416,13 +1408,13 @@ $delta_{tot}$ before-minus-after. We claim
 Work near infinity. Every target is present in both configurations (deleted set
 disjoint from the band, initially absent band vertices excluded). At such a
 target, bounded support gives
-$G_v(z)=z^{-1}sum_{kge0}h_k(v)z^{-k}$, $|h_k(v)|le L^k$, and
-$F^0_{v,i}(z)-z-a_v=-sum_{kge0}b_k(v,i)z^{-k-1}$, $|b_k(v,i)|le
-B_{v,i}L^k$ where $B_{v,i}$ is the limiting residue mass, uniformly bounded by
+G_v(z)=z^{-1}sum_{kge0}h_k(v)z^{-k}, |h_k(v)|le L^k, and
+F^0_{v,i}(z)-z-a_v=-sum_{kge0}b_k(v,i)z^{-k-1}, |b_k(v,i)|le
+B_{v,i}L^k where B_{v,i} is the limiting residue mass, uniformly bounded by
 Lemma~lem:rooted-locality; these hold as the limits of negative Cauchy
-transforms of residue measures supported in $[-L,L]$. Each coefficient is
+transforms of residue measures supported in [-L,L]. Each coefficient is
 eventually fixed by a finite neighborhood, the bounds are uniform, and the series
-converge to the limiting transforms for large $|z|$; the product defining $J_i$
+converge to the limiting transforms for large |z|; the product defining J_i
 has exponentially bounded Laurent coefficients, and the coefficient at each order
 depends on a radius bounded by that order (Lemma~lem:rooted-locality plus
 series multiplication).
@@ -1431,26 +1423,26 @@ For each coefficient the ordered local telescoping of
 Lemma~lem:integral-change plus (inv:mass-transport) proves the
 coefficient version of (inv:analytic-change); bounded degree makes the
 number of affected roots at a fixed deleted vertex grow at most exponentially in
-the order, and the coefficient bounds give absolute summability for large $|z|$.
+the order, and the coefficient bounds give absolute summability for large |z|.
 Finite measure allows sum/integral interchange, proving
 (inv:analytic-change) on a nonempty open region (possibly depending on
-$H_0$).
+H_0).
 
 Both sides are holomorphic on the full upper half-plane. On the right, the sum
-over $v$ converges absolutely at each $z$ by the single-deletion estimates,
+over v converges absolutely at each z by the single-deletion estimates,
 its analytic partial sums (e.g. over increasing neighborhoods) are locally
 bounded by them, and normal-family convergence makes it holomorphic; the outer
-integral over the finite-measure $E_{hi}$ with uniform compact bounds is
-holomorphic. On the left, $|G_v(z)|le(Imz)^{-1}$ and the band
-residue bound similarly bounds $J_i$ on compacts; these and finite band measure
+integral over the finite-measure E_{hi} with uniform compact bounds is
+holomorphic. On the left, |G_v(z)|le(Imz)^{-1} and the band
+residue bound similarly bounds J_i on compacts; these and finite band measure
 justify holomorphy. The identity theorem extends (inv:analytic-change) to
 the whole upper half-plane.
 
-Finally fix $z$ and choose $lambda_v$ so that
-$lambda_vdelta_{tot}Z_v(z)=|delta_{tot}Z_v(z)|$ (arbitrary
+Finally fix z and choose lambda_v so that
+lambda_vdelta_{tot}Z_v(z)=|delta_{tot}Z_v(z)| (arbitrary
 when zero). These fixed directions give a right side bounded by the
-single-deletion constant times $nu(E_{hi})$. Apply separately to $G$
-and $J_i$ and add; the constants were independent of $H_0$ and of comparison
+single-deletion constant times nu(E_{hi}). Apply separately to G
+and J_i and add; the constants were independent of H_0 and of comparison
 environments, proving (inv:rank-comparison).
 
 
@@ -1474,47 +1466,47 @@ ball.
 
 ## The probability extension and the coloring theorem
 
-Let $(X,mu)$ be a standard probability space and $mathcal R$calR$ a countable
-nonsingular measured equivalence relation on $X$, with Borel representatives off
+Let (X,mu) be a standard probability space and mathcal RcalR a countable
+nonsingular measured equivalence relation on X, with Borel representatives off
 null sets. Write the Radon--Nikodym cocycle as
 
  mu(gamma S)=int_SDelta(gamma v,v),dmu(v)
 
-for measurable partial isomorphisms $gamma$ with graph in $mathcal R$.calR$. Versions
-satisfy the cocycle identity off a saturated null set (cover $mathcal R$calR$ by
+for measurable partial isomorphisms gamma with graph in mathcal R.calR. Versions
+satisfy the cocycle identity off a saturated null set (cover mathcal RcalR by
 countably many partial isomorphisms, use the chain rule for compositions, and
 agree on overlaps). Null sets along the countable relation are discarded with
 their saturations, which are null by nonsingularity.
 
-Let $G$ be an undirected measurable graph in $mathcal R$,calR$, loopless, degree at
-most $D_0<∞$, with radius-$m$ ball $B_G(v,m)$. Suppose
+Let G be an undirected measurable graph in mathcal R,calR, loopless, degree at
+most D_0<∞, with radius-m ball B_G(v,m). Suppose
 
- K^{-1}leDelta(u,v)le Kquadon every edge of G,qquad Kge1.
+ K^{-1}leDelta(u,v)le Kquadon every edge of G, Kge1.
 
-Consider a measurable field $T$ of Hermitian matrices on the orbit spaces
-$ell^2([v]_{mathcal R}calR)$, norm at most one, zero diagonal, supported on $G$.
+Consider a measurable field T of Hermitian matrices on the orbit spaces
+ell^2([v]_{mathcal R}calR), norm at most one, zero diagonal, supported on G.
 Changing the root conjugates them by a diagonal unitary, and all quantities used
 (principal norms, polynomial costs) are invariant.
 
-The relative Bernoulli extension of $(X,mu,mathcal R)$calR)$ is
+The relative Bernoulli extension of (X,mu,mathcal R)calR) is
 [
  widetilde X={(v,omega):vin X, omegain[0,1]^{[v]_{mathcal R}calR}},
  qquadpi(v,omega)=v,
 ]
-with probability $widetildemu$ whose conditional measure over $v$ is the
-product Lebesgue measure $kappa_v$. The extension identifies $(v,omega)$ with
-$(u,omega)$ when $(u,v)inmathcal R$,calR$, treating $omega$ as one function on
+with probability widetildemu whose conditional measure over v is the
+product Lebesgue measure kappa_v. The extension identifies (v,omega) with
+(u,omega) when (u,v)inmathcal R,calR, treating omega as one function on
 the class, so each extended orbit projects bijectively onto a base orbit; graph
 and matrices pull back. These are standard measurable spaces and fields
 (enumerate the relation and use one coordinate per site); changing the root
-reindexes a product family and preserves $kappa_v$. Lifts of base partial
+reindexes a product family and preserves kappa_v. Lifts of base partial
 isomorphisms have the same Radon--Nikodym derivative, enumerate the extended
 relation, and (by conditional integration) show the cocycle is the pullback of
-$Delta$. One Lebesgue coordinate at a site encodes countably many independent
+Delta. One Lebesgue coordinate at a site encodes countably many independent
 coordinates when needed.
 
-For a coloring $c:widetilde Xto{1,...,r}$, its value at a site $u$ in the
-orbit of $(v,omega)$ is $c(u,omega)$. On a finite principal set $I$ in this
+For a coloring c:widetilde Xto{1,...,r}, its value at a site u in the
+orbit of (v,omega) is c(u,omega). On a finite principal set I in this
 orbit let
 [
  T_{I,c}=sum_{i=1}^r1_{{c=i}cap I},T|_I,1_{{c=i}cap I}
@@ -1523,10 +1515,10 @@ retain entries joining equal-colored vertices.
 
 [Bernoulli coloring]
 
-Let $(X,mu,mathcal R)$,calR)$, $G$, $Delta$, $T$ satisfy the above assumptions.
-Fix an integer $rge1$ and $t>3/sqrt r$, and put $beta=80$. For every $mge1$
-and every $alpha,eta>0$ there is a measurable coloring
-$c:widetilde Xto{1,...,r}$ on the relative Bernoulli extension with
+Let (X,mu,mathcal R),calR), G, Delta, T satisfy the above assumptions.
+Fix an integer rge1 and t>3/sqrt r, and put beta=80. For every mge1
+and every alpha,eta>0 there is a measurable coloring
+c:widetilde Xto{1,...,r} on the relative Bernoulli extension with
 
  kappa_v{omega:c(v,omega)=i}=frac1rquad(iin{1,...,r}, a.e. v)
 
@@ -1537,67 +1529,67 @@ and
 
 
 We first build the cost, then prove the theorem. Until then the probability base
-may already include some Bernoulli coordinates; keep the notation $(X,mu)$,
-$mathcal R$calR$ for the current base, with the same bounds.
+may already include some Bernoulli coordinates; keep the notation (X,mu),
+mathcal RcalR for the current base, with the same bounds.
 
 ## Homogeneous costs from an invariant lift
 
-A partial coloring is $c:Xto{0,1,...,r}$ ($0$ unassigned). Recall the
+A partial coloring is c:Xto{0,1,...,r} (0 unassigned). Recall the
 finite-site data of Section~sec:polynomials: the shift is
 [
  tau_v=t,&c(v)=0, beta t,&c(v)>0,
 ]
-allowed choices are both signs of all $r$ colors at an unassigned site, or both
+allowed choices are both signs of all r colors at an unassigned site, or both
 signs of the assigned color otherwise, uniform and independent. With positive
-weights $W_v$, the outcome matrix on a choice block is the principal restriction
-of $diag(sqrt W)(pm T)diag(sqrt W)-diag(tau W)$. For the expected
-characteristic polynomial $P$ and deletion $Q=P_{-v}$ use
+weights W_v, the outcome matrix on a choice block is the principal restriction
+of diag(sqrt W)(pm T)diag(sqrt W)-diag(tau W). For the expected
+characteristic polynomial P and deletion Q=P_{-v} use
 [
- F_v=P/Q,quad G_v=1/F_v,quad m_v=int x_+rho_v(dx),quad d_v=s(P)-s(Q).
+ F_v=P/Q, G_v=1/F_v, m_v=int x_+rho_v(dx), d_v=s(P)-s(Q).
 ]
-At an unassigned site the superscript $i$ means permanent assignment to color
-$i$ (including the increased shift); $F^0_{v,i}$ conditions on color $i$ without
+At an unassigned site the superscript i means permanent assignment to color
+i (including the increased shift); F^0_{v,i} conditions on color i without
 increasing the shift, and
 [
  F_{v,i}=F^0_{v,i}+(beta-1)tW_v
 ]
-includes it; $d_v^{,i}=0$ on assigned sites. On bounded invariant graphs the
+includes it; d_v^{,i}=0 on assigned sites. On bounded invariant graphs the
 same notation refers to limits from Lemma~lem:bounded-costs.
 
 The original measure need not be invariant, so the integral-change formula does
 not yet apply. Use the Maharam invariant lift [cite]; for the
 multiplicative-coordinate formula see cite[Section~2, before Theorem~2.1]{RoySamorodnitsky06}.
-Take logarithmic height $sinmathbb R$bbR$ with multiplicative coordinate $y=e^s$:
+Take logarithmic height sinmathbb RbbR with multiplicative coordinate y=e^s:
 
- dnu(v,s)=dmu(v)e^s,ds,qquad W(v,s)=e^{-s}.
+ dnu(v,s)=dmu(v)e^s,ds, W(v,s)=e^{-s}.
 
-In the lifted relation the site of $u$ at root $(v,s)$ is
+In the lifted relation the site of u at root (v,s) is
 
  (u,s-logDelta(u,v)).
 
-The measure $nu$ is invariant: the factor $Delta(u,v)$ in
-(eq:coloring-cocycle) cancels $e^{-logDelta(u,v)}$ from the height
-density. The destination weight is $Delta(u,v)W(v,s)$, so neighboring weights
-have ratios in $[K^{-1},K]$; coloring and matrix data are independent of height.
+The measure nu is invariant: the factor Delta(u,v) in
+(eq:coloring-cocycle) cancels e^{-logDelta(u,v)} from the height
+density. The destination weight is Delta(u,v)W(v,s), so neighboring weights
+have ratios in [K^{-1},K]; coloring and matrix data are independent of height.
 
-For $0<l<H<∞$ retain sites $lle Wle H$ with the restricted relation,
+For 0<l<H<∞ retain sites lle Wle H with the restricted relation,
 graph, and principal data. The measure is finite, since
-$nu{lle Wle H}=l^{-1}-H^{-1}$, so Section~sec:invariant applies. Write
-the cut costs $m^{l,H},d^{l,H},(d^i)^{l,H}$.
+nu{lle Wle H}=l^{-1}-H^{-1}, so Section~sec:invariant applies. Write
+the cut costs m^{l,H},d^{l,H},(d^i)^{l,H}.
 
 [Homogeneous costs]
 
-Fix $D_0,K,r,t$, $beta=80$, and the above data and partial coloring. For each
-field $f$ in $m,d,d^1,...,d^r$ there is a bounded measurable $barbar f$f$ on $X$
+Fix D_0,K,r,t, beta=80, and the above data and partial coloring. For each
+field f in m,d,d^1,...,d^r there is a bounded measurable barbar ff on X
 such that on every band
-$mathcal U_calU_{b_1,b_2}={(v,s):b_1le W(v,s)le b_2}$, $0<b_1<b_2<∞$,
+mathcal U_calU_{b_1,b_2}={(v,s):b_1le W(v,s)le b_2}, 0<b_1<b_2<∞,
 
  frac{f^{l,H}}{W}→barbar f(v)f(v)
  quadin L^1(mathcal U_calU_{b_1,b_2},nu)
  quadjointly as l↓0, H↑∞.
 
 The errors vanish uniformly over all such bases and configurations with the
-fixed parameters. There is $C_0$ depending only on those parameters bounding all
+fixed parameters. There is C_0 depending only on those parameters bounding all
 normalized cutoff fields on retained sites and all limits; the limits are
 independent of height and satisfy
 
@@ -1607,103 +1599,103 @@ independent of height and satisfy
 
 
 Proof. 
-The two cutoffs use different controls: weights above $H$ have invariant measure
-$H^{-1}$ (rank comparison controls the upper end); with the upper cutoff fixed,
+The two cutoffs use different controls: weights above H have invariant measure
+H^{-1} (rank comparison controls the upper end); with the upper cutoff fixed,
 the edge-ratio bound makes sufficiently small weights invisible in any chosen
-finite neighborhood of a fixed band. Prove uniform convergence on $1le Wle e$
-(the same handles any band bounded away from zero). Keep $l$ fixed and compare
-$H_2ge H_1ge e$; passing from $H_2$ to $H_1$ deletes invariant measure at most
-$1/H_1$, and Lemma~lem:rank-comparison gives, on band sites, integrated
-absolute differences of order $1/H_1$ for
+finite neighborhood of a fixed band. Prove uniform convergence on 1le Wle e
+(the same handles any band bounded away from zero). Keep l fixed and compare
+H_2ge H_1ge e; passing from H_2 to H_1 deletes invariant measure at most
+1/H_1, and Lemma~lem:rank-comparison gives, on band sites, integrated
+absolute differences of order 1/H_1 for
 [
  G_v(z)quadandquad
- J_i(v,z)=(F^0_{v,i}(z)-z-a_v)G_v(z),quad a_v=tau_vW_v,
+ J_i(v,z)=(F^0_{v,i}(z)-z-a_v)G_v(z), a_v=tau_vW_v,
 ]
-targeting unassigned band sites for $J_i$, with constants uniform in
-$l,H_1,H_2$, the configuration, and $z$ on upper-half-plane compacts.
+targeting unassigned band sites for J_i, with constants uniform in
+l,H_1,H_2, the configuration, and z on upper-half-plane compacts.
 
 The second-moment estimate of Lemma~lem:rooted-locality,
 
  int x^2rho_v(dx)le((beta t)^2+K)W_v^2,
 
 also holds for the altered site choices, so these moments are uniformly bounded
-on the band. They give a uniform positive lower bound for $|G_v(z)|$ on compacts:
+on the band. They give a uniform positive lower bound for |G_v(z)| on compacts:
 a fixed bounded interval carries uniformly positive mass and
-$-ImG_v(z)=(Imz)int|z-x|^{-2}rho_v(dx)$. The
+-ImG_v(z)=(Imz)int|z-x|^{-2}rho_v(dx). The
 residue bound of Lemmas~lem:polynomial-quotients,lem:rooted-locality
-bounds $J_i$ on compacts, so division is controlled in
+bounds J_i on compacts, so division is controlled in
 [
- F_v=1/G_v,qquad F^0_{v,i}=z+a_v+J_i/G_v,qquad
+ F_v=1/G_v, F^0_{v,i}=z+a_v+J_i/G_v,
  F_{v,i}=F^0_{v,i}+(beta-1)tW_v.
 ]
 The integrated differences of these quotients thus tend to zero as
-$H_1to∞$, uniformly in $l$.
+H_1to∞, uniformly in l.
 
 Recover costs from transform comparisons. Probabilities with the uniform
-second-moment bound are weakly compact, and $int x_+$ is continuous there by
+second-moment bound are weakly compact, and int x_+ is continuous there by
 uniform integrability; Cauchy transforms on a countable dense subset separate
-them. So a prescribed accuracy of $m_v$ follows from finitely many transform
-comparisons to sufficient accuracy, giving the integrated comparison for $m$.
+them. So a prescribed accuracy of m_v follows from finitely many transform
+comparisons to sufficient accuracy, giving the integrated comparison for m.
 
-For $d$ use the argument of the quotient. In a finite restriction let $n_P(x),
-n_Q(x)$ count roots above $x$ with multiplicity. Interlacing and quotient
-properties give, for $y>0$,
+For d use the argument of the quotient. In a finite restriction let n_P(x),
+n_Q(x) count roots above x with multiplicity. Interlacing and quotient
+properties give, for y>0,
 
  arg F(x+iy)=int_{mathbb R}bbR(y)/((x-u)^2+y^2)(n_P-n_Q)(u),du,
- qquad 0learg Flepi,
+  0learg Flepi,
 
 by writing the arguments of the root factors (branch fixed by
-$ImF>0$). Here $0le n_P-n_Qle1$, and
+ImF>0). Here 0le n_P-n_Qle1, and
 Lemma~lem:positive-cost with (eq:coloring-second-moment) gives
 
- int_B^∞(n_P-n_Q)(x),dx=O(B^{-1})quad(Bto∞),
+ int_B^∞(n_P-n_Q)(x),dx=O(B^{-1})(Bto∞),
 
 uniformly on the band (including shifted options): the bound
-$2int(x-B)_+rho_v+(a_v+B)rho_v{x>B}$ is controlled by the second moment and
-the band bound on $a_v$. On $[0,B]$,
+2int(x-B)_+rho_v+(a_v+B)rho_v{x>B} is controlled by the second moment and
+the band bound on a_v. On [0,B],
 
  frac1piint_0^Barg F(x+iy),dxquadas y↓0
 
 approximates the integral of the spectral-shift function uniformly: with
-$p_y(u)=y/(pi(u^2+y^2))$ and
-$delta_B(y)=|p_y*1_{[0,B]}-1_{[0,B]}|_{L^1(mathbb R)}bbR)to0$ for fixed $B$, the
-$L^1$ smoothing error plus (eq:coloring-shift-tail) give, uniformly,
+p_y(u)=y/(pi(u^2+y^2)) and
+delta_B(y)=|p_y*1_{[0,B]}-1_{[0,B]}|_{L^1(mathbb R)}bbR)to0 for fixed B, the
+L^1 smoothing error plus (eq:coloring-shift-tail) give, uniformly,
 
  |d_v-frac1piint_0^Barg F_v(x+iy),dx|lefrac CB+delta_B(y),
- quad Bge1,
+  Bge1,
 
-and the same for shifted options. For fixed $B,y$, convergence of costs and
+and the same for shifted options. For fixed B,y, convergence of costs and
 locally uniform quotient convergence (Lemma~lem:bounded-costs) pass this
 to present band vertices of a bounded infinite cut and to present unassigned band
-vertices for the options; $ImFge y$ keeps the argument
-continuous. First choose $B$ large, then $y$ small, then $H_1$ large for the
-fixed segment $[0,B]+iy$, proving the high-cutoff comparison.
+vertices for the options; ImFge y keeps the argument
+continuous. First choose B large, then y small, then H_1 large for the
+fixed segment [0,B]+iy, proving the high-cutoff comparison.
 
 Remove the lower cutoff. For fixed finite upper cutoff,
 Lemma~lem:bounded-costs approximates costs uniformly by bounded
-neighborhoods; a radius-$R$ neighborhood of a fixed-band root cannot contain a
-vertex of weight below $K^{-R}$, so a small enough lower cutoff is invisible to
+neighborhoods; a radius-R neighborhood of a fixed-band root cannot contain a
+vertex of weight below K^{-R}, so a small enough lower cutoff is invisible to
 each fixed neighborhood calculation. This proves uniform convergence as
-$l↓0$ with that upper cutoff fixed. For joint convergence, compare both
+l↓0 with that upper cutoff fixed. For joint convergence, compare both
 large upper cutoffs to one fixed large cutoff by the preceding estimates, then
-compare lower cutoffs there, giving uniform joint $L^1$-Cauchy estimates;
-dividing by $W$ preserves convergence on a band.
+compare lower cutoffs there, giving uniform joint L^1-Cauchy estimates;
+dividing by W preserves convergence on a band.
 
 The uniform bound follows from (eq:coloring-second-moment) and
-Lemma~lem:positive-cost: $m_v/W_v$ is uniformly bounded and
-$d_vle2m_v+a_v$, likewise for the options, independent of band, cutoff, and
+Lemma~lem:positive-cost: m_v/W_v is uniformly bounded and
+d_vle2m_v+a_v, likewise for the options, independent of band, cutoff, and
 configuration.
 
-Identify the height dependence. Translating height by real $b$ scales weights by
-$e^{-b}$ and shifts cutoffs correspondingly; all matrices and costs scale by the
+Identify the height dependence. Translating height by real b scales weights by
+e^{-b} and shifts cutoffs correspondingly; all matrices and costs scale by the
 same positive factor, so
 [
  frac{f^{l,H}(v,s+b)}{W(v,s+b)}=frac{f^{e^bl,e^bH}(v,s)}{W(v,s)}.
 ]
 Joint convergence makes the limit invariant a.e. under every rational height
-translation simultaneously; a measurable function on $mathbb R$bbR$ invariant a.e.
+translation simultaneously; a measurable function on mathbb RbbR invariant a.e.
 under all rational translations is a.e. constant (by convolution). Applying on
-almost every base fiber gives $barbar f(v)$,f(v)$, independent of height; measurability
+almost every base fiber gives barbar f(v),f(v), independent of height; measurability
 follows from the band limits (e.g. integrate a representative over a fixed
 height interval).
 
@@ -1715,16 +1707,16 @@ list of normalized fields; dividing by the positive root weight yields
 
 
 Two compatibility properties follow. First, for a standard nonsingular
-extension $(widehatwidehat X,X,widehatmu,mathcal Rhat)$calRhat)$ with $pi:widehatwidehat
-XXto X$, $pi_*widehatmu=mu$, each class projecting bijectively onto a base
-class, and cocycle $widehatDelta(widehatwidehat u,u,widehatwidehat v)=v)=Delta(piwidehatwidehat
-u,u,piwidehatwidehat v)$:v)$: pulling back the data (including unused Bernoulli
-coordinates), every bounded-cut ball is exactly a pullback, band $L^1$ errors
+extension (widehatwidehat X,X,widehatmu,mathcal Rhat)calRhat) with pi:widehatwidehat
+XXto X, pi_*widehatmu=mu, each class projecting bijectively onto a base
+class, and cocycle widehatDelta(widehatwidehat u,u,widehatwidehat v)=v)=Delta(piwidehatwidehat
+u,u,piwidehatwidehat v):v): pulling back the data (including unused Bernoulli
+coordinates), every bounded-cut ball is exactly a pullback, band L^1 errors
 agree, and uniqueness of limits gives the same pullback identity a.e. for each
-normalized limit field. Second, for a color permutation $sigma$ (with
-$sigma(0)=0$), relabeling the finite polynomial laws gives
+normalized limit field. Second, for a color permutation sigma (with
+sigma(0)=0), relabeling the finite polynomial laws gives
 [
- barbar m(m(sigmacirc c)=barbar m(c),m(c),quad barbar d(d(sigmacirc c)=barbar d(c),d(c),quad
+ barbar m(m(sigmacirc c)=barbar m(c),m(c), barbar d(d(sigmacirc c)=barbar d(c),d(c),
  barbar d^d^{i}(sigmacirc c)=barbar d^d^{sigma^{-1}(i)}(c)
 ]
 a.e.; the last records the renaming of virtual options.
@@ -1734,41 +1726,41 @@ Define the nonnegative total cost
  mathcal S(c)=calS(c)=int_Xbarbar m(v)m(v),dmu(v),
 
 uniformly bounded by the above. The next lemma records strip averaging and
-locality, which make $mathcal S$calS$ usable.
+locality, which make mathcal ScalS usable.
 
 [Strip averages and locality]
 
 Under Lemma~lem:homogeneous-costs, put
-$Σ_n={e^{-n}le Wle e^n}$ and $f_{Σ_n}=f^{e^{-n},e^n}$. For every
-$f$ in $m,d,d^1,...,d^r$,
+Σ_n={e^{-n}le Wle e^n} and f_{Σ_n}=f^{e^{-n},e^n}. For every
+f in m,d,d^1,...,d^r,
 
  frac1{2n}int_{Σ_n}|f_{Σ_n}-Wbarbar f|f|,dnu→0
- quad(nto∞),
+ (nto∞),
 
 uniformly over configurations and bases with the fixed parameters. If
-$c_1,c_2$ are partial colorings on the same base with the same graph, matrix,
-and cocycle data, and $E_R$ is where they agree on the radius-$R$ ball, then
+c_1,c_2 are partial colorings on the same base with the same graph, matrix,
+and cocycle data, and E_R is where they agree on the radius-R ball, then
 
  int_{E_R}|barbar f(c_1)-f(c_1)-barbar f(c_2)|f(c_2)|,dmulegamma(R),qquadgamma(R)to0,
 
-with one $gamma$ for the finite list of fields, uniform over configurations.
+with one gamma for the finite list of fields, uniform over configurations.
 Both hold after taking a product with an auxiliary probability space keeping the
 auxiliary parameter fixed along the relation; the cost fields are measurable
 there.
 
 
 Proof. 
-For the strip integral, $|f_{Σ_n}-Wbarbar f|f|,dnu=|f_{Σ_n}/W-barbar
-f|f|,dmu,ds$. Divide $[-n,n]$ into unit intervals; translating each to a fixed
+For the strip integral, |f_{Σ_n}-Wbarbar f|f|,dnu=|f_{Σ_n}/W-barbar
+f|f|,dmu,ds. Divide [-n,n] into unit intervals; translating each to a fixed
 interval uses homogeneity from Lemma~lem:homogeneous-costs. Intervals a
 large fixed distance from both ends have translated cutoffs uniformly near zero
 and infinity, so uniform joint convergence on the fixed band bounds their
 average error by any positive number; the bounded number of end intervals has
-uniformly bounded error, and its contribution over $2n$ tends to zero, proving
+uniformly bounded error, and its contribution over 2n tends to zero, proving
 (eq:coloring-strip).
 
 For (eq:coloring-locality), approximate both limiting normalized fields
-on a fixed unit height band in $L^1$ by common fixed lower and upper cutoffs
+on a fixed unit height band in L^1 by common fixed lower and upper cutoffs
 (errors uniform by Lemma~lem:homogeneous-costs). On this cut,
 Lemma~lem:bounded-costs approximates each field uniformly by a finite-radius
 calculation, and agreement of the base colorings on a large enough ball makes
@@ -1779,7 +1771,7 @@ then radius for each accuracy proves (eq:coloring-locality).
 Finally, bounded-cut calculations are measurable (limits of rooted finite
 calculations), remain so with measurable parameters, and an auxiliary parameter
 fixed along the relation changes no degree, edge ratio, or spectral estimate; the
-uniform estimates hold on its product probability space, and the $L^1$ limit
+uniform estimates hold on its product probability space, and the L^1 limit
 gives jointly measurable limit fields there.
 
 
@@ -1792,11 +1784,11 @@ invariant lift sums this without assuming invariance of the original measure.
 [Initial cost and bad balls]
 
 Assume the hypotheses and parameters of Theorem~thm:bernoulli-coloring. The
-all-unassigned coloring has $mathcal S=0$.calS=0$. Fix $mge1$, $alpha>0$. For a
-measurable partial coloring $c$ let $Bad_m(c)$ be the set of $v$
-where every vertex of $B_G(v,m)$ is assigned and
-$|T_{B_G(v,m),c}|>2beta t+alpha$. Then there is $C_4$ depending only on
-$m,D_0,K,alpha$ with
+all-unassigned coloring has mathcal S=0.calS=0. Fix mge1, alpha>0. For a
+measurable partial coloring c let Bad_m(c) be the set of v
+where every vertex of B_G(v,m) is assigned and
+|T_{B_G(v,m),c}|>2beta t+alpha. Then there is C_4 depending only on
+m,D_0,K,alpha with
 
  mu(Bad_m(c))le C_4mathcal S(c).calS(c).
 
@@ -1806,25 +1798,25 @@ The same holds on probability extensions with pulled-back data.
 Proof. 
 For the all-unassigned coloring Lemma~lem:finite-cost-endpoints gives zero
 cost on every finite principal restriction; finite restriction limits and
-Lemma~lem:homogeneous-costs give $mathcal S=0$.calS=0$.
+Lemma~lem:homogeneous-costs give mathcal S=0.calS=0.
 
 Partition bad centers into a bounded number of measurable classes with pairwise
-graph distance greater than $2m+1$. Construction: a bounded-degree Borel graph
+graph distance greater than 2m+1. Construction: a bounded-degree Borel graph
 has a countable measurable partition into independent sets (assign to each vertex
 a finite prefix of an injective standard-space code long enough to separate it
 from its neighbors, including the prefix length; equal codes are not adjacent);
 visiting these in sequence and giving each vertex the first color not used by an
 earlier-colored neighbor from a palette of size one more than the degree bound
 gives a measurable proper coloring. Apply to the graph joining vertices at
-distance at most $2m+1$, and intersect its color classes with
-$Bad_m(c)$; there are boundedly many in terms of $D_0,m$.
+distance at most 2m+1, and intersect its color classes with
+Bad_m(c); there are boundedly many in terms of D_0,m.
 
-Fix one class $C$. In the lift retain $Σ_n$ and select lifted centers above
-$C$ whose height is at least $mlog K$ from both ends of $[-n,n]$; each radius-$m$
+Fix one class C. In the lift retain Σ_n and select lifted centers above
+C whose height is at least mlog K from both ends of [-n,n]; each radius-m
 ball lies in the strip, and the balls are disjoint and nonadjacent along each
 orbit ((eq:coloring-height-change) plus the cocycle law give exactly one
 lifted site above each base site in a fixed orbit, and distinct centers project
-to distinct centers of $C$ at distance $>2m+1$). Delete every strip site outside
+to distinct centers of C at distance >2m+1). Delete every strip site outside
 these balls; by Lemma~lem:integral-change deletion cannot increase total
 integrated cost, so
 
@@ -1832,19 +1824,19 @@ integrated cost, so
 
 Each retained component is one selected finite ball; invariant mass transport
 sends rooted costs to its center, so the right side equals the integral over
-selected lifted centers of $s(P_{ball})$.
+selected lifted centers of s(P_{ball}).
 
 For a bad ball, Lemma~lem:finite-cost-endpoints gives
-$s(P_{ball})ge(alpha/2)min_{ball}W$, and the edge-ratio
+s(P_{ball})ge(alpha/2)min_{ball}W, and the edge-ratio
 bound gives
 
  s(P_{ball})gefracalpha2K^{-m}W_{center}.
 
-Divide (eq:coloring-ball-deletion) by $2n$, let $nto∞$:
-(eq:coloring-strip) makes the left side tend to $mathcal S(c)$,calS(c)$, while on
-the right $W,dnu=dmu,ds$ and the allowed height interval has length
-$2n-2mlog K$, so (eq:coloring-ball-root-cost) gives
-$(alpha/2)K^{-m}mu(C)lemathcal S(c)$.calS(c)$. Summing over the bounded number of
+Divide (eq:coloring-ball-deletion) by 2n, let nto∞:
+(eq:coloring-strip) makes the left side tend to mathcal S(c),calS(c), while on
+the right W,dnu=dmu,ds and the allowed height interval has length
+2n-2mlog K, so (eq:coloring-ball-root-cost) gives
+(alpha/2)K^{-m}mu(C)lemathcal S(c).calS(c). Summing over the bounded number of
 classes proves (eq:coloring-bad-balls). Every step uses only the stated
 probability-space, cocycle, degree, and matrix bounds, so it applies to the
 extensions.
@@ -1852,73 +1844,73 @@ extensions.
 
 ## Assigning a small fraction of the remaining vertices
 
-At a current partial coloring $c$ with $E_0={c=0}$ determined by coordinates
-already used, choose at each $vin E_0$ a color $i(v)$ minimizing $barbar d^i(v)$,d^i(v)$,
+At a current partial coloring c with E_0={c=0} determined by coordinates
+already used, choose at each vin E_0 a color i(v) minimizing barbar d^i(v),d^i(v),
 breaking ties by a fresh uniform permutation of the colors (taking the first
-minimizer in the permutation). The costs of $c$ are defined first on the old
+minimizer in the permutation). The costs of c are defined first on the old
 coordinates then pulled back to the tie coordinates, and bounded-cut definitions
-and limits commute with this pullback, so $i(v)$ uses only old and tie
+and limits commute with this pullback, so i(v) uses only old and tie
 coordinates.
 
-Add fresh uniform times $U_vin(0,1)$ independent of old and tie coordinates.
-For $0letheta<1$ define $c^theta$ by assigning $i(v)$ at each $vin E_0$ with
-$U_v<theta$, leaving other sites as in $c$. This is measurable on the relative
+Add fresh uniform times U_vin(0,1) independent of old and tie coordinates.
+For 0letheta<1 define c^theta by assigning i(v) at each vin E_0 with
+U_v<theta, leaving other sites as in c. This is measurable on the relative
 Bernoulli extension; null exceptions are filled by measurable defaults before
 adding the next independent coordinates, with negligible saturations.
 
 [One assignment step]
 
-Under Theorem~thm:bernoulli-coloring, let $c$ be a partial coloring
-determined before the fresh times, with $i(v)$ and $c^theta$ as above. There is
-a nonnegative $g$, depending only on the fixed graph and polynomial parameters,
-with $g(h)to0$ as $h↓0$ and, for every $0<h<1/2$,
+Under Theorem~thm:bernoulli-coloring, let c be a partial coloring
+determined before the fresh times, with i(v) and c^theta as above. There is
+a nonnegative g, depending only on the fixed graph and polynomial parameters,
+with g(h)to0 as h↓0 and, for every 0<h<1/2,
 
  mathcal S(c^h)calS(c^h)le(1+24h)mathcal S(c)+hg(h),calS(c)+hg(h),
 
-the same $g$ working for all steps and current configurations.
+the same g working for all steps and current configurations.
 
 
 Proof. 
-Apply Lemma~lem:integral-change to $cto c^h$ on the lift restricted to
-$Σ_n$, ordering changed sites by $U_v$ (a.e. distinct along every countable
+Apply Lemma~lem:integral-change to cto c^h on the lift restricted to
+Σ_n, ordering changed sites by U_v (a.e. distinct along every countable
 orbit off a saturated null set; the order need not be a well-order, as allowed).
-At a comparison site $v$ the pre-change configuration is $c^{U_v}$ with $U_v$
+At a comparison site v the pre-change configuration is c^{U_v} with U_v
 fixed as a threshold along the orbit.
 
 We claim
 
  
  &int_{Σ_n}(m_{Σ_n}(c^h)-m_{Σ_n}(c)),dnu
- &quad=int_0^h(dtheta)/(1-theta)
+ &=int_0^h(dtheta)/(1-theta)
    int_{Σ_n}1_{{vin E_0,U_vgetheta}}
    [d_{Σ_n}^{i(v)}(c^theta)-d_{Σ_n}(c^theta)],dnu,
  
  
 
-$nu$ including label laws and the height measure, $v$ the base site of the lifted
-root. To justify, condition at a root in $E_0$ on old coordinates, all tie
+nu including label laws and the height measure, v the base site of the lifted
+root. To justify, condition at a root in E_0 on old coordinates, all tie
 coordinates, every fresh time except this root's, and the height. For a fixed
-threshold $theta$, leave the root unassigned and call the bracket
-$D(theta)$; whenever the root time $u_0getheta$, $D$ is independent of its
+threshold theta, leave the root unassigned and call the bracket
+D(theta); whenever the root time u_0getheta, D is independent of its
 value (chosen colors precede the fresh times, other sites use their own times,
 and neither the graph nor the restricted weight data use the new root
 coordinate), holding for deterministic rooted finite-subset calculations and
-hence for bounded-strip costs; the same at $u_0=theta$ since the update is
-strict. When $u_0<h$ the ordered comparison increment at the root is $D(u_0)$
+hence for bounded-strip costs; the same at u_0=theta since the update is
+strict. When u_0<h the ordered comparison increment at the root is D(u_0)
 outside the null event of a tie, and the last uniform time gives
-$int_0^hD(u_0),du_0=int_0^h(dtheta)/(1-theta)int_theta^1D(theta),du_0$.
+int_0^hD(u_0),du_0=int_0^h(dtheta)/(1-theta)int_theta^1D(theta),du_0.
 The bounded-strip computations are measurable jointly in threshold and data, so
 Fubini (first a.e. root and height, then the rest) proves
 (eq:coloring-finite-strip-step), before choosing any homogeneous limit
-version depending on $theta$.
+version depending on theta.
 
-Pass to that limit in an integrated expression. Divide by $2n$ and use
-Lemma~lem:strip-locality. On the right adjoin $theta$ with normalized
-Lebesgue measure on $[0,h]$, fixed along the relation, and regard $c^theta$ as
+Pass to that limit in an integrated expression. Divide by 2n and use
+Lemma~lem:strip-locality. On the right adjoin theta with normalized
+Lebesgue measure on [0,h], fixed along the relation, and regard c^theta as
 one configuration on the augmented base; the lemma applies uniformly there, its
 absolute errors controlling those after the mask, after the finite selection
-$i(v)$, and after the factor $(1-theta)^{-1}le2$. For a height-independent
-limit field, $W$ cancels the height density leaving an interval of length $2n$,
+i(v), and after the factor (1-theta)^{-1}le2. For a height-independent
+limit field, W cancels the height density leaving an interval of length 2n,
 so
 
  
@@ -1931,32 +1923,32 @@ so
 All fields have jointly measurable versions on the augmented space; no
 evaluation of an unspecified limit at the random diagonal is used.
 
-Compare the intermediate costs with the old ones. Let $D_R$ bound the cardinality
-of a radius-$R$ ball. On the event that this ball has no update $cto
-c^theta$, the colorings agree there; (eq:coloring-locality) controls the
+Compare the intermediate costs with the old ones. Let D_R bound the cardinality
+of a radius-R ball. On the event that this ball has no update cto
+c^theta, the colorings agree there; (eq:coloring-locality) controls the
 integrated difference of their cost fields on that event, and summing over the
 finite option list controls the selected field. Conditional on old and tie
-coordinates, the probability of an update in the ball is at most $D_Rh$ (union
+coordinates, the probability of an update in the ball is at most D_Rh (union
 bound); on the update event the cost differences are uniformly bounded by
-Lemma~lem:homogeneous-costs. These hold for fixed $theta$ or jointly with
-normalized measure on $[0,h]$, so replacing the two costs in
-(eq:coloring-limit-step) by their values at $c$ creates absolute error at
+Lemma~lem:homogeneous-costs. These hold for fixed theta or jointly with
+normalized measure on [0,h], so replacing the two costs in
+(eq:coloring-limit-step) by their values at c creates absolute error at
 most
 
  Ch(gamma(R)+D_Rh),
 
-$C$ depending only on the fixed parameters, the factor $(1-theta)^{-1}le2$.
+C depending only on the fixed parameters, the factor (1-theta)^{-1}le2.
 
-The old costs and the minimizing $i(v)$ are independent of the fresh root time;
-the survival probability $1-theta$ cancels the denominator. By
+The old costs and the minimizing i(v) are independent of the fresh root time;
+the survival probability 1-theta cancels the denominator. By
 (eq:coloring-limit-assignment) the old-cost contribution is at most
 [
  hint_{E_0}(barbar d^d^{i(v)}(c)-barbar d(c))d(c)),dmu(v)
  le24hint_{E_0}barbar m(c)m(c),dmule24hmathcal S(c).calS(c).
 ]
-For each accuracy in (eq:coloring-step-error), take $R$ large to control
-$gamma(R)$ then $h$ small to control $D_Rh$; the infimum over integer $R$ gives
-a nonnegative $g(h)to0$, proving (eq:coloring-step-bound) uniformly.
+For each accuracy in (eq:coloring-step-error), take R large to control
+gamma(R) then h small to control D_Rh; the infimum over integer R gives
+a nonnegative g(h)to0, proving (eq:coloring-step-bound) uniformly.
 
 
 ## Completion of the coloring
@@ -1964,49 +1956,49 @@ a nonnegative $g(h)to0$, proving (eq:coloring-step-bound) uniformly.
 Proof. [Proof of Theorem~thm:bernoulli-coloring]
 Work on the relative Bernoulli extension of the original base, reserving
 independent coordinates at each site for each step. Start from all-unassigned
-(cost zero by Lemma~lem:bad-balls). Choose $T_0>0$, $hin(0,1/2)$, and apply
-Lemma~lem:hazard-step with fresh coordinates for $N=lceil T_0/hrceil$
-steps. With $S_j$ the cost after $j$ steps, iteration gives
+(cost zero by Lemma~lem:bad-balls). Choose T_0>0, hin(0,1/2), and apply
+Lemma~lem:hazard-step with fresh coordinates for N=lceil T_0/hrceil
+steps. With S_j the cost after j steps, iteration gives
 
  S_Nle(g(h))/(24)((1+24h)^N-1),
 
-tending to $0$ as $h↓0$ for fixed $T_0$ (since
-$(1+24h)^Nleexp(24(T_0+h))$).
+tending to 0 as h↓0 for fixed T_0 (since
+(1+24h)^Nleexp(24(T_0+h))).
 
 An unassigned site receives a color at its first successful hazard; conditional
-on the base point, the probability of remaining unassigned after $N$ steps is
-$(1-h)^Nle e^{-T_0}$, independent of the selected color. If $D_m$ bounds a
-radius-$m$ ball, the conditional probability that any site of it stays unassigned
-is at most $D_me^{-T_0}$. The unconditional probability that the ball is fully
-assigned but violates the norm bound is at most $C_4S_N$ by
-Lemma~lem:bad-balls. Take $T_0$ large then $h$ small so
-$D_me^{-T_0}+C_4S_N<eta$. Complete the partial coloring by fresh independent
+on the base point, the probability of remaining unassigned after N steps is
+(1-h)^Nle e^{-T_0}, independent of the selected color. If D_m bounds a
+radius-m ball, the conditional probability that any site of it stays unassigned
+is at most D_me^{-T_0}. The unconditional probability that the ball is fully
+assigned but violates the norm bound is at most C_4S_N by
+Lemma~lem:bad-balls. Take T_0 large then h small so
+D_me^{-T_0}+C_4S_N<eta. Complete the partial coloring by fresh independent
 uniform colors at remaining sites; every previously complete ball is unchanged,
 so (eq:coloring-ball-conclusion) holds.
 
-For the marginals, fix a permutation $sigma$ of ${1,...,r}$ and let
-$Θ_sigma$ apply it to color names in every tie-breaking permutation and
+For the marginals, fix a permutation sigma of {1,...,r} and let
+Θ_sigma apply it to color names in every tie-breaking permutation and
 every final uniform color input, fixing all times. It fixes the base point,
-commutes with rerooting, preserves each $kappa_v$, fixes the height, and acts
-isometrically on band $L^1$. Inductively suppose $ccircΘ_sigma=sigmacirc
-c$. Exact pullback of the finite-cut calculations and color covariance give
+commutes with rerooting, preserves each kappa_v, fixes the height, and acts
+isometrically on band L^1. Inductively suppose ccircΘ_sigma=sigmacirc
+c. Exact pullback of the finite-cut calculations and color covariance give
 [
  ((d^{sigma(i)})^{l,H}(c))circΘ_sigma=(d^i)^{l,H}(c),
  quad1le ile r,
 ]
-and the band $L^1$ isometry plus uniqueness of limits gives
-$(barbar d^d^{sigma(i)}(c))circΘ_sigma=barbar d^i(c)$d^i(c)$ a.e. So the minimizing
-set at the transformed input is the $sigma$-image of the original, the
+and the band L^1 isometry plus uniqueness of limits gives
+(barbar d^d^{sigma(i)}(c))circΘ_sigma=barbar d^i(c)d^i(c) a.e. So the minimizing
+set at the transformed input is the sigma-image of the original, the
 transformed tie-breaker chooses the corresponding color, and the update test is
 unchanged (times fixed).
 
-There are finitely many color permutations and, with $T_0,h$ fixed, finitely
+There are finitely many color permutations and, with T_0,h fixed, finitely
 many steps. Take a common conull set for the required identities, intersect its
 images under the finite color action, and discard the saturation of the
 complement (null by nonsingularity); the induction then holds throughout,
 starting from all-unassigned. The final uniform inputs transform the same way, so
-the completed coloring obeys $ccircΘ_sigma=sigmacirc c$. Since
-$Θ_sigma$ preserves the conditional product measure over each base point,
+the completed coloring obeys ccircΘ_sigma=sigmacirc c. Since
+Θ_sigma preserves the conditional product measure over each base point,
 disintegration shows all output colors at the root have equal conditional
 probabilities a.e.; summing to one gives (eq:coloring-marginal) and
 completes the proof.
@@ -2015,44 +2007,44 @@ completes the proof.
 # A Bernoulli model and transfer of state moments
 
 
-Throughout this section $Asubset M$ is a masa, $M$ has separable predual, and
-$varphi$ is a faithful normal state with $Asubset M_varphi$. Keep the
-notation of Section~sec:kernels: $N$ is generated by the groupoid
-normalizers of $A$, $E_N:Mto N$ is the $varphi$-preserving conditional
-expectation, $M^circ=ker E_N$, and $N_{rm alg}$ is the algebra of finite sums
+Throughout this section Asubset M is a masa, M has separable predual, and
+varphi is a faithful normal state with Asubset M_varphi. Keep the
+notation of Section~sec:kernels: N is generated by the groupoid
+normalizers of A, E_N:Mto N is the varphi-preserving conditional
+expectation, M^circ=ker E_N, and N_{rm alg} is the algebra of finite sums
 of base functions times kernel groupoid normalizers; the Cartan inclusion
-$Asubset N$ is represented by a countable nonsingular measured relation
-$mathcal R$calR$ on $(X,mu)$ with its scalar cocycle.
+Asubset N is represented by a countable nonsingular measured relation
+mathcal RcalR on (X,mu) with its scalar cocycle.
 
 The coloring of Section~sec:coloring lives on the Bernoulli extension of
-this relation. We place that extension and $M$ in a common von Neumann algebra,
-get a norm bound for the part of an operator in $M^circ$, then show every fixed
+this relation. We place that extension and M in a common von Neumann algebra,
+get a norm bound for the part of an operator in M^circ, then show every fixed
 scalar moment of finitely many model cylinder functions is a limit of averaged
-moments in $M$, which brings the model color projections back into $A$.
+moments in M, which brings the model color projections back into A.
 
 ## The model and a free compression bound
 
-Let $widetilde X$ be the relative Bernoulli extension of $(X,mu,mathcal R)$calR)$
+Let widetilde X be the relative Bernoulli extension of (X,mu,mathcal R)calR)
 from Section~sec:coloring with one Lebesgue label at each site of a
-relation class. Write $widetildemu$ for its probability and $B=L^∞(
-widetilde X,widetildemu)$. Re-rooting preserves the label assignment, so each
-extended class projects bijectively onto its base class. Let $widetilde N$ be
+relation class. Write widetildemu for its probability and B=L^∞(
+widetilde X,widetildemu). Re-rooting preserves the label assignment, so each
+extended class projects bijectively onto its base class. Let widetilde N be
 the von Neumann algebra of the extended relation with the cocycle pulled back.
 
 [The Bernoulli free-product model]
 
-Pulling back base kernel operators embeds $N$ as a von Neumann subalgebra of
-$widetilde N$, and there is a faithful normal conditional expectation
-$widetilde E_N:widetilde Nto N$ whose restriction to $B$ integrates labels
-conditional on the base point. With $widetildevarphi$ the diagonal integral in
-$widetilde N$, $(varphi|_N)circwidetilde E_N=widetildevarphi$. The reduced
-amalgamated free product $mathcal K=M*_NcalK=M*_Nwidetilde N$ has a faithful normal
-conditional expectation $E_N^{mathcal K}calK$ onto $N$ restricting to $E_N$ and
-$widetilde E_N$; the state $Phi=(varphi|_N)circ E_N^{mathcal K}calK$ is faithful
-and normal, restricts to $varphi$ and $widetildevarphi$, and satisfies
-$Bsubsetmathcal K_calK_Phi$. The two constituents are free over $N$: the
+Pulling back base kernel operators embeds N as a von Neumann subalgebra of
+widetilde N, and there is a faithful normal conditional expectation
+widetilde E_N:widetilde Nto N whose restriction to B integrates labels
+conditional on the base point. With widetildevarphi the diagonal integral in
+widetilde N, (varphi|_N)circwidetilde E_N=widetildevarphi. The reduced
+amalgamated free product mathcal K=M*_NcalK=M*_Nwidetilde N has a faithful normal
+conditional expectation E_N^{mathcal K}calK onto N restricting to E_N and
+widetilde E_N; the state Phi=(varphi|_N)circ E_N^{mathcal K}calK is faithful
+and normal, restricts to varphi and widetildevarphi, and satisfies
+Bsubsetmathcal K_calK_Phi. The two constituents are free over N: the
 expectation of every alternating product of centered letters from
-$ker E_N$ and $kerwidetilde E_N$ is zero.
+ker E_N and kerwidetilde E_N is zero.
 
 
 Proof. 
@@ -2060,94 +2052,94 @@ Use the right counting representations of Section~sec:kernels. Under the
 bijection of extended and base classes, a pulled-back base kernel acts by the
 same orbit operator on each fiber, so pullback is a normal faithful
 representation (an amplification over conditional label spaces; matrix
-coefficients disintegrate to normal functionals). The pulled-back copy of $N$
-lies in $widetilde N$ since this holds for generating kernel normalizers.
+coefficients disintegrate to normal functionals). The pulled-back copy of N
+lies in widetilde N since this holds for generating kernel normalizers.
 
 The isometry from the original space to the extension space making vector fields
-constant in labels intertwines the two copies of $N$ and sends the diagonal state
-vector to the new one. Compression by it maps $widetilde N$ normally into $N$.
-On finite sums of $B$-functions times lifted base groupoid normalizers,
-compression integrates coefficients over labels and lands in $N$; these form a
-weakly dense $*$-algebra (lifted partial isomorphisms enumerate the extended
+constant in labels intertwines the two copies of N and sends the diagonal state
+vector to the new one. Compression by it maps widetilde N normally into N.
+On finite sums of B-functions times lifted base groupoid normalizers,
+compression integrates coefficients over labels and lands in N; these form a
+weakly dense *-algebra (lifted partial isomorphisms enumerate the extended
 relation, cuts by extension-measurable sets and scalar phases give its kernel
 normalizers), so normality and weak closure give the range assertion for
-$widetilde N$.
+widetilde N.
 
-The compression is unital, $N$-bimodular, identity on $N$, so it is the asserted
+The compression is unital, N-bimodular, identity on N, so it is the asserted
 normal conditional expectation; it preserves the diagonal state, which is
 faithful by the faithful normal diagonal expectation of a countable nonsingular
-relation algebra, so $widetilde E_N$ is faithful. The diagonal expectation is
-$B$-bimodular and $B$ abelian, so $Bsubsetwidetilde N_{widetildevarphi}$.
+relation algebra, so widetilde E_N is faithful. The diagonal expectation is
+B-bimodular and B abelian, so Bsubsetwidetilde N_{widetildevarphi}.
 
-Both expectations onto $N$ are faithful and normal and the constituents are
-$sigma$-finite, so the reduced amalgamated free-product construction gives
-$mathcal K$,calK$, $E_N^{mathcal K}calK$, generation, and freeness over $N$
+Both expectations onto N are faithful and normal and the constituents are
+sigma-finite, so the reduced amalgamated free-product construction gives
+mathcal K,calK, E_N^{mathcal K}calK, generation, and freeness over N
 cite[Section~2]{Ued99},cite[Section~2]{Ued13} (these do not require tracial
-states); faithfulness and normality of $Phi$ follow from those of the
-expectation and $varphi|_N$.
+states); faithfulness and normality of Phi follow from those of the
+expectation and varphi|_N.
 
-For completeness, the product has $Phi$-preserving expectations onto each
+For completeness, the product has Phi-preserving expectations onto each
 constituent: compress the state representation to the closure of one
-constituent's state vectors. The span of $N$ and alternating centered words is
+constituent's state vectors. The span of N and alternating centered words is
 weakly dense, and every reduced word with a centered letter from the other
-constituent compresses to zero (expand at the ends, absorb $N$-coefficients, use
+constituent compresses to zero (expand at the ends, absorb N-coefficients, use
 freeness). The retained algebra acts normally and faithfully on its state-vector
 space, so compression is normal, bimodular, state-preserving onto that
 representation; by Takesaki [cite] the modular group restricts to each
-constituent (agreeing with cite[eq.~(2.1)]{Ued13}). Since $B$ is in the
-centralizer of $widetildevarphi$, it is in $mathcal K_calK_Phi$.
+constituent (agreeing with cite[eq.~(2.1)]{Ued13}). Since B is in the
+centralizer of widetildevarphi, it is in mathcal K_calK_Phi.
 
 
 [Free compression]
 
-In the model of Lemma~lem:bernoulli-model, let $pin B$ be a projection
-with $widetilde E_N(p)=lambda1$, $0lelambdale1$. For every $win M^circ$,
+In the model of Lemma~lem:bernoulli-model, let pin B be a projection
+with widetilde E_N(p)=lambda1, 0lelambdale1. For every win M^circ,
 
  |pwp|le(lambda+2sqrtlambda)|w|.
 
 
 
 Proof. 
-In $L^2(mathcal K,calK,Phi)$ let $P$ be the orthogonal projection onto the closed
-span of reduced-word vectors whose first letter is a centered $M$-letter. The
-orthogonal complement is densely spanned by $N$-vectors and reduced words whose
-first letter is a centered $widetilde N$-letter (generation and freeness over
-$N$). Left multiplication by $w$ sends these generators of the complement into
-$ranP$, so $(1-P)w(1-P)=0$. Left multiplication by
-$p-lambda1$ on a word starting with a centered $M$-letter inserts a centered
-$widetilde N$-letter at the front, orthogonal to $ranP$; hence
-$PpP=lambda P$ and $|Pp|^2=|PpP|lelambda$. Insert $P+(1-P)$ around $w$ in
-$pwp$: the $(1-P)w(1-P)$ term vanishes and the three remaining terms have norms
-at most $lambda|w|$, $sqrtlambda|w|$, $sqrtlambda|w|$, proving
+In L^2(mathcal K,calK,Phi) let P be the orthogonal projection onto the closed
+span of reduced-word vectors whose first letter is a centered M-letter. The
+orthogonal complement is densely spanned by N-vectors and reduced words whose
+first letter is a centered widetilde N-letter (generation and freeness over
+N). Left multiplication by w sends these generators of the complement into
+ranP, so (1-P)w(1-P)=0. Left multiplication by
+p-lambda1 on a word starting with a centered M-letter inserts a centered
+widetilde N-letter at the front, orthogonal to ranP; hence
+PpP=lambda P and |Pp|^2=|PpP|lelambda. Insert P+(1-P) around w in
+pwp: the (1-P)w(1-P) term vanishes and the three remaining terms have norms
+at most lambda|w|, sqrtlambda|w|, sqrtlambda|w|, proving
 (eq:free-compression).
 
 
 ## Finite symbols and their evaluation in the masa
 
-Fix an integer $sge2$ and partition each Lebesgue label into $s$ equally likely
-symbols identified with the $s$th roots of unity (dyadic, $s$ a power of two,
+Fix an integer sge2 and partition each Lebesgue label into s equally likely
+symbols identified with the sth roots of unity (dyadic, s a power of two,
 suffices later). A finite cylinder function is a bounded function on
-$widetilde X$ depending measurably on the base point and on these symbols at a
+widetilde X depending measurably on the base point and on these symbols at a
 fixed finite list of translated sites, each given by a partial isomorphism with
-graph in $mathcal R$calR$ (possibly also on whether the base point lies in its
-domain). Let $mathcal Y$calY$ be the algebra of finite sums of such functions times
+graph in mathcal RcalR (possibly also on whether the base point lies in its
+domain). Let mathcal YcalY be the algebra of finite sums of such functions times
 lifted base kernel groupoid normalizers; covariance and composition make it a
-$*$-algebra, and $widetilde E_N(mathcal Y)calY)subset N_{rm alg}$.
+*-algebra, and widetilde E_N(mathcal Y)calY)subset N_{rm alg}.
 
-Choose increasing finite Borel partitions $mathcal P_j$calP_j$ of $X$ generating $A$
+Choose increasing finite Borel partitions mathcal P_jcalP_j of X generating A
 and separating points off the fixed null exceptions. Write
-$mathcal P=calP={e_a}_a$; for independent uniform $s$th roots of unity $z_a$ let
-$u_{mathcal P}calP=sum_az_ae_ain A$. Replace the symbol at a translated site by
-the value of $u_{mathcal P}calP$ there and each lifted base normalizer by its
-original in $N$; this defines an evaluation $ev_{mathcal P}calP:mathcal
-YcalYto N_{rm alg}$, and finite cylinder functions evaluate to elements of $A$.
+mathcal P=calP={e_a}_a; for independent uniform sth roots of unity z_a let
+u_{mathcal P}calP=sum_az_ae_ain A. Replace the symbol at a translated site by
+the value of u_{mathcal P}calP there and each lifted base normalizer by its
+original in N; this defines an evaluation ev_{mathcal P}calP:mathcal
+YcalYto N_{rm alg}, and finite cylinder functions evaluate to elements of A.
 
-Evaluation respects multiplication and adjoints and fixes $N_{rm alg}$;
+Evaluation respects multiplication and adjoints and fixes N_{rm alg};
 substitution is consistent under re-rooting (hence with covariance) and respects
 a.e. identities: conditional on a.e. base point every assignment of symbols to
 a finite list of distinct sites has positive probability, so a cylinder identity
 holds for every such assignment off a base null set, including those from
-$u_{mathcal P}calP$; for sums with normalizers, check finitely supported root
+u_{mathcal P}calP; for sums with normalizers, check finitely supported root
 columns in the counting representations with common cocycle phases, equality
 a.e. giving equality by faithfulness of the diagonal state. So the definition
 is independent of the expression. Evaluated coefficients have norm at most their
@@ -2159,16 +2151,16 @@ The transfer result needs a uniform approximation estimate for fixed scalar
 words, which also serves the final clipping.
 
 [Uniform right multiplication in the state norm]
-Let $varphi$ be a faithful normal state on $M$. For every $C<∞$, fixed
-$zin M$, and $eta>0$ there is $delta>0$ such that
+Let varphi be a faithful normal state on M. For every C<∞, fixed
+zin M, and eta>0 there is delta>0 such that
 
  |D|le C, |D|_varphi<delta ⇒ |Dz|_varphi<eta,
 
-uniformly over such $D$. For $U,Din M$, $bin M_varphi$,
+uniformly over such D. For U,Din M, bin M_varphi,
 
- |UD|_varphile|U||D|_varphi,qquad |Db|_varphile|b||D|_varphi.
+ |UD|_varphile|U||D|_varphi, |Db|_varphile|b||D|_varphi.
 
-The right multiplier bound is uniform when $z$ ranges over a norm-bounded set
+The right multiplier bound is uniform when z ranges over a norm-bounded set
 precompact in state norm. Hence a bounded family of state-norm errors tending
 uniformly to zero stays so after right multiplication by a fixed finite word
 whose factors range over such precompact sets or over bounded subsets of the
@@ -2180,27 +2172,27 @@ indices of other approximation sequences.
 
 
 Proof. 
-Let $Ω$ be the GNS vector; faithfulness makes it separating, hence cyclic
-for $M'$. Pick $c'in M'$ with $|zΩ-c'Ω|<alpha$; then for
-$|D|le C$,
-$|Dz|_varphile Calpha+|Dc'Ω|=Calpha+|c'DΩ|
-le Calpha+|c'||D|_varphi$, and choosing $alpha$ first gives
+Let Ω be the GNS vector; faithfulness makes it separating, hence cyclic
+for M'. Pick c'in M' with |zΩ-c'Ω|<alpha; then for
+|D|le C,
+|Dz|_varphile Calpha+|Dc'Ω|=Calpha+|c'DΩ|
+le Calpha+|c'||D|_varphi, and choosing alpha first gives
 (eq:uniform-fixed-right), including parametrized families.
 
-The left multiplier bound is immediate. For $bin M_varphi$ and $D_0ge0$,
-centrality gives $varphi(b^*D_0b)=varphi(D_0bb^*)$, and
-$cmapstovarphi(D_0c)$ is positive on $M_varphi$ ($cge0$ gives
-$varphi(c^{1/2}D_0c^{1/2})$), so $varphi(b^*D_0b)le|b|^2varphi(D_0)$,
+The left multiplier bound is immediate. For bin M_varphi and D_0ge0,
+centrality gives varphi(b^*D_0b)=varphi(D_0bb^*), and
+cmapstovarphi(D_0c) is positive on M_varphi (cge0 gives
+varphi(c^{1/2}D_0c^{1/2})), so varphi(b^*D_0b)le|b|^2varphi(D_0),
 proving (eq:state-left-centralizer).
 
-For a norm-bounded state-norm precompact $mathcal C$calC$ take a finite approximation
-$z_1,...,z_m$; if $z$ is close to $z_i$ then
-$|Dz|_varphile|Dz_i|_varphi+C|z-z_i|_varphi$, and the finitely many
-fixed-factor moduli give one for $mathcal C$calC$ (a convergent sequence plus limit is
-precompact, covering every bounded $z_j$ with $|z_j-z|_varphito0$). Apply
+For a norm-bounded state-norm precompact mathcal CcalC take a finite approximation
+z_1,...,z_m; if z is close to z_i then
+|Dz|_varphile|Dz_i|_varphi+C|z-z_i|_varphi, and the finitely many
+fixed-factor moduli give one for mathcal CcalC (a convergent sequence plus limit is
+precompact, covering every bounded z_j with |z_j-z|_varphito0). Apply
 right multiplier bounds in the order of factors; errors stay uniformly bounded in
-operator norm so each modulus applies. For a scalar word with middle error $D$,
-$|varphi(UDV)|le|U||DV|_varphi$ and telescoping reduce it to the preceding
+operator norm so each modulus applies. For a scalar word with middle error D,
+|varphi(UDV)|le|U||DV|_varphi and telescoping reduce it to the preceding
 bounds. Strong-* approximation ensures state-norm convergence of a replaced
 letter and adjoint; for each fixed approximant the finite words may be expanded,
 and its index sent to infinity after the uniform estimates.
@@ -2208,11 +2200,11 @@ and its index sent to infinity after the uniform estimates.
 
 [Transfer of fixed scalar moments]
 
-Let $Asubset M$, $varphi$, $mathcal K$calK$ be as in Lemma~lem:bernoulli-model.
-Fix a symbol size $s$, the cylinder algebra $mathcal Y$,calY$, and the generating
-$(mathcal P_j)$.calP_j)$. For a fixed integer $kge1$ and fixed factors $Z_1,...,Z_k$,
-each in $M$ or $mathcal Y$,calY$, set $Z_h^{(j)}=Z_h$ for an $M$-factor and
-$Z_h^{(j)}=ev_{mathcal P_j}calP_j(Z_h)$ for a $mathcal Y$-factor,calY$-factor, using the
+Let Asubset M, varphi, mathcal KcalK be as in Lemma~lem:bernoulli-model.
+Fix a symbol size s, the cylinder algebra mathcal Y,calY, and the generating
+(mathcal P_j).calP_j). For a fixed integer kge1 and fixed factors Z_1,...,Z_k,
+each in M or mathcal Y,calY, set Z_h^{(j)}=Z_h for an M-factor and
+Z_h^{(j)}=ev_{mathcal P_j}calP_j(Z_h) for a mathcal Y-factor,calY-factor, using the
 same symbols for all evaluations in a word. Then
 
  lim_{jto∞}mathbb E_bbE_{{z_a}}
@@ -2228,159 +2220,159 @@ Proposition~prop:atomless-kernels.
 
 Uniform approximation in fixed words. Lemma~lem:state-multipliers
 controls bounded state-norm errors with the right word from fixed operators and
-bounded $A$-functions. Expanding each finite normalizer sum if needed,
-telescoping and $|varphi(UDV)|le|U||DV|_varphi$ bound word errors
-uniformly over evaluated diagonal functions. If a fixed $N$-letter is
-approximated by a bounded strongly convergent sequence from $N_{rm alg}$, the
+bounded A-functions. Expanding each finite normalizer sum if needed,
+telescoping and |varphi(UDV)|le|U||DV|_varphi bound word errors
+uniformly over evaluated diagonal functions. If a fixed N-letter is
+approximated by a bounded strongly convergent sequence from N_{rm alg}, the
 sequence and its limit are state-norm precompact, so the lemma applies to the
 whole sequence as right factors; expand each normalizer sum only after fixing an
 approximant, pass to the partition limit, then send the approximation error to
-zero. The same holds in the model with $Phi$ and bounded $B$-factors
-($Bsubsetmathcal K_calK_Phi$).
+zero. The same holds in the model with Phi and bounded B-factors
+(Bsubsetmathcal K_calK_Phi).
 
 Reduction to centered alternating words. Induct on the number of factors.
-A length-one $M$-word is an identity. A length-one $mathcal Y$-word:calY$-word: average its
+A length-one M-word is an identity. A length-one mathcal Y-word:calY-word: average its
 cylinder coefficients; at a.e. base point distinct sites of a fixed finite
-translated list eventually lie in distinct atoms of $mathcal P_j$,calP_j$, so the
+translated list eventually lie in distinct atoms of mathcal P_j,calP_j, so the
 averaged coefficient converges boundedly a.e. to its conditional label average,
 which is also state-norm convergence with the fixed normalizers allowed as right
 multipliers; this proves (eq:moment-transfer) in length one.
 
 Fold adjacent same-algebra factors. For an alternating word subtract and add the
-$N$-expectation of each model factor (the same expansion on the evaluated side;
+N-expectation of each model factor (the same expansion on the evaluated side;
 a centered model factor need not stay centered after evaluation). Expectations
-of $mathcal Y$-factorscalY$-factors lie in $N_{rm alg}$ and are fixed by evaluation; each
+of mathcal Y-factorscalY-factors lie in N_{rm alg} and are fixed by evaluation; each
 term with an expectation folds into a neighbor, reducing length. An arbitrary
-$N$-element multiplying a $mathcal Y$-factorcalY$-factor is first approximated boundedly
-strong-* by $N_{rm alg}$ (generation and Kaplansky), legitimate on both sides by
+N-element multiplying a mathcal Y-factorcalY-factor is first approximated boundedly
+strong-* by N_{rm alg} (generation and Kaplansky), legitimate on both sides by
 the uniform word estimates. The induction leaves fully centered alternating
 words, whose model expectation is zero by freeness.
 
-Approximate each centered $M$-factor boundedly strong-* by an entire analytic
-element for $sigma^varphi$ in $M^circ$ (convolve its modular orbit with
+Approximate each centered M-factor boundedly strong-* by an entire analytic
+element for sigma^varphi in M^circ (convolve its modular orbit with
 Gaussian approximate identities; complex translations of the Gaussian give the
-entire continuation; $E_N$ commutes with the modular group, so the convolutions
-and their complex translates stay centered). Each centered $mathcal Y$-factorcalY$-factor is
-a finite sum of $vb$, $v$ a base kernel groupoid normalizer, $b$ a cylinder
+entire continuation; E_N commutes with the modular group, so the convolutions
+and their complex translates stay centered). Each centered mathcal Y-factorcalY-factor is
+a finite sum of vb, v a base kernel groupoid normalizer, b a cylinder
 function with zero conditional label average: move each coefficient right of its
-normalizer by covariance, absorb bounded base coefficients into $b$, and subtract
-the summand expectation (sum zero). Cut the initial support of $v$ as in
+normalizer by covariance, absorb bounded base coefficients into b, and subtract
+the summand expectation (sum zero). Cut the initial support of v as in
 Section~sec:kernels to be entire analytic, preserving centering with
 arbitrarily small state errors in the fixed words. Cylinder functions and their
-evaluations lie in the centralizers, so complex modular shifts of $vb$ are
+evaluations lie in the centralizers, so complex modular shifts of vb are
 evaluation-compatible (the shift only changes the cut normalizer by the same
 bounded base formula in both algebras). Fix these analytic approximants before the
 partition limit; their complex-shift norms enter later bounds.
 
 Odd words. For an odd alternating word of length at least three the first
 and last factors lie in the same constituent. Apply the KMS identity
-$varphi(ZY)=varphi(Ysigma^varphi_{-i}(Z))$ to the first analytic factor (and
-the $Phi$-analogue), fold the last with the shifted first, and use the induction
+varphi(ZY)=varphi(Ysigma^varphi_{-i}(Z)) to the first analytic factor (and
+the Phi-analogue), fold the last with the shifted first, and use the induction
 hypothesis on the shorter word, whose model value equals the original centered
 word's and is zero.
 
-Even words. Rotate to start in $M^circ$; expand other-side factors as
-$vb$ and absorb each $v$ into the preceding centered $M$-factor (modular shifts
-only add fixed base coefficients). For word length $2m$ it remains to prove the
+Even words. Rotate to start in M^circ; expand other-side factors as
+vb and absorb each v into the preceding centered M-factor (modular shifts
+only add fixed base coefficients). For word length 2m it remains to prove the
 averaged evaluations of
 
  varphi(y_1d_1... y_md_m)
 
-tend to zero, where the $y_h$ are fixed bounded entire analytic elements of
-$M^circ$ and each $d_hin A$ is a substitution in a centered finite cylinder
+tend to zero, where the y_h are fixed bounded entire analytic elements of
+M^circ and each d_hin A is a substitution in a centered finite cylinder
 function; constants may depend on this fixed word and the analytic approximants.
 
 Fourier expansion and index equalities. Split each cylinder domain by the
 domains of its translated sites and the equality pattern among them; on each
 piece its distinct symbols are independent. Fourier expansion on the finite
 product of cyclic groups writes the centered coefficient as a finite sum of
-monomials with bounded $A$-coefficients; centering removes the constant term and
+monomials with bounded A-coefficients; centering removes the constant term and
 we delete zero-exponent occurrences, leaving each monomial with at least one
-occurrence, all nonzero mod $s$, and within each block $d_h$ the occurrence sites
+occurrence, all nonzero mod s, and within each block d_h the occurrence sites
 pointwise distinct on the indicated support.
 
 After further splitting, regard each translation at an occurrence as an
-automorphism of $A$ (the extension off the support need not be in $mathcal R$).calR$).
-Proposition~prop:atomless-kernels makes the $y_h$ annihilate the atomic part
-of $A$ on both sides, so only diffuse supports contribute. Partial isomorphisms
+automorphism of A (the extension off the support need not be in mathcal R).calR).
+Proposition~prop:atomless-kernels makes the y_h annihilate the atomic part
+of A on both sides, so only diffuse supports contribute. Partial isomorphisms
 map these supports into the diffuse part off null sets; splitting a nonnull common
 domain piece into two positive-measure pieces gives nonnull diffuse complements
 (the other piece and its image), and the isomorphism theorem for atomless standard
 probability spaces cite[Theorem~17.41]{Kec95} extends each restricted map to a
 measure-class automorphism of the diffuse part (identity on the atomic part),
-giving finitely many terms with $lambda_o(f)=fcircgamma_o$, $gamma_o$ a
-nonsingular bijection, at each occurrence $o$.
+giving finitely many terms with lambda_o(f)=fcircgamma_o, gamma_o a
+nonsingular bijection, at each occurrence o.
 
-Expand (eq:cyclic-centered-word) in one partition index $a_o$ per
-occurrence. Averaging the independent $z_a$ keeps exactly the equality patterns
-with zero total Fourier exponent mod $s$ in each class; since individual
+Expand (eq:cyclic-centered-word) in one partition index a_o per
+occurrence. Averaging the independent z_a keeps exactly the equality patterns
+with zero total Fourier exponent mod s in each class; since individual
 exponents are nonzero, every class has at least two occurrences. Inclusion--
 exclusion over coarsenings removes inequalities between classes, so it suffices to
 prove vanishing of a sum with prescribed equalities within the classes of a
 partition of occurrences, each class of size at least two, no inequalities
-between classes; occurrence $o$ contributes $lambda_o(e_{a_o})$.
+between classes; occurrence o contributes lambda_o(e_{a_o}).
 
 Keep selected pair equalities as explicit indices. If selected pairs form a forest
 in each class, extend to a spanning tree and encode the other tree edges with
-independent phases: for an edge between $a,b$,
-$mathbf1_bf1_{{a=b}}=mathbb E_bbE_{boldsymbolzeta}(zeta_azeta_bbar)$
-with independent uniform circle phases (fresh vector per edge). At occurrence $o$
-let $w_o(a)$ be the product of incident auxiliary edge phases (conjugates per
-orientation, empty product $1$), so $|w_o(a)|=1$, and set
-$U_o=sum_aw_o(a)lambda_o(e_a)$; since ${lambda_o(e_a)}_a$ is a partition,
-$U_o$ is a diagonal unitary, with $sum_a$ unretained giving $U_o$ and at retained
-$j$, $w_o(j)lambda_o(e_j)=U_olambda_o(e_j)$. So each $U_o$ is independent of the
-retained indices; different $U_o$ may be correlated through the phases, and the
+independent phases: for an edge between a,b,
+mathbf1_bf1_{{a=b}}=mathbb E_bbE_{boldsymbolzeta}(zeta_azeta_bbar)
+with independent uniform circle phases (fresh vector per edge). At occurrence o
+let w_o(a) be the product of incident auxiliary edge phases (conjugates per
+orientation, empty product 1), so |w_o(a)|=1, and set
+U_o=sum_aw_o(a)lambda_o(e_a); since {lambda_o(e_a)}_a is a partition,
+U_o is a diagonal unitary, with sum_a unretained giving U_o and at retained
+j, w_o(j)lambda_o(e_j)=U_olambda_o(e_j). So each U_o is independent of the
+retained indices; different U_o may be correlated through the phases, and the
 estimates hold uniformly over their joint realizations before averaging. Products
 of these factors commute within a block.
 
 A pair in one block. If two occurrences of a class lie in one block, retain
 their pair equality, encode the rest, and sum the pair index to
-$p_{mathcal P}calP=mathbf1_bf1_{support}sum_alambda_o(e_a)lambda_{o'}(e_a)$;
+p_{mathcal P}calP=mathbf1_bf1_{support}sum_alambda_o(e_a)lambda_{o'}(e_a);
 the sites are distinct on the support and partitions separate points, so
-$mu(p_{mathcal P}calP)to0$. KMS-rotate this projection to the front (only fixed
+mu(p_{mathcal P}calP)to0. KMS-rotate this projection to the front (only fixed
 analytic letters shift; diagonal factors stay in the centralizer) to get
-$varphi(p_{mathcal P}calPZ)$ with $|Z|$ uniformly bounded; centrality gives
-$|varphi(p_{mathcal P}calPZ)|=|varphi(p_{mathcal P}calPZp_{mathcal P}calP)|le
-|Z|mu(p_{mathcal P}calP)$, so these sums tend to zero uniformly in the phases.
+varphi(p_{mathcal P}calPZ) with |Z| uniformly bounded; centrality gives
+|varphi(p_{mathcal P}calPZ)|=|varphi(p_{mathcal P}calPZp_{mathcal P}calP)|le
+|Z|mu(p_{mathcal P}calP), so these sums tend to zero uniformly in the phases.
 
 A shortest pair between blocks. It remains to treat patterns with at most
-one occurrence of any class per block, so $mge2$. Number blocks by
-$mathbb Z/mbbZ/mmathbb Z$bbZ$ and for distinct $h,k$ let
-$d_m(h,k)=min{(k-h)bmod m,(h-k)bmod m}$ with residues in ${1,...,m-1}$.
+one occurrence of any class per block, so mge2. Number blocks by
+mathbb Z/mbbZ/mmathbb ZbbZ and for distinct h,k let
+d_m(h,k)=min{(k-h)bmod m,(h-k)bmod m} with residues in {1,...,m-1}.
 Among same-class pairs choose one minimizing this cyclic distance, call the
-minimum $ell$, and choose the endpoint order so one shortest arc runs forward,
-labeling its blocks $0,1,...,ell$ (either order allowed in an antipodal tie).
+minimum ell, and choose the endpoint order so one shortest arc runs forward,
+labeling its blocks 0,1,...,ell (either order allowed in an antipodal tie).
 
-For interior $h=1,...,ell-1$ pick an occurrence and another of its class; its
+For interior h=1,...,ell-1 pick an occurrence and another of its class; its
 mate lies outside the chosen arc including endpoints (a mate on the arc in block
-$k$ would give $d_m(h,k)le|h-k|<ell$, contradicting minimality). This also makes
-interior classes distinct from each other and from the endpoint class; $ell=1$
+k would give d_m(h,k)le|h-k|<ell, contradicting minimality). This also makes
+interior classes distinct from each other and from the endpoint class; ell=1
 has no interior blocks.
 
 Retain the endpoint pair and these interior-to-mate pairs; these lie in distinct
 classes, form a forest, and are included in the spanning trees. Retained indices
-are $J=(j_0,j_1,...,j_{ell-1})$, $j_0$ linking the endpoints. Write
-$E_0=lambda_0(e_{j_0})$, $E_ell'=lambda_ell(e_{j_0})$,
-$E_h=lambda_h(e_{j_h})$ ($1le h<ell$); each $j_h$ ($hge1$) also occurs in one
+are J=(j_0,j_1,...,j_{ell-1}), j_0 linking the endpoints. Write
+E_0=lambda_0(e_{j_0}), E_ell'=lambda_ell(e_{j_0}),
+E_h=lambda_h(e_{j_h}) (1le h<ell); each j_h (hge1) also occurs in one
 further retained projection in an exterior block (several may share a block).
 
-KMS-rotate to begin at $E_0$ and insert $E_0$ at the far end inside the state by
+KMS-rotate to begin at E_0 and insert E_0 at the far end inside the state by
 centrality. Absorb fixed diagonal coefficients into neighboring fixed letters
 (within a block place retained projections first, then auxiliary unitaries, then
 the fixed coefficient; absorb it into the following letter). The fixed
 intervening letters (including modular translates) stay bounded analytic
-elements of $M^circ$; relabel as $y_h$, and write $V_hin A$ for the auxiliary
-unitaries (independent of explicit indices). Split at $E_ell'$ using
-$(E_ell')^2=E_ell'$. Before averaging the unitaries the summands are
-$varphi(X_JY_J)$ with
+elements of M^circ; relabel as y_h, and write V_hin A for the auxiliary
+unitaries (independent of explicit indices). Split at E_ell' using
+(E_ell')^2=E_ell'. Before averaging the unitaries the summands are
+varphi(X_JY_J) with
 
  X_J&=E_0V_0(prod_{h=1}^{ell-1}y_hE_hV_h)y_ell E_ell',notag
  Y_J&=E_ell'V_ell B_{j_1,...,j_{ell-1}}E_0,
  
 
-the product in $X_J$ increasing and omitted when $ell=1$; in
-$B_{j_1,...,j_{ell-1}}$ each exterior mate projection for these indices
+the product in X_J increasing and omitted when ell=1; in
+B_{j_1,...,j_{ell-1}} each exterior mate projection for these indices
 occurs exactly once and every other factor is independent of them; diagonal
 factors in a block commute. (Figure~fig:shortest-pair shows one cyclic
 pattern.)
@@ -2392,14 +2384,14 @@ centering
   word/.style={-{Stealth[length=1.5mm]},line width=0.35pt},
   equality/.style={densely dashed,line width=0.7pt},
   every node/.style={font=small}]
-node[block] (d0) at (0,1.5) {$d_0$};
-node[block] (d1) at (2,1.5) {$d_1$};
-node[block] (d2) at (4,1.5) {$d_2$};
-node[block] (d3) at (6,1.5) {$d_3$};
-node[block] (d4) at (6,-1) {$d_4$};
-node[block] (d5) at (4,-1) {$d_5$};
-node[block] (d6) at (2,-1) {$d_6$};
-node[block] (d7) at (0,-1) {$d_7$};
+node[block] (d0) at (0,1.5) {d_0};
+node[block] (d1) at (2,1.5) {d_1};
+node[block] (d2) at (4,1.5) {d_2};
+node[block] (d3) at (6,1.5) {d_3};
+node[block] (d4) at (6,-1) {d_4};
+node[block] (d5) at (4,-1) {d_5};
+node[block] (d6) at (2,-1) {d_6};
+node[block] (d7) at (0,-1) {d_7};
 draw[word] (d0.east) -- (d1.west);
 draw[word] (d1.east) -- (d2.west);
 draw[word] (d2.east) -- (d3.west);
@@ -2408,19 +2400,19 @@ draw[word] (d4.west) -- (d5.east);
 draw[word] (d5.west) -- (d6.east);
 draw[word] (d6.west) -- (d7.east);
 draw[word] (d7.north) -- (d0.south);
-coordinate (o51) at ($(d5.north)+(-0.2,0)$);
-coordinate (o52) at ($(d5.north)+(0.2,0)$);
-draw[equality] (d0.north) to[out=90,in=90] node[above,pos=0.5] {$j_0$} (d3.north);
-draw[equality] (d1.south) to[out=-65,in=110] node[left,pos=0.45] {$j_1$} (o51);
-draw[equality] (d2.south) to[out=-70,in=70] node[right,pos=0.45] {$j_2$} (o52);
+coordinate (o51) at ((d5.north)+(-0.2,0));
+coordinate (o52) at ((d5.north)+(0.2,0));
+draw[equality] (d0.north) to[out=90,in=90] node[above,pos=0.5] {j_0} (d3.north);
+draw[equality] (d1.south) to[out=-65,in=110] node[left,pos=0.45] {j_1} (o51);
+draw[equality] (d2.south) to[out=-70,in=70] node[right,pos=0.45] {j_2} (o52);
 foreach point in {d0.north,d3.north,d1.south,d2.south,o51,o52}
   fill (point) circle[radius=1.4pt];
-node[fill=white,inner sep=2pt] at (3,2.1) {short arc: $X_J$};
-node at (3,-1.7) {complementary arc: $Y_J$};
+node[fill=white,inner sep=2pt] at (3,2.1) {short arc: X_J};
+node at (3,-1.7) {complementary arc: Y_J};
 
 Caption: Selected equality links in one fixed cyclic word with eight blocks and
-$ell=3$. The short arc $0to1to2to3$ and its complement indicate the positions
-of $X_J$ and $Y_J$; the two marks in block $5$ are distinct mate occurrences.
+ell=3. The short arc 0to1to2to3 and its complement indicate the positions
+of X_J and Y_J; the two marks in block 5 are distinct mate occurrences.
 Further occurrences and links are omitted; distinct prescribed classes may use
 equal numerical indices after inequalities are removed.
 
@@ -2432,73 +2424,73 @@ State Cauchy--Schwarz gives
  le(sum_Jvarphi(X_JX_J^*))^{1/2}
    (sum_Jvarphi(Y_J^*Y_J))^{1/2}.
 
-The second square sum is uniformly bounded: drop $E_ell'$ from the positive bound
-of $Y_J^*Y_J$, order the exterior mate occurrences left to right (commutativity in
-a block), write $mathbf j=(j_1,...,j_bfj=(j_1,...,j_{ell-1})$, and at the first remaining
-mate factor $B_{mathbf j}bfj=C_0F_jB'_{mathbf j'}bfj'$ ($F_j$ its projection partition,
-$C_0,B'_{mathbf j'}bfj'$ independent of $j$), so
-$sum_jB_{mathbf j}bfj^*B_{mathbf j}bfj
+The second square sum is uniformly bounded: drop E_ell' from the positive bound
+of Y_J^*Y_J, order the exterior mate occurrences left to right (commutativity in
+a block), write mathbf j=(j_1,...,j_bfj=(j_1,...,j_{ell-1}), and at the first remaining
+mate factor B_{mathbf j}bfj=C_0F_jB'_{mathbf j'}bfj' (F_j its projection partition,
+C_0,B'_{mathbf j'}bfj' independent of j), so
+sum_jB_{mathbf j}bfj^*B_{mathbf j}bfj
 ={B'_{mathbf j'}bfj'}^*(sum_jF_jC_0^*C_0F_j)B'_{mathbf j'}bfj'
-le|C_0|^2{B'_{mathbf j'}bfj'}^*B'_{mathbf j'}bfj'$; iterating over the remaining mates
+le|C_0|^2{B'_{mathbf j'}bfj'}^*B'_{mathbf j'}bfj'; iterating over the remaining mates
 (with the fixed intervening norm bounds) gives
-$sum_{mathbf j}bfjB_{mathbf j}bfj^*B_{mathbf j}bfjle C1$ (for $ell=1$ just the fixed
-exterior word bound). Summing the $E_0$ corners in the state (a centralizer
-partition) leaves a bound independent of $mathcal P$calP$ and the unitaries.
+sum_{mathbf j}bfjB_{mathbf j}bfj^*B_{mathbf j}bfjle C1 (for ell=1 just the fixed
+exterior word bound). Summing the E_0 corners in the state (a centralizer
+partition) leaves a bound independent of mathcal PcalP and the unitaries.
 
 The first square sum in (eq:shortest-pair-cs) gives the smallness, from a
 collision between the two endpoint indices; the interior pinchings converge first
-at fixed coarse endpoint data, which are refined only afterward. Removing $V_0$
-by centrality and summing interior indices in $X_JX_J^*$ gives
+at fixed coarse endpoint data, which are refined only afterward. Removing V_0
+by centrality and summing interior indices in X_JX_J^* gives
 
  sum_{j_0}varphi(lambda_0(e_{j_0})
    C_{mathcal P}calP(lambda_ell(e_{j_0}))lambda_0(e_{j_0})),
 
-where $C_{mathcal P}calP:Ato M$ is $C_{mathcal P}calP(f)=L_1$ with
-$L_ell=y_ell fy_ell^*$ and
-$L_h=y_hS_{lambda_h(mathcal P)}calP)(V_hL_{h+1}V_h^*)y_h^*$ for
-$h=ell-1,...,1$ (for $ell=1$, $C_{mathcal P}calP(f)=y_1fy_1^*$).
+where C_{mathcal P}calP:Ato M is C_{mathcal P}calP(f)=L_1 with
+L_ell=y_ell fy_ell^* and
+L_h=y_hS_{lambda_h(mathcal P)}calP)(V_hL_{h+1}V_h^*)y_h^* for
+h=ell-1,...,1 (for ell=1, C_{mathcal P}calP(f)=y_1fy_1^*).
 
-Fix a coarse $mathcal Q$calQ$ from the generating sequence and let $mathcal P$calP$
+Fix a coarse mathcal QcalQ from the generating sequence and let mathcal PcalP
 refine it. Enlarging each input projection in (eq:endpoint-collision) to its
 coarse piece and regrouping output corners by centrality bounds that expression by
 
  sum_{einmathcal Q}calQvarphi(lambda_0(e)
    C_{mathcal P}calP(lambda_ell(e))lambda_0(e)).
 
-For fixed $fin A$ the nested $C_{mathcal P}calP(f)$ converges in state norm to the
-expression with each pinching replaced by $E_A$ and the $V_h$ removed, uniformly
+For fixed fin A the nested C_{mathcal P}calP(f) converges in state norm to the
+expression with each pinching replaced by E_A and the V_h removed, uniformly
 over the unitaries: work innermost outward; each translated sequence
-$lambda_h(mathcal P_j)$calP_j)$ is generating, so Lemma~lem:pinching gives
-state-norm convergence to $E_A$ on the fixed input; conjugation by $V_h$ commutes
-with pinching, preserves the state norm, and fixes the limiting $A$-valued datum;
+lambda_h(mathcal P_j)calP_j) is generating, so Lemma~lem:pinching gives
+state-norm convergence to E_A on the fixed input; conjugation by V_h commutes
+with pinching, preserves the state norm, and fixes the limiting A-valued datum;
 pinching is state-norm contractive, and Lemma~lem:state-multipliers
-propagates the error through the fixed $y_h,y_h^*$, uniformly in the remaining
+propagates the error through the fixed y_h,y_h^*, uniformly in the remaining
 phases.
 
-Define positive kernels $K_h(f)=E_A(y_hfy_h^*)$ ($1le hleell$). For each fixed
-$f=lambda_ell(e)$, $einmathcal Q$,calQ$, the preceding convergence and finiteness of
-$mathcal Q$calQ$ give the limit of (eq:coarse-endpoint-bound):
+Define positive kernels K_h(f)=E_A(y_hfy_h^*) (1le hleell). For each fixed
+f=lambda_ell(e), einmathcal Q,calQ, the preceding convergence and finiteness of
+mathcal QcalQ give the limit of (eq:coarse-endpoint-bound):
 
  sum_{einmathcal Q}calQint_Xlambda_0(e)(K_1circ...circ K_ell)
    (lambda_ell(e)),dmu.
 
-Proposition~prop:atomless-kernels (applied to the adjoints of the $y_h$)
+Proposition~prop:atomless-kernels (applied to the adjoints of the y_h)
 makes these kernels and their composition atomless a.e.; in compositions a
 preceding kernel avoids the fixed null set where a following one may fail to be
 atomless.
 
-Refine $mathcal Q$calQ$ along the generating sequence. The integrand sum tests
-whether $gamma_0$ of the output site and $gamma_ell$ of the input site lie in
+Refine mathcal QcalQ along the generating sequence. The integrand sum tests
+whether gamma_0 of the output site and gamma_ell of the input site lie in
 the same piece; these conditions decrease to equality of the two points. Since
-$gamma_ell$ is one-to-one, this graph has zero conditional mass for each
+gamma_ell is one-to-one, this graph has zero conditional mass for each
 atomless row of the composed kernel. Nonsingularity makes fixed null exceptions
 irrelevant, and bounded convergence shows (eq:coarse-kernel-collision)
 tends to zero.
 
-For each fixed coarse $mathcal Q$,calQ$, the limsup of the first square sum, even after
-taking the supremum over the auxiliary unitaries at each fine $mathcal P$,calP$, was
+For each fixed coarse mathcal Q,calQ, the limsup of the first square sum, even after
+taking the supremum over the auxiliary unitaries at each fine mathcal P,calP, was
 bounded by (eq:coarse-kernel-collision). First take that fine-partition
-limsup, then refine $mathcal Q$:calQ$: the first square sum vanishes uniformly in the
+limsup, then refine mathcal Q:calQ: the first square sum vanishes uniformly in the
 auxiliary phases. The second square sum was uniformly bounded, so
 (eq:shortest-pair-cs) proves vanishing of the equality-pattern sum. Averaging
 the auxiliary phases, undoing the finite inclusion--exclusion and Fourier
@@ -2514,28 +2506,28 @@ Theorem~thm:bernoulli-coloring.
 [Cylinder approximation of projection partitions]
 
 In the Bernoulli extension of Lemma~lem:bernoulli-model, let
-$p_1,...,p_rin B$ be projections with $sum_ip_i=1$, $p_ip_h=0$ for $ine h$.
-For every $delta>0$ there is a projection partition $b_1,...,b_rin B$, each a
+p_1,...,p_rin B be projections with sum_ip_i=1, p_ip_h=0 for ine h.
+For every delta>0 there is a projection partition b_1,...,b_rin B, each a
 finite cylinder function for one common finite dyadic symbol size, with
-$|b_i-p_i|_Phi<delta$ ($1le ile r$). Every evaluation of this cylinder
-partition by the finite-symbol substitution gives a projection partition of $1$ in
-$A$ off the fixed null sets of its definition.
+|b_i-p_i|_Phi<delta (1le ile r). Every evaluation of this cylinder
+partition by the finite-symbol substitution gives a projection partition of 1 in
+A off the fixed null sets of its definition.
 
 
 Proof. 
-The extension $sigma$-algebra is generated by base measurable sets and countably
+The extension sigma-algebra is generated by base measurable sets and countably
 many label-coordinate readings along the partial graphings, so dyadic
 approximations of label coordinates show the finite cylinder set algebra
 (arbitrary base measurable sets allowed) is dense in probability. Approximate the
-sets of the $p_i$ by cylinder sets, disjointize in order, and put the complement
-in one piece, keeping exactly $r$ pieces covering the identity; symmetric
+sets of the p_i by cylinder sets, disjointize in order, and put the complement
+in one piece, keeping exactly r pieces covering the identity; symmetric
 differences can be made arbitrarily small, giving the state-norm bounds. Finitely
 many pieces and coordinates allow one common finite dyadic symbol size.
 
-The identities $b_i=b_i^*=b_i^2$, $b_ib_h=0$ ($ine h$), $sum_ib_i=1$ are
+The identities b_i=b_i^*=b_i^2, b_ib_h=0 (ine h), sum_ib_i=1 are
 cylinder identities, and the full-support argument in the construction of
-$ev_{mathcal P}calP$ shows substitution preserves them, so the evaluations
-form a genuine projection partition in $A$.
+ev_{mathcal P}calP shows substitution preserves them, so the evaluations
+form a genuine projection partition in A.
 
 
 # Completion of the paving argument
@@ -2550,110 +2542,110 @@ multipliers.
 ## Continuous calculus for varying sequences
 
 
-Let $mathcal L$calL$ be a von Neumann algebra with faithful normal state $omega$, and
-let $mathscr R_scrR_omega$ be the set of uniformly bounded sequences $(R_n)$ in
-$mathcal L$calL$ with $|D_nR_n|_omegato0$ whenever $(D_n)$ is uniformly bounded
-and $|D_n|_omegato0$. Then $mathscr R_scrR_omega$ is an algebra closed in
+Let mathcal LcalL be a von Neumann algebra with faithful normal state omega, and
+let mathscr R_scrR_omega be the set of uniformly bounded sequences (R_n) in
+mathcal LcalL with |D_nR_n|_omegato0 whenever (D_n) is uniformly bounded
+and |D_n|_omegato0. Then mathscr R_scrR_omega is an algebra closed in
 supremum norm, containing all constant sequences and all uniformly bounded
-sequences in the centralizer of $omega$, and closed under adding uniformly
-bounded sequences that converge to zero in $|cdot|_omega$. If
-$(U_n),(V_n)inmathscr R_scrR_omega$ are self-adjoint, uniformly bounded, and
-$|U_n-V_n|_omegato0$, then for every continuous $f$ on a compact interval
-containing their spectra, $(f(U_n)),(f(V_n))inmathscr R_scrR_omega$ and
-$|f(U_n)-f(V_n)|_omegato0$.
+sequences in the centralizer of omega, and closed under adding uniformly
+bounded sequences that converge to zero in |cdot|_omega. If
+(U_n),(V_n)inmathscr R_scrR_omega are self-adjoint, uniformly bounded, and
+|U_n-V_n|_omegato0, then for every continuous f on a compact interval
+containing their spectra, (f(U_n)),(f(V_n))inmathscr R_scrR_omega and
+|f(U_n)-f(V_n)|_omegato0.
 
 
 Proof. 
-Constant and centralizer sequences lie in $mathscr R_scrR_omega$ by
-Lemma~lem:state-multipliers. If $(R_n),(S_n)inmathscr R_scrR_omega$ and
-$(D_n)$ is bounded state-norm-null, then $(D_nR_n)$ is too, so
-$|D_nR_nS_n|_omegato0$; addition is immediate. Closure in supremum norm
-follows by approximating the right multiplier uniformly before the limit in $n$.
-If $(E_n)$ is bounded state-norm-null, $|D_nE_n|_omegale|D_n||E_n|_omega
-to0$, so $(R_n+E_n)inmathscr R_scrR_omega$.
-
-For polynomial $f$, membership follows from the algebra property, and differences
+Constant and centralizer sequences lie in mathscr R_scrR_omega by
+Lemma~lem:state-multipliers. If (R_n),(S_n)inmathscr R_scrR_omega and
+(D_n) is bounded state-norm-null, then (D_nR_n) is too, so
+|D_nR_nS_n|_omegato0; addition is immediate. Closure in supremum norm
+follows by approximating the right multiplier uniformly before the limit in n.
+If (E_n) is bounded state-norm-null, |D_nE_n|_omegale|D_n||E_n|_omega
+to0, so (R_n+E_n)inmathscr R_scrR_omega.
+
+For polynomial f, membership follows from the algebra property, and differences
 from the telescoping
-$U_n^j-V_n^j=sum_{h=0}^{j-1}U_n^h(U_n-V_n)V_n^{j-1-h}$: each right multiplier is
-in $mathscr R_scrR_omega$ and left multipliers are uniformly bounded, so each summand
+U_n^j-V_n^j=sum_{h=0}^{j-1}U_n^h(U_n-V_n)V_n^{j-1-h}: each right multiplier is
+in mathscr R_scrR_omega and left multipliers are uniformly bounded, so each summand
 tends to zero in state norm. Uniform polynomial approximation on the common
-spectral interval gives both assertions for continuous $f$.
+spectral interval gives both assertions for continuous f.
 
 
 ## Paving in the presence of a central state
 
 
-Let $A⊆ M$ be a masa and $varphi$ a faithful normal state on $M$ with $A$
-in its centralizer. For $0<eps<1$ put
-$r_{mathrm e}rme=lceil20000^2eps^{-2}rceil$. For every self-adjoint contraction
-$xin M$ and every $rho>0$ there are a partition $P=(p_1,...,p_{r_{mathrm e}rme})$
-in $A$ and a projection $qin M$ with
+Let A⊆ M be a masa and varphi a faithful normal state on M with A
+in its centralizer. For 0<eps<1 put
+r_{mathrm e}rme=lceil20000^2eps^{-2}rceil. For every self-adjoint contraction
+xin M and every rho>0 there are a partition P=(p_1,...,p_{r_{mathrm e}rme})
+in A and a projection qin M with
 [
- varphi(1-q)<rho,qquad |q(S_P(x)-E_Ax)q|leeps,
+ varphi(1-q)<rho, |q(S_P(x)-E_Ax)q|leeps,
 ]
-$E_A$ the $varphi$-preserving conditional expectation onto $A$.
+E_A the varphi-preserving conditional expectation onto A.
 
 
 Proof. 
 By Lemma~lem:separable-reduction it suffices to work with separable predual;
-the reduction's conditional expectation restricts $E_A$, so a diagonal choice
+the reduction's conditional expectation restricts E_A, so a diagonal choice
 there proves the stated one. Use the notation of
 Sections~sec:kernels,sec:transfer.
 
-Fix $r=r_{mathrm e}rme$ and write $a=E_Ax$, $y=E_Nx-E_Axin N$,
-$w=x-E_Nxinker E_N$. These are self-adjoint, $|a|le1$,
-$|y|,|w|le2$, and $E_Ay=0$ (uniqueness of the $varphi$-preserving
-expectation onto $A$ gives $E_AE_N=E_A$).
+Fix r=r_{mathrm e}rme and write a=E_Ax, y=E_Nx-E_Axin N,
+w=x-E_Nxinker E_N. These are self-adjoint, |a|le1,
+|y|,|w|le2, and E_Ay=0 (uniqueness of the varphi-preserving
+expectation onto A gives E_AE_N=E_A).
 
 Finite-support Cartan approximations.: 
-Choose self-adjoint contractions in $N_{alg}$ converging strongly to
-$E_Nx$: apply Kaplansky density to the norm closure of the finite normalizer
+Choose self-adjoint contractions in N_{alg} converging strongly to
+E_Nx: apply Kaplansky density to the norm closure of the finite normalizer
 algebra, approximate its self-adjoint contractions in norm by self-adjoint
-elements of $N_{alg}$, and rescale; bounded strong approximation is
+elements of N_{alg}, and rescale; bounded strong approximation is
 sequential in the faithful separable state representation. Each approximant
 involves finitely many partial isomorphisms; we also bound their edge
-Radon--Nikodym ratios: for a finite list of maps $gamma$, delete from $X$ each
-domain set where $Delta(gamma v,v)$ lies outside $[b^{-1},b]$, with $e_bin A$
-the indicator of the rest. As $bto∞$, $e_b$ increase strongly to $1$;
-compressing the approximant by $e_b$ keeps it self-adjoint and contractive and
+Radon--Nikodym ratios: for a finite list of maps gamma, delete from X each
+domain set where Delta(gamma v,v) lies outside [b^{-1},b], with e_bin A
+the indicator of the rest. As bto∞, e_b increase strongly to 1;
+compressing the approximant by e_b keeps it self-adjoint and contractive and
 restricts each surviving edge to the required ratio range (enlarge the list by
-inverses if needed). Choosing $b$ large for each approximant preserves strong
+inverses if needed). Choosing b large for each approximant preserves strong
 convergence.
 
 Subtracting the diagonal gives
 
- y_n=y_n^*in N_{alg},quad E_Ay_n=0,quad |y_n|le2,quad
+ y_n=y_n^*in N_{alg}, E_Ay_n=0, |y_n|le2,
  |y_n-y|_varphi→0.
 
-Each $y_n/2$ is a measurable zero-diagonal Hermitian contraction on orbit
-$ell^2$ spaces, supported on a finite-degree undirected graph with bounded edge
-cocycle ratios (bounds may depend on $n$, which Theorem~thm:bernoulli-coloring
-allows without affecting $r$).
+Each y_n/2 is a measurable zero-diagonal Hermitian contraction on orbit
+ell^2 spaces, supported on a finite-degree undirected graph with bounded edge
+cocycle ratios (bounds may depend on n, which Theorem~thm:bernoulli-coloring
+allows without affecting r).
 
 One partition for the two operator terms.: 
-Use the relative Bernoulli extension $widetilde N$ and amalgamated product
-$mathcal K=M*_NcalK=M*_Nwidetilde N$, $Phi=varphi|_Ncirc E_N^{mathcal K}calK$, from
+Use the relative Bernoulli extension widetilde N and amalgamated product
+mathcal K=M*_NcalK=M*_Nwidetilde N, Phi=varphi|_Ncirc E_N^{mathcal K}calK, from
 Lemma~lem:bernoulli-model. Apply Theorem~thm:bernoulli-coloring to
-$y_n/2$ with
+y_n/2 with
 
- t=frac4{sqrt r},qquad alpha=fraceps{32},qquad m=n,qquad
+ t=frac4{sqrt r}, alpha=fraceps{32}, m=n,
  eta=(fraceps{16})^{2n}.
 
-The color projections $p_1^n,...,p_r^nin B$ form a partition, lie in the
-centralizer of $Phi$, and satisfy
+The color projections p_1^n,...,p_r^nin B form a partition, lie in the
+centralizer of Phi, and satisfy
 
  widetilde E_N(p_i^n)=r^{-1}1qquad(1le ile r).
 
-With $beta=80$ the norm bound, after restoring the factor two, obeys
+With beta=80 the norm bound, after restoring the factor two, obeys
 
  2(2beta t+alpha)=(1280)/(sqrt r)+fraceps{16}
  le(253)/(2000)eps<fraceps4.
 
 
-Put $widehatwidehat Y_n=Y_n=sum_ip_i^ny_np_i^ninwidetilde N$. At a root where the
-radius-$n$ ball has the pinching bound, $widehatwidehat Y_n^n$Y_n^n$ on the root vector agrees
-with the $n$th power of that principal matrix (every length-$n$ path stays in the
-ball); elsewhere $|widehatwidehat Y_nY_n|le2$. Integrating squared vector norms,
+Put widehatwidehat Y_n=Y_n=sum_ip_i^ny_np_i^ninwidetilde N. At a root where the
+radius-n ball has the pinching bound, widehatwidehat Y_n^nY_n^n on the root vector agrees
+with the nth power of that principal matrix (every length-n path stays in the
+ball); elsewhere |widehatwidehat Y_nY_n|le2. Integrating squared vector norms,
 
  Phi(widehatwidehat Y_n^Y_n^{2n})le(fraceps4)^{2n}
        +2^{2n}(fraceps{16})^{2n},
@@ -2661,7 +2653,7 @@ ball); elsewhere $|widehatwidehat Y_nY_n|le2$. Integrating squared vector norms,
 using diagonal integration in the relation representation, not a trace on the
 nonsingular relation algebra.
 
-Set $Y_n=sum_ip_i^n y p_i^n$ and $W_n=sum_ip_i^n w p_i^n$. Pinching by a
+Set Y_n=sum_ip_i^n y p_i^n and W_n=sum_ip_i^n w p_i^n. Pinching by a
 centralizer partition is state-norm contractive, so (eq:cartan-approximants)
 gives
 
@@ -2672,35 +2664,35 @@ and by Lemma~lem:free-compression and (eq:color-marginals-completion),
  |W_n|le2(frac1r+frac2{sqrt r})
  le(40001)/(200000000)eps<fraceps4,
 
-orthogonality of the $p_i^n$ making the block-sum norm the max of block norms.
-Thus the same $r$ projections control the complementary term; no second
+orthogonality of the p_i^n making the block-sum norm the max of block norms.
+Thus the same r projections control the complementary term; no second
 refinement.
 
 Clipping before fixing a transferred moment.: 
-Let $f:mathbb RbbRto[-eps/2,eps/2]$, $f(s)=max{-eps/2,min{s,eps/2}}$.
+Let f:mathbb RbbRto[-eps/2,eps/2], f(s)=max{-eps/2,min{s,eps/2}}.
 Spectral calculus and (eq:cartan-high-moment) give
 [
  Phi(1_{{|widehatwidehat Y_n|>Y_n|>eps/2}})
  le(frac2eps)^{2n}Phi(widehatwidehat Y_n^Y_n^{2n})le2^{-2n}+4^{-2n},
 ]
-and since $|widehatwidehat Y_nY_n|le2$,
+and since |widehatwidehat Y_nY_n|le2,
 
  |widehatwidehat Y_n-f(Y_n-f(widehatwidehat Y_n)Y_n)|_Phi→0.
 
 
-Apply Lemma~lem:multiplier-calculus in $mathcal K$:calK$: each $(p_i^n)in
-mathscr R_scrR_Phi$ and $r$ is fixed, so $(Y_n)$ and $(W_n)$ (finite sums of products
-of centralizer sequences and fixed operators) lie in $mathscr R_scrR_Phi$;
-(eq:pinched-approximation) places $(widehatwidehat Y_n)$Y_n)$ there too.
+Apply Lemma~lem:multiplier-calculus in mathcal K:calK: each (p_i^n)in
+mathscr R_scrR_Phi and r is fixed, so (Y_n) and (W_n) (finite sums of products
+of centralizer sequences and fixed operators) lie in mathscr R_scrR_Phi;
+(eq:pinched-approximation) places (widehatwidehat Y_n)Y_n) there too.
 Equations~(eq:pinched-approximation)--(eq:approximate-clipping) and
 continuous calculus give
 
  |Y_n-f(Y_n)|_Phi→0.
 
 
-The self-adjoint $Z_n=Y_n+W_n=sum_ip_i^n(x-a)p_i^n$ thus differ by a bounded
-state-norm-null sequence from $Q_n=f(Y_n)+W_n$ with $|Q_n|le3eps/4$, and both
-sequences lie in $mathscr R_scrR_Phi$. Telescoping powers, for each fixed $kge1$,
+The self-adjoint Z_n=Y_n+W_n=sum_ip_i^n(x-a)p_i^n thus differ by a bounded
+state-norm-null sequence from Q_n=f(Y_n)+W_n with |Q_n|le3eps/4, and both
+sequences lie in mathscr R_scrR_Phi. Telescoping powers, for each fixed kge1,
 
  limsup_{nto∞}Phi(Z_n^{2k})le((3eps)/(4))^{2k}.
 
@@ -2708,44 +2700,44 @@ Only a fixed power is used here; (eq:cartan-approximants) needed no
 prescribed rate.
 
 Transfer and the final spectral cut.: 
-Fix $rho>0$. Choose $k$ with $(3/4)^{2k}<rho/4$, then $n$ with
-$Phi(Z_n^{2k})<tfracrho2eps^{2k}$. With $n,k$ fixed, approximate
-$(p_i^n)$ in $B$ by a finite-symbol cylinder partition (Lemma~lem:cylinder-partitions);
+Fix rho>0. Choose k with (3/4)^{2k}<rho/4, then n with
+Phi(Z_n^{2k})<tfracrho2eps^{2k}. With n,k fixed, approximate
+(p_i^n) in B by a finite-symbol cylinder partition (Lemma~lem:cylinder-partitions);
 bounded state-norm approximation of self-adjoint projections is strong-*
 approximation, so the fixed polynomial moment changes little and we keep the
 strict bound with slack.
 
 Proposition~prop:moment-transfer applies to the expansion of this moment.
 For a fine enough base partition the averaged evaluated moment stays strictly below
-$rhoeps^{2k}$. Each evaluation is a genuine projection partition in $A$ with
+rhoeps^{2k}. Each evaluation is a genuine projection partition in A with
 nonnegative even moment, so some deterministic evaluation gives a partition
-$P=(p_1,...,p_r)$ with
+P=(p_1,...,p_r) with
 [
  varphi([sum_{i=1}^r p_i(x-a)p_i]^{2k})<rhoeps^{2k}.
 ]
-Let $z=sum_ip_i(x-a)p_i=z^*$ and $q=1_{[-eps,eps]}(z)in M$. Then
-$varphi(1-q)leeps^{-2k}varphi(z^{2k})<rho$ and $|qzq|leeps$. Since $ain A$
-commutes with the partition, $z=S_P(x)-a$. The number of colors is fixed before
-$rho$.
+Let z=sum_ip_i(x-a)p_i=z^* and q=1_{[-eps,eps]}(z)in M. Then
+varphi(1-q)leeps^{-2k}varphi(z^{2k})<rho and |qzq|leeps. Since ain A
+commutes with the partition, z=S_P(x)-a. The number of colors is fixed before
+rho.
 
 
 ## Arbitrary inclusions
 
 Proof. [Proof of Theorem~thm:main]
-If $x=0$, take $p_1=1$, $p_2=...=p_{r_eps}=0$, $a=0$, $q=1$. Otherwise replace
-$x$ by $x/|x|$. Proposition~prop:expected with
+If x=0, take p_1=1, p_2=...=p_{r_eps}=0, a=0, q=1. Otherwise replace
+x by x/|x|. Proposition~prop:expected with
 Proposition~prop:state-reduction supplies the paving in the original
 representation with
 [
  r=Bigllceil400000000eps^{-2}Bigrrceil+Bigllceil4/epsBigrrceil+1
 ]
-projections, depending only on $eps$. The supporting corner and all partitions,
-diagonal elements, and compressions may depend on $F$ and $delta$ as
+projections, depending only on eps. The supporting corner and all partitions,
+diagonal elements, and compressions may depend on F and delta as
 Definition~def:so-paving permits. The diagonal has norm at most one and
 [
  rle400000000eps^{-2}+4eps^{-1}+3le400000007eps^{-2}<500000000eps^{-2}.
 ]
-Multiplying the diagonal by the original $|x|$ restores scale and gives every
+Multiplying the diagonal by the original |x| restores scale and gives every
 assertion of Theorem~thm:main.
 
 
@@ -2775,7 +2767,7 @@ Mathematica Scandinavica 37 (1975), 271--283.
 
 bibitem[Haa79]{Haa79}
 Uffe Haagerup,
-$L^p$-spaces associated with an arbitrary von Neumann algebra,
+L^p-spaces associated with an arbitrary von Neumann algebra,
 in Alg`ebres d'op'erateurs et leurs applications en physique
 math'ematique (Marseille, 1977), Colloques Internationaux du C.N.R.S.
 274, C.N.R.S., Paris, 1979, pp.~175--184.
@@ -2809,7 +2801,7 @@ OpenAI Math Release preprint, 2026, Theorem~1.1.
 
 bibitem[Pop14]{Popa14}
 Sorin Popa,
-A $II_1$ factor approach to the Kadison--Singer problem,
+A II_1 factor approach to the Kadison--Singer problem,
 Communications in Mathematical Physics 332 (2014), 379--414.
 
 bibitem[PV15]{PV15}