TheoremBase

The bound follows from |lambda|<=1 and #P(w)<=k^2-k, with continuity by an explicit head/tail epsilon-eta argument. For dissipation, each inner and outer piece is stripped to a cyclically reduced word; the cross terms are split by a weighted AM-GM with weights 24^{-(k-m)}; the second part is resummed over the stripped word using the count N(u,k)<=16k(2d)^{k-m}; and the estimate is taken to the limit from truncations.

Proof

Each result cited below is universally quantified over the data in its own statement.

Conventions. Throughout, d∈Nd\in\mathbb{N} is fixed and θ=θd=16144 d\theta=\theta_{d}=\frac{1}{6144\,d}, so that by The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §weights one has c∅=1c_{\varnothing}=1 and cw=γkc_{w}=\gamma_{k} for w∈Wd,k∘w\in W^{\circ}_{d,k}, k∈Nk\in\mathbb{N}, where γk=θk(k+1)4\gamma_{k}=\frac{\theta^{k}}{(k+1)^{4}}; by the same clause 0<γk≤θk0<\gamma_{k}\le\theta^{k}. In particular the weight of a nonempty cyclically reduced word depends only on its length. For a word ww of length kk and p∈[k]p\in[k] we write εp=ε(wp)\varepsilon_{p}=\varepsilon(w_{p}), and we use P(w)P(w), σp\sigma_{p}, τq\tau_{q}, ypq=ypq(w)y_{pq}=y_{pq}(w) and xpq=xpq(w)x_{pq}=x_{pq}(w) as in The Free Unitary Heat Generator on Unitary Laws §positions and The Free Unitary Heat Generator on Unitary Laws §pieces. A sum over an empty index set is read as 00. For finite sums of real numbers the following rules are used without further mention: termwise inequalities pass to the sums by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §comparison; a sum of nonnegative terms is nonnegative and does not decrease when the index set is enlarged, by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §nonnegative and Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §monotone; sums are additive and homogeneous, may be reindexed along bijections and agree with sums over a numerical range, by claims 3, 4, 2 and 1 of Properties of a Sum over a Finite Index Set; they split over disjoint unions by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §disjoint-union; iterated sums are sums over dependent pairs by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs; and terms equal to 00 may be dropped by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing. For complex numbers we use the multiplicativity of the modulus and the triangle inequality, claims 4 and 7 of Properties of Complex Conjugation and Modulus, and the fact that the modulus of a finite sum is at most the sum of the moduli, Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §modulus. Since every sign is 11 or −1-1, claim 8 of Properties of Complex Conjugation and Modulus gives ∣εp∣=1|\varepsilon_{p}|=1.

Step 0 (elementary facts).

(E1) For every letter l∈[2d]l\in[2d] one has (l−1)−1=l(l^{-1})^{-1}=l; moreover ε(l)=1\varepsilon(l)=1 if and only if l≤dl\le d, in which case g(l)=lg(l)=l, and if ε(l)=−1\varepsilon(l)=-1 then l=d+g(l)l=d+g(l). All of this is read off Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters: if l≤dl\le d, then ε(l)=1\varepsilon(l)=1, g(l)=lg(l)=l and l−1=d+ll^{-1}=d+l, and since d<d+ld<d+l the inverse letter of d+ld+l is the unique jj with d+l=d+jd+l=d+j, that is ll; if d<ld<l, then l=d+jl=d+j with j=g(l)∈[d]j=g(l)\in[d], ε(l)=−1\varepsilon(l)=-1, l−1=j≤dl^{-1}=j\le d, and j−1=d+j=lj^{-1}=d+j=l.

(E2) Let r∈Rr\in\mathbb{R} with 0≤r≤10\le r\le1 and let n<n′n<n' be natural numbers. Then rn′=rn rn′−nr^{n'}=r^{n}\,r^{n'-n}, by induction on n′−nn'-n from the recursion, claim 1 of Properties of Natural Number Powers in a Field; and 0≤rn0\le r^{n}, 0≤rn′−n≤1n′−n=10\le r^{n'-n}\le1^{n'-n}=1 by claims 5 and 2 of that lemma. Hence rn′≤rnr^{n'}\le r^{n} whenever n≤n′n\le n'.

(E3) For every k∈Nk\in\mathbb{N} one has k≤2kk\le2^{k} and k4≤16kk^{4}\le16^{k}. Indeed 1≤2=211\le2=2^{1}, and if k≤2kk\le2^{k} then, as 1=1k≤2k1=1^{k}\le2^{k} by claims 2 and 5 of Properties of Natural Number Powers in a Field, k+1≤2k+2k=2k+1k+1\le2^{k}+2^{k}=2^{k+1} by claim 1 of that lemma; this is an induction on kk. By claim 5 of that lemma, k4≤(2k)4k^{4}\le(2^{k})^{4}, and (2k)4=2k2k2k2k=4k4k=16k(2^{k})^{4}=2^{k}2^{k}2^{k}2^{k}=4^{k}4^{k}=16^{k} by claims 1 and 3 of that lemma. Likewise 1+k≤2k1+k\le2^{k} for every k∈Nk\in\mathbb{N}: 1+1=211+1=2^{1}, and if 1+k≤2k1+k\le2^{k} then 1+(k+1)≤2k+1≤2k+2k=2k+11+(k+1)\le2^{k}+1\le2^{k}+2^{k}=2^{k+1}. Consequently, for m,r∈Nm,r\in\mathbb{N}, m+r≤m+mr=m(1+r)≤m 2rm+r\le m+mr=m(1+r)\le m\,2^{r}, where r≤mrr\le mr because 1≤m1\le m and 0≤r0\le r.

(E4) Let α,β∈R\alpha,\beta\in\mathbb{R} and let a,a′∈Ra,a'\in\mathbb{R} be positive with a a′=1a\,a'=1. Then αβ≤12aα2+12a′β2\alpha\beta\le\frac12a\alpha^{2}+\frac12a'\beta^{2}, because aα2−2αβ+a′β2=a′(aα−β)2≥0a\alpha^{2}-2\alpha\beta+a'\beta^{2}=a'(a\alpha-\beta)^{2}\ge0, as one checks by expanding and using aa′=1aa'=1.

(E5) Let s∈Rs\in\mathbb{R} with 0≤s<120\le s<\frac12 and let n∈Nn\in\mathbb{N}. Then

∑r=1nsr≤s1−s.\sum_{r=1}^{n}s^{r}\le\frac{s}{1-s}.

This holds because by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric the left side equals s−sn+11−s\frac{s-s^{n+1}}{1-s} with 1−s>01-s>0 and 0≤sn+10\le s^{n+1}.

(E6) For k∈Nk\in\mathbb{N}, ∑i∈[k]1=∑i=1k1=k\sum_{i\in[k]}1=\sum_{i=1}^{k}1=k: the first equality is claim 1 of Properties of a Sum over a Finite Index Set, and the finite sum of kk ones is, by definition, the image of kk in R\mathbb{R} under the canonical map.

(E7) First, 0<θ≤10<\theta\le1: θ\theta is positive by The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §weights, and θ≤1\theta\le1 by Order Reversal under Reciprocals, and Summability of the Reciprocals of the Squares §reciprocal applied to 0<1≤6144 d0<1\le6144\,d. Next, let n,n′∈Nn,n'\in\mathbb{N} with n≤n′n\le n'. Then 0<n+1≤n′+10<n+1\le n'+1, so 0≤(n+1)4≤(n′+1)40\le(n+1)^{4}\le(n'+1)^{4} by claim 5 of Properties of Natural Number Powers in a Field and (n+1)4≠0(n+1)^{4}\ne0 by claim 4 of that lemma; hence Order Reversal under Reciprocals, and Summability of the Reciprocals of the Squares §reciprocal gives

0<1(n′+1)4≤1(n+1)4.0<\frac{1}{(n'+1)^{4}}\le\frac{1}{(n+1)^{4}}.

Multiplying inequalities between nonnegative reals, we obtain: (i) if n≤Kn\le K for some K∈NK\in\mathbb{N}, then θK(K+1)4≤γn\frac{\theta^{K}}{(K+1)^{4}}\le\gamma_{n}, since 0≤θK≤θn0\le\theta^{K}\le\theta^{n} by (E2) and claim 5 of Properties of Natural Number Powers in a Field; (ii) if n<n′n<n', then γn′≤θ n′−nγn\gamma_{n'}\le\theta^{\,n'-n}\gamma_{n}, since γn′=θ n′−n θn 1(n′+1)4\gamma_{n'}=\theta^{\,n'-n}\,\theta^{n}\,\frac{1}{(n'+1)^{4}} by the factorisation in (E2), and 1(n′+1)4≤1(n+1)4\frac{1}{(n'+1)^{4}}\le\frac{1}{(n+1)^{4}} with 0≤θ n′−nθn0\le\theta^{\,n'-n}\theta^{n}.

Step 1 (the number of pairs, and claim 1). Let w∈W2dw\in W_{2d} have length k∈Nk\in\mathbb{N}. We show

∑(p,q)∈P(w)1≤k2−k.(1.1)\sum_{(p,q)\in P(w)}1\le k^{2}-k.\tag{1.1}

Let Dk={(p,p):p∈[k]}D_{k}=\{(p,p):p\in[k]\} and Qk=([k]×[k])∖DkQ_{k}=([k]\times[k])\setminus D_{k}. Since p<qp<q excludes p=qp=q, P(w)⊆QkP(w)\subseteq Q_{k}. By The Product of Two Sums over Finite Index Sets is a Sum over the Cartesian Product with aa and bb constantly 11, and (E6), ∑[k]×[k]1=k⋅k\sum_{[k]\times[k]}1=k\cdot k; and p↦(p,p)p\mapsto(p,p) is a bijection from [k][k] onto DkD_{k}, so ∑Dk1=k\sum_{D_{k}}1=k. If k=1k=1, then QkQ_{k} and hence P(w)P(w) are empty, and (1.1) reads 0≤00\le0. If 1<k1<k, then QkQ_{k} is nonempty (it contains (1,2)(1,2)), and splitting [k]×[k]=Qk∪Dk[k]\times[k]=Q_{k}\cup D_{k} gives ∑Qk1=k2−k\sum_{Q_{k}}1=k^{2}-k, whence (1.1) by monotonicity in the index set if P(w)P(w) is nonempty; if P(w)P(w) is empty, the left side of (1.1) is 00, and 0≤∑Qk1=k2−k0\le\sum_{Q_{k}}1=k^{2}-k by nonnegativity.

Claim 1. Let λ∈Ld\lambda\in\mathcal{L}_{d} and w∈W2dw\in W_{2d}. If w=∅w=\varnothing, then Θλ(∅)=0\Theta\lambda(\varnothing)=0 by The Free Unitary Heat Generator on Unitary Laws §generator, and ∣0∣=0=∣∅∣2|0|=0=|\varnothing|^{2} by claim 3 of Properties of Complex Conjugation and Modulus and Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §length. If ww has length k∈Nk\in\mathbb{N}, then by The Free Unitary Heat Generator on Unitary Laws §generator, the triangle inequality, the modulus of a finite sum, multiplicativity, ∣εp∣=∣εq∣=1|\varepsilon_{p}|=|\varepsilon_{q}|=1, ∣λ(v)∣≤1|\lambda(v)|\le1 for all v∈W2dv\in W_{2d} (Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §bound) and (1.1),

∣Θλ(w)∣≤k2∣λ(w)∣+∑(p,q)∈P(w)∣λ(ypq)∣ ∣λ(xpq)∣≤k2+k2−k≤k2=∣w∣2.|\Theta\lambda(w)|\le\frac{k}{2}|\lambda(w)|+\sum_{(p,q)\in P(w)}|\lambda(y_{pq})|\,|\lambda(x_{pq})|\le\frac{k}{2}+k^{2}-k\le k^{2}=|w|^{2}.

Now let k∈Nk\in\mathbb{N} and w∈Wd,k∘w\in W^{\circ}_{d,k}. Since 0≤∣Θλ(w)∣≤k20\le|\Theta\lambda(w)|\le k^{2}, Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field (weak form) gives ∣Θλ(w)∣2≤k4|\Theta\lambda(w)|^{2}\le k^{4}, so, as 0<cw≤θk0<c_{w}\le\theta^{k} (Conventions), 0≤cw∣Θ^λ(w)∣2≤θkk40\le c_{w}|\widehat{\Theta}\lambda(w)|^{2}\le\theta^{k}k^{4} by The Free Unitary Heat Generator on Unitary Laws §hat. The set Wd,k∘W^{\circ}_{d,k} is nonempty by Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §finite, so by Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §count with M=θkk4M=\theta^{k}k^{4}, claim 3 of Properties of Natural Number Powers in a Field and (E3), the kk-th block sum of w↦cw∣Θ^λ(w)∣2w\mapsto c_{w}|\widehat{\Theta}\lambda(w)|^{2} is at most

k4θk(2d)k=k4(2d θ)k=k4(13072)k≤16k(13072)k=(1192)k,(1.2)k^{4}\theta^{k}(2d)^{k}=k^{4}(2d\,\theta)^{k}=k^{4}\Bigl(\frac{1}{3072}\Bigr)^{k}\le16^{k}\Bigl(\frac{1}{3072}\Bigr)^{k}=\Bigl(\frac{1}{192}\Bigr)^{k},\tag{1.2}

and it is nonnegative. The geometric series of ratio 1192\frac{1}{192} converges by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric, so the series of block sums converges by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §comparison; that is, Θ^λ∈Ed\widehat{\Theta}\lambda\in E_{d} by The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §sums and The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §space. This proves claim 1.

Step 2 (the pieces of a cyclically reduced word). Let k∈Nk\in\mathbb{N}, w∈Wd,k∘w\in W^{\circ}_{d,k} and (p,q)∈P(w)(p,q)\in P(w). Then 1≤p<q≤k1\le p<q\le k, so 1<k1<k, and by The Free Unitary Heat Generator on Unitary Laws §positions

σp≤p+1≤q≤τq+1,τq≤q≤k,p≤σp,q−1≤τq,\sigma_{p}\le p+1\le q\le\tau_{q}+1,\qquad\tau_{q}\le q\le k,\qquad p\le\sigma_{p},\qquad q-1\le\tau_{q},

the last two because σp∈{p,p+1}\sigma_{p}\in\{p,p+1\} and τq∈{q−1,q}\tau_{q}\in\{q-1,q\}. We show:

(2a) σp≤τq\sigma_{p}\le\tau_{q}; hence ypq=w[σp,τq]y_{pq}=w_{[\sigma_{p},\tau_{q}]} has length ∣ypq∣=τq−σp+1≥1|y_{pq}|=\tau_{q}-\sigma_{p}+1\ge1.

(2b) ∣xpq∣=k−∣ypq∣|x_{pq}|=k-|y_{pq}| and 1≤∣xpq∣1\le|x_{pq}|; hence 1≤∣ypq∣≤k−11\le|y_{pq}|\le k-1 and 1≤∣xpq∣≤k−11\le|x_{pq}|\le k-1.

(2c) q−p−1≤∣ypq∣q-p-1\le|y_{pq}| and k−q+p−1≤∣xpq∣k-q+p-1\le|x_{pq}|, where q−p−1≥0q-p-1\ge0 and k−q+p−1≥0k-q+p-1\ge0.

(2d) ypqy_{pq} and xpqx_{pq} are reduced.

(2e) σp\sigma_{p} is determined by pp and εp\varepsilon_{p}, τq\tau_{q} by qq and εq\varepsilon_{q}, and conversely q=τqq=\tau_{q} if εq=1\varepsilon_{q}=1 and q=τq+1q=\tau_{q}+1 if εq=−1\varepsilon_{q}=-1.

Proof of (2a): otherwise σp=τq+1\sigma_{p}=\tau_{q}+1 by the displayed chain, and then all its inequalities are equalities: σp=p+1\sigma_{p}=p+1, so εp=1\varepsilon_{p}=1; q=p+1q=p+1; and τq=q−1\tau_{q}=q-1, so εq=−1\varepsilon_{q}=-1. By (E1), wp=g(wp)w_{p}=g(w_{p}) and wq=d+g(wq)=d+g(wp)=d+wp=(wp)−1w_{q}=d+g(w_{q})=d+g(w_{p})=d+w_{p}=(w_{p})^{-1}, using g(wp)=g(wq)g(w_{p})=g(w_{q}) from the definition of P(w)P(w). As p<kp<k and q=p+1q=p+1, this contradicts the reducedness of ww (Reduced and Cyclically Reduced Words in Unitary Letters §reduced, Reduced and Cyclically Reduced Words in Unitary Letters §cyclically-reduced). The length formula is the definition of w[a,b]w_{[a,b]} in The Free Unitary Heat Generator on Unitary Laws.

Proof of (2b): by The Free Unitary Heat Generator on Unitary Laws §pieces, xpqx_{pq} is the concatenation of w[τq+1,k]w_{[\tau_{q}+1,k]}, of length k−τqk-\tau_{q} (read as 00 if τq=k\tau_{q}=k), and w[1,σp−1]w_{[1,\sigma_{p}-1]}, of length σp−1\sigma_{p}-1 (read as 00 if σp=1\sigma_{p}=1). Hence ∣xpq∣=k−τq+σp−1=k−∣ypq∣|x_{pq}|=k-\tau_{q}+\sigma_{p}-1=k-|y_{pq}| by (2a). Suppose ∣xpq∣=0|x_{pq}|=0. Then τq=k\tau_{q}=k and σp=1\sigma_{p}=1. Since τq≤q≤k\tau_{q}\le q\le k, q=k=τqq=k=\tau_{q}, so εk=1\varepsilon_{k}=1 (τq=q−1\tau_{q}=q-1 would give τq<k\tau_{q}<k); and σp=1\sigma_{p}=1 excludes σp=p+1≥2\sigma_{p}=p+1\ge2, so εp=−1\varepsilon_{p}=-1 and p=σp=1p=\sigma_{p}=1. By (E1), wk=g(wk)=g(w1)w_{k}=g(w_{k})=g(w_{1}) and w1=d+g(w1)w_{1}=d+g(w_{1}), so (w1)−1=g(w1)=wk(w_{1})^{-1}=g(w_{1})=w_{k}; since 1<k1<k this contradicts Reduced and Cyclically Reduced Words in Unitary Letters §cyclically-reduced. So 1≤∣xpq∣1\le|x_{pq}|, and the bounds follow from ∣ypq∣+∣xpq∣=k|y_{pq}|+|x_{pq}|=k.

Proof of (2c): ∣ypq∣=τq−σp+1≥(q−1)−(p+1)+1=q−p−1|y_{pq}|=\tau_{q}-\sigma_{p}+1\ge(q-1)-(p+1)+1=q-p-1 and ∣ypq∣≤q−p+1|y_{pq}|\le q-p+1, so ∣xpq∣=k−∣ypq∣≥k−q+p−1|x_{pq}|=k-|y_{pq}|\ge k-q+p-1; the nonnegativity follows from p<qp<q, q≤kq\le k and 1≤p1\le p.

Proof of (2d): the ii-th and (i+1)(i+1)-st letters of ypqy_{pq} are wσp+i−1w_{\sigma_{p}+i-1} and wσp+iw_{\sigma_{p}+i}, consecutive letters of the reduced word ww, so ypqy_{pq} is reduced. Two consecutive letters of xpqx_{pq} are consecutive letters of ww unless both factors of xpqx_{pq} are nonempty and the two letters are the last letter wkw_{k} of the first factor and the first letter w1w_{1} of the second; in that case w1≠(wk)−1w_{1}\ne(w_{k})^{-1}, since otherwise (w1)−1=((wk)−1)−1=wk(w_{1})^{-1}=((w_{k})^{-1})^{-1}=w_{k} by (E1), contradicting Reduced and Cyclically Reduced Words in Unitary Letters §cyclically-reduced (here 1<k1<k). So xpqx_{pq} is reduced.

(2e) is the definition of σp\sigma_{p} and τq\tau_{q} in The Free Unitary Heat Generator on Unitary Laws §positions.

Step 3 (stripping the pieces, and the basic difference estimate). For k∈Nk\in\mathbb{N}, w∈Wd,k∘w\in W^{\circ}_{d,k}, (p,q)∈P(w)(p,q)\in P(w) and ι∈{in,out}\iota\in\{\mathrm{in},\mathrm{out}\} put zpqin(w)=ypq(w)z^{\mathrm{in}}_{pq}(w)=y_{pq}(w) and zpqout(w)=xpq(w)z^{\mathrm{out}}_{pq}(w)=x_{pq}(w). Each zpqι(w)z^{\iota}_{pq}(w) is reduced by (2d), so by Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §strip-exists we may choose, and fix once and for all, words bpqι(w)∈W2db^{\iota}_{pq}(w)\in W_{2d} and upqι(w)∈Wd∘u^{\iota}_{pq}(w)\in W^{\circ}_{d} with

zpqι(w)=bpqι(w) upqι(w) bpqι(w)∗,∣zpqι(w)∣=2∣bpqι(w)∣+∣upqι(w)∣.z^{\iota}_{pq}(w)=b^{\iota}_{pq}(w)\,u^{\iota}_{pq}(w)\,b^{\iota}_{pq}(w)^{*},\qquad|z^{\iota}_{pq}(w)|=2|b^{\iota}_{pq}(w)|+|u^{\iota}_{pq}(w)|.

Then:

(3a) ∣upqι(w)∣≤∣zpqι(w)∣≤k−1|u^{\iota}_{pq}(w)|\le|z^{\iota}_{pq}(w)|\le k-1 by (2b); consequently, using ∣ypq∣+∣xpq∣=k|y_{pq}|+|x_{pq}|=k, the natural numbers k−∣upqin(w)∣k-|u^{\mathrm{in}}_{pq}(w)| and k−∣upqout(w)∣k-|u^{\mathrm{out}}_{pq}(w)| satisfy k−∣upqin(w)∣≥∣xpq∣≥1k-|u^{\mathrm{in}}_{pq}(w)|\ge|x_{pq}|\ge1 and k−∣upqout(w)∣≥∣ypq∣≥1k-|u^{\mathrm{out}}_{pq}(w)|\ge|y_{pq}|\ge1.

(3b) λ(zpqι(w))=λ(upqι(w))\lambda(z^{\iota}_{pq}(w))=\lambda(u^{\iota}_{pq}(w)) for every λ∈Ld\lambda\in\mathcal{L}_{d}, by Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §strip-invariance.

(3c) Let μ,ν∈Ld\mu,\nu\in\mathcal{L}_{d} and put Δv=μ(v)−ν(v)\Delta_{v}=\mu(v)-\nu(v) for v∈W2dv\in W_{2d}; then Δ∅=1−1=0\Delta_{\varnothing}=1-1=0 by Laws of d-Tuples of Unitaries §normalised. For k∈Nk\in\mathbb{N} and w∈Wd,k∘w\in W^{\circ}_{d,k} put

R(w)=∑(p,q)∈P(w)(∣Δupqin(w)∣+∣Δupqout(w)∣).R(w)=\sum_{(p,q)\in P(w)}\Bigl(\bigl|\Delta_{u^{\mathrm{in}}_{pq}(w)}\bigr|+\bigl|\Delta_{u^{\mathrm{out}}_{pq}(w)}\bigr|\Bigr).

Then

∣Θμ(w)−Θν(w)+k2Δw∣≤R(w).(3.1)\Bigl|\Theta\mu(w)-\Theta\nu(w)+\frac{k}{2}\Delta_{w}\Bigr|\le R(w).\tag{3.1}

Indeed, by The Free Unitary Heat Generator on Unitary Laws §generator and additivity of finite sums, Θμ(w)−Θν(w)+k2Δw=−∑(p,q)∈P(w)εpεq(μ(ypq)μ(xpq)−ν(ypq)ν(xpq))\Theta\mu(w)-\Theta\nu(w)+\frac{k}{2}\Delta_{w}=-\sum_{(p,q)\in P(w)}\varepsilon_{p}\varepsilon_{q}\bigl(\mu(y_{pq})\mu(x_{pq})-\nu(y_{pq})\nu(x_{pq})\bigr) (both sides 00 if P(w)P(w) is empty). By field arithmetic and (3b),

μ(ypq)μ(xpq)−ν(ypq)ν(xpq)=Δypq μ(xpq)+ν(ypq) Δxpq=Δupqin(w) μ(xpq)+ν(ypq) Δupqout(w),\mu(y_{pq})\mu(x_{pq})-\nu(y_{pq})\nu(x_{pq})=\Delta_{y_{pq}}\,\mu(x_{pq})+\nu(y_{pq})\,\Delta_{x_{pq}}=\Delta_{u^{\mathrm{in}}_{pq}(w)}\,\mu(x_{pq})+\nu(y_{pq})\,\Delta_{u^{\mathrm{out}}_{pq}(w)},

whose modulus is at most ∣Δupqin(w)∣+∣Δupqout(w)∣|\Delta_{u^{\mathrm{in}}_{pq}(w)}|+|\Delta_{u^{\mathrm{out}}_{pq}(w)}| by the triangle inequality, multiplicativity and Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §bound. The modulus of the sum, with ∣εpεq∣=1|\varepsilon_{p}\varepsilon_{q}|=1, gives (3.1).

Step 4 (claim 2). By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §metric, (Ld,dL)(\mathcal{L}_{d},d_{\mathcal{L}}) is a metric space; by The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §inner-product, EdE_{d} is a real inner product space with norm ∥⋅∥d\lVert\cdot\rVert_{d}, so (x,y)↦∥x−y∥d(x,y)\mapsto\lVert x-y\rVert_{d} is a metric on EdE_{d} by The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity §metric. We verify continuity at an arbitrary λ∈Ld\lambda\in\mathcal{L}_{d}. Fix λ∈Ld\lambda\in\mathcal{L}_{d} and a real ϵ>0\epsilon>0. We choose, in this order, first K∈NK\in\mathbb{N} depending only on ϵ\epsilon, and then η>0\eta>0 depending only on ϵ\epsilon and KK.

Choice of KK. By Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric the sequence ((1192)n)n∈N((\frac{1}{192})^{n})_{n\in\mathbb{N}} converges to 00, so there is K∈NK\in\mathbb{N} with (1192)n<191 ϵ21536(\frac{1}{192})^{n}<\frac{191\,\epsilon^{2}}{1536} for all n≥Kn\ge K; in particular

768191(1192)K+1<ϵ22.(4.1)\frac{768}{191}\Bigl(\frac{1}{192}\Bigr)^{K+1}<\frac{\epsilon^{2}}{2}.\tag{4.1}

Choice of η\eta. Put Λ=(6144 d)K(K+1)4\Lambda=(6144\,d)^{K}(K+1)^{4}, which is positive by claims 5 and 4 of Properties of Natural Number Powers in a Field, C=4K5Λ (2d)KC=4K^{5}\Lambda\,(2d)^{K}, which is positive, and η=min⁡{1,ϵ22C}>0\eta=\min\{1,\frac{\epsilon^{2}}{2C}\}>0.

Now let μ∈Ld\mu\in\mathcal{L}_{d} with D:=dL(λ,μ)<ηD:=d_{\mathcal{L}}(\lambda,\mu)<\eta. Apply (3c) to the pair (λ,μ)(\lambda,\mu) in place of (μ,ν)(\mu,\nu), so that Δv=λ(v)−μ(v)\Delta_{v}=\lambda(v)-\mu(v); by The Embedding of Unitary Laws into the Word Gauge Space and the Gauge Distance §distance and The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §sobolev, D=∥ιd(λ)−ιd(μ)∥dD=\lVert\iota_{d}(\lambda)-\iota_{d}(\mu)\rVert_{d} where ιd(λ)−ιd(μ)\iota_{d}(\lambda)-\iota_{d}(\mu) is the map v↦Δvv\mapsto\Delta_{v} on Wd∘W^{\circ}_{d}. Let g=Θ^λ−Θ^μ∈Edg=\widehat{\Theta}\lambda-\widehat{\Theta}\mu\in E_{d} (claim 1 and The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §space), the map v↦Θλ(v)−Θμ(v)v\mapsto\Theta\lambda(v)-\Theta\mu(v) on Wd∘W^{\circ}_{d}; g(∅)=0g(\varnothing)=0 by The Free Unitary Heat Generator on Unitary Laws §generator. For k∈Nk\in\mathbb{N} put Hk=∑w∈Wd,k∘cw∣g(w)∣2H_{k}=\sum_{w\in W^{\circ}_{d,k}}c_{w}|g(w)|^{2} and Xk=∑w∈Wd,k∘cw∣Δw∣2X_{k}=\sum_{w\in W^{\circ}_{d,k}}c_{w}|\Delta_{w}|^{2}, both nonnegative. By The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §norm and The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §sums, the square of a nonnegative square root being the number itself, and g(∅)=0=Δ∅g(\varnothing)=0=\Delta_{\varnothing},

∥g∥d2=∑k=1∞Hk,D2=∑k=1∞Xk.\lVert g\rVert_{d}^{2}=\sum_{k=1}^{\infty}H_{k},\qquad D^{2}=\sum_{k=1}^{\infty}X_{k}.

Tail. Let k∈Nk\in\mathbb{N} and w∈Wd,k∘w\in W^{\circ}_{d,k}. By the triangle inequality and claim 1, ∣g(w)∣≤∣Θλ(w)∣+∣Θμ(w)∣≤2k2|g(w)|\le|\Theta\lambda(w)|+|\Theta\mu(w)|\le2k^{2}, so ∣g(w)∣2≤4k4|g(w)|^{2}\le4k^{4} by Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field; by Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §count and the computation (1.2), 0≤Hk≤4(1192)k0\le H_{k}\le4(\frac{1}{192})^{k}. Apply Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §tail-bound with μk=(1192)k\mu_{k}=(\frac{1}{192})^{k}, whose series converges by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric, M=4M=4 and wk=192kHk∈[0,4]w_{k}=192^{k}H_{k}\in[0,4] (note (1192)k192k=1k=1(\frac{1}{192})^{k}192^{k}=1^{k}=1 by claims 3 and 2 of Properties of Natural Number Powers in a Field); with the tail formula of Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric and (4.1),

∑k=1∞Hk−∑k=1KHk≤4⋅(1192)K+11−1192=768191(1192)K+1<ϵ22.\sum_{k=1}^{\infty}H_{k}-\sum_{k=1}^{K}H_{k}\le4\cdot\frac{(\frac{1}{192})^{K+1}}{1-\frac{1}{192}}=\frac{768}{191}\Bigl(\frac{1}{192}\Bigr)^{K+1}<\frac{\epsilon^{2}}{2}.

Head. Let v∈Wd∘v\in W^{\circ}_{d} with 1≤∣v∣=n≤K1\le|v|=n\le K. Then cv∣Δv∣2≤Xn≤∑j=1nXj≤∑j=1∞Xj=D2c_{v}|\Delta_{v}|^{2}\le X_{n}\le\sum_{j=1}^{n}X_{j}\le\sum_{j=1}^{\infty}X_{j}=D^{2}, by monotonicity in the index set, nonnegativity of the XjX_{j} and Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates. By (E7)(i), θK(K+1)4≤γn=cv\frac{\theta^{K}}{(K+1)^{4}}\le\gamma_{n}=c_{v}, so θK(K+1)4∣Δv∣2≤D2\frac{\theta^{K}}{(K+1)^{4}}|\Delta_{v}|^{2}\le D^{2}; multiplying by Λ>0\Lambda>0 and using θK(K+1)4Λ=θK(6144 d)K=(6144 d θ)K=1K=1\frac{\theta^{K}}{(K+1)^{4}}\Lambda=\theta^{K}(6144\,d)^{K}=(6144\,d\,\theta)^{K}=1^{K}=1 (claims 3 and 2 of Properties of Natural Number Powers in a Field) gives ∣Δv∣2≤ΛD2|\Delta_{v}|^{2}\le\Lambda D^{2}, which also holds for v=∅v=\varnothing as Δ∅=0\Delta_{\varnothing}=0. Let e≥0e\ge0 be the nonnegative square root of ΛD2\Lambda D^{2} (Existence and Uniqueness of the Nonnegative Square Root); by Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field (weak form), ∣Δv∣≤e|\Delta_{v}|\le e for every v∈Wd∘v\in W^{\circ}_{d} with ∣v∣≤K|v|\le K.

Let k∈[K]k\in[K] and w∈Wd,k∘w\in W^{\circ}_{d,k}. Every upqι(w)u^{\iota}_{pq}(w) lies in Wd∘W^{\circ}_{d} and has length at most k−1<Kk-1<K by (3a), so R(w)≤∑(p,q)∈P(w)2e≤2(k2−k)eR(w)\le\sum_{(p,q)\in P(w)}2e\le2(k^{2}-k)e by (1.1). By (3.1) and the triangle inequality, ∣g(w)∣≤k2∣Δw∣+R(w)≤(k2+2k2−2k)e≤2k2e|g(w)|\le\frac{k}{2}|\Delta_{w}|+R(w)\le(\frac{k}{2}+2k^{2}-2k)e\le2k^{2}e, hence ∣g(w)∣2≤4k4e2=4k4ΛD2≤4K4ΛD2|g(w)|^{2}\le4k^{4}e^{2}=4k^{4}\Lambda D^{2}\le4K^{4}\Lambda D^{2} by Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field and claim 5 of Properties of Natural Number Powers in a Field. Since 0≤θ≤10\le\theta\le1 by (E7), claims 5 and 2 of Properties of Natural Number Powers in a Field and the Conventions give 0<cw≤θk≤1k=10<c_{w}\le\theta^{k}\le1^{k}=1, so 0≤cw∣g(w)∣2≤∣g(w)∣2≤4K4ΛD20\le c_{w}|g(w)|^{2}\le|g(w)|^{2}\le4K^{4}\Lambda D^{2}. By Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §count, Hk≤4K4ΛD2(2d)k≤4K4Λ(2d)KD2H_{k}\le4K^{4}\Lambda D^{2}(2d)^{k}\le4K^{4}\Lambda(2d)^{K}D^{2}, where (2d)k≤(2d)K(2d)^{k}\le(2d)^{K} because (2d)K=(2d)k(2d)K−k(2d)^{K}=(2d)^{k}(2d)^{K-k} for k<Kk<K and (2d)K−k≥1K−k=1(2d)^{K-k}\ge1^{K-k}=1 (as in (E2), with claim 5 of Properties of Natural Number Powers in a Field). Summing over k∈[K]k\in[K] with (E6),

∑k=1KHk≤K⋅4K4Λ(2d)KD2=C D2.\sum_{k=1}^{K}H_{k}\le K\cdot4K^{4}\Lambda(2d)^{K}D^{2}=C\,D^{2}.

Since 0≤D<η0\le D<\eta and 0<η≤10<\eta\le1, Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field (strict form) gives D2<η2≤η≤ϵ22CD^{2}<\eta^{2}\le\eta\le\frac{\epsilon^{2}}{2C}, so ∑k=1KHk≤ϵ22\sum_{k=1}^{K}H_{k}\le\frac{\epsilon^{2}}{2}.

Conclusion. Adding head and tail, ∥g∥d2<ϵ22+ϵ22=ϵ2\lVert g\rVert_{d}^{2}<\frac{\epsilon^{2}}{2}+\frac{\epsilon^{2}}{2}=\epsilon^{2}, so ∥g∥d<ϵ\lVert g\rVert_{d}<\epsilon by Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field (strict form), both numbers being nonnegative. By symmetry of the metric of EdE_{d}, ∥Θ^μ−Θ^λ∥d=∥g∥d<ϵ\lVert\widehat{\Theta}\mu-\widehat{\Theta}\lambda\rVert_{d}=\lVert g\rVert_{d}<\epsilon. As μ\mu with dL(λ,μ)<ηd_{\mathcal{L}}(\lambda,\mu)<\eta was arbitrary, λ↦Θ^λ\lambda\mapsto\widehat{\Theta}\lambda is continuous at λ\lambda relative to Ld\mathcal{L}_{d}, and, λ\lambda being arbitrary, continuous on Ld\mathcal{L}_{d} (Continuous Map Between Metric Spaces). This proves claim 2.

Step 5 (row sums). Put ρ=124\rho=\frac{1}{24}, and for j∈{0}∪Nj\in\{0\}\cup\mathbb{N} let ρj=ρ\rho_{j}=\rho if j=0j=0 and ρj=ρj\rho_{j}=\rho^{j} if j∈Nj\in\mathbb{N}. For k∈Nk\in\mathbb{N} and w∈Wd,k∘w\in W^{\circ}_{d,k} put, with the exponents natural numbers by (3a),

S(w)=∑(p,q)∈P(w)(ρ k−∣upqin(w)∣+ρ k−∣upqout(w)∣).S(w)=\sum_{(p,q)\in P(w)}\Bigl(\rho^{\,k-|u^{\mathrm{in}}_{pq}(w)|}+\rho^{\,k-|u^{\mathrm{out}}_{pq}(w)|}\Bigr).

We show

S(w)≤47 k276.(5.1)S(w)\le\frac{47\,k}{276}.\tag{5.1}

If P(w)P(w) is empty, S(w)=0S(w)=0. Otherwise let (p,q)∈P(w)(p,q)\in P(w). By (3a) and (E2), ρ k−∣upqin(w)∣≤ρ∣xpq∣\rho^{\,k-|u^{\mathrm{in}}_{pq}(w)|}\le\rho^{|x_{pq}|} and ρ k−∣upqout(w)∣≤ρ∣ypq∣\rho^{\,k-|u^{\mathrm{out}}_{pq}(w)|}\le\rho^{|y_{pq}|}. By (2b) and (2c), ∣xpq∣≥1|x_{pq}|\ge1 and ∣xpq∣≥k−q+p−1|x_{pq}|\ge k-q+p-1, so (E2) gives ρ∣xpq∣≤ρk−q+p−1\rho^{|x_{pq}|}\le\rho_{k-q+p-1} (compare with ρk−q+p−1\rho^{k-q+p-1} if k−q+p−1≥1k-q+p-1\ge1, and with ρ1\rho^{1} if k−q+p−1=0k-q+p-1=0); likewise ∣ypq∣≥1|y_{pq}|\ge1 and ∣ypq∣≥q−p−1|y_{pq}|\ge q-p-1 give ρ∣ypq∣≤ρq−p−1\rho^{|y_{pq}|}\le\rho_{q-p-1}. The set P(w)P(w) is contained in the set of pairs (p,q)(p,q) with p∈[k−1]p\in[k-1] and p<q≤kp<q\le k, and all terms are nonnegative, so

S(w)≤∑p=1k−1 ∑q=p+1k(ρk−q+p−1+ρq−p−1).S(w)\le\sum_{p=1}^{k-1}\ \sum_{q=p+1}^{k}\bigl(\rho_{k-q+p-1}+\rho_{q-p-1}\bigr).

For fixed pp, as qq runs over {p+1,…,k}\{p+1,\dots,k\}, j=q−p−1j=q-p-1 runs bijectively over {0,…,k−p−1}\{0,\dots,k-p-1\} and j′=k−q+p−1j'=k-q+p-1 runs bijectively over {p−1,…,k−2}\{p-1,\dots,k-2\}, both subsets of {0,…,k−2}\{0,\dots,k-2\}. Hence each of the two inner sums is at most ∑j=0k−2ρj=ρ+∑j=1k−2ρj≤ρ+ρ1−ρ=124+123=47552\sum_{j=0}^{k-2}\rho_{j}=\rho+\sum_{j=1}^{k-2}\rho^{j}\le\rho+\frac{\rho}{1-\rho}=\frac{1}{24}+\frac{1}{23}=\frac{47}{552}, by (E5) with s=ρs=\rho (the last sum being empty if k=2k=2). Therefore S(w)≤(k−1)⋅2⋅47552≤47k276S(w)\le(k-1)\cdot2\cdot\frac{47}{552}\le\frac{47k}{276}, using (E6).

Step 6 (counting the words whose pieces strip to a given word). Let m,k∈Nm,k\in\mathbb{N} with m<km<k and let u∈Wd,m∘u\in W^{\circ}_{d,m}. Let I(u,k)I(u,k) be the set of triples ξ=(w,(p,q),ι)\xi=(w,(p,q),\iota) with w∈Wd,k∘w\in W^{\circ}_{d,k}, (p,q)∈P(w)(p,q)\in P(w), ι∈{in,out}\iota\in\{\mathrm{in},\mathrm{out}\} and upqι(w)=uu^{\iota}_{pq}(w)=u; it is a subset of the finite set Wd,k∘×([k]×[k])×{in,out}W^{\circ}_{d,k}\times([k]\times[k])\times\{\mathrm{in},\mathrm{out}\} (Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §finite, claim 1 of Finiteness of Cartesian Products, Tuple Sets, and Permutation Sets), hence finite by claim 3 of Basic Properties of Finite Sets. Put N(u,k)=∑ξ∈I(u,k)1N(u,k)=\sum_{\xi\in I(u,k)}1. We show

N(u,k)≤16 k (2d)k−m.(6.1)N(u,k)\le16\,k\,(2d)^{k-m}.\tag{6.1}

Let T={t∈{0}∪N:m+2t≤k−1}T=\{t\in\{0\}\cup\mathbb{N}:m+2t\le k-1\}, which contains 00, and Π={in,out}×[k]×{1,−1}×{1,−1}×T\Pi=\{\mathrm{in},\mathrm{out}\}\times[k]\times\{1,-1\}\times\{1,-1\}\times T. To ξ=(w,(p,q),ι)∈I(u,k)\xi=(w,(p,q),\iota)\in I(u,k) assign π(ξ)=(ι,p,εp,εq,t)\pi(\xi)=(\iota,p,\varepsilon_{p},\varepsilon_{q},t) with t=∣bpqι(w)∣t=|b^{\iota}_{pq}(w)|; since ∣zpqι(w)∣=m+2t≤k−1|z^{\iota}_{pq}(w)|=m+2t\le k-1 by Step 3 and (3a), π(ξ)∈Π\pi(\xi)\in\Pi. For π∈Π\pi\in\Pi let Iπ={ξ∈I(u,k):π(ξ)=π}I_{\pi}=\{\xi\in I(u,k):\pi(\xi)=\pi\}; these sets are pairwise disjoint with union I(u,k)I(u,k).

Fix π=(ι,p,e,e′,t)∈Π\pi=(\iota,p,e,e',t)\in\Pi with IπI_{\pi} nonempty, and put L=m+2tL=m+2t and nπ=k−m−tn_{\pi}=k-m-t. Let ξ=(w,(p,q),ι)∈Iπ\xi=(w,(p,q),\iota)\in I_{\pi} and z=zpqι(w)z=z^{\iota}_{pq}(w), so ∣z∣=L|z|=L. By (2e), σ:=σp\sigma:=\sigma_{p} depends only on (p,e)(p,e); by (2a) and (2b), τ:=τq\tau:=\tau_{q} equals L+σ−1L+\sigma-1 if ι=in\iota=\mathrm{in} and k−L+σ−1k-L+\sigma-1 if ι=out\iota=\mathrm{out}; and by (2e), qq is determined by τ\tau and e′e'. So σ\sigma, τ\tau and qq depend only on π\pi. Let r(ξ)=w[1,σ−1] w[τ+1,k]r(\xi)=w_{[1,\sigma-1]}\,w_{[\tau+1,k]} (empty factors omitted) if ι=in\iota=\mathrm{in}, and r(ξ)=w[σ,τ]=ypqr(\xi)=w_{[\sigma,\tau]}=y_{pq} if ι=out\iota=\mathrm{out}; in both cases r(ξ)r(\xi) is the other piece rearranged, of length k−L≥1k-L\ge1 by (2b). Put φπ(ξ)=bpqι(w) r(ξ)\varphi_{\pi}(\xi)=b^{\iota}_{pq}(w)\,r(\xi), a word of length t+(k−L)=nπ≥1t+(k-L)=n_{\pi}\ge1, that is an element of [2d]nπ[2d]^{n_{\pi}}. The map φπ\varphi_{\pi} is injective on IπI_{\pi}: from v=φπ(ξ)v=\varphi_{\pi}(\xi) one reads off b=bpqι(w)b=b^{\iota}_{pq}(w) as the first tt letters and r=r(ξ)r=r(\xi) as the last k−Lk-L letters, then z=b u b∗z=b\,u\,b^{*} (Step 3), and then ww letter by letter: if ι=in\iota=\mathrm{in}, wi=riw_{i}=r_{i} for i<σi<\sigma, wi=zi−σ+1w_{i}=z_{i-\sigma+1} for σ≤i≤τ\sigma\le i\le\tau and wi=ri−Lw_{i}=r_{i-L} for τ<i≤k\tau<i\le k; if ι=out\iota=\mathrm{out}, wi=zk−τ+iw_{i}=z_{k-\tau+i} for i<σi<\sigma, wi=ri−σ+1w_{i}=r_{i-\sigma+1} for σ≤i≤τ\sigma\le i\le\tau and wi=zi−τw_{i}=z_{i-\tau} for τ<i≤k\tau<i\le k (these are the definitions of w[a,b]w_{[a,b]} and of the pieces in The Free Unitary Heat Generator on Unitary Laws §pieces, read backwards); finally ξ=(w,(p,q),ι)\xi=(w,(p,q),\iota) with (p,q,ι)(p,q,\iota) determined by π\pi. So φπ\varphi_{\pi} is a bijection from IπI_{\pi} onto a nonempty set Bπ⊆[2d]nπB_{\pi}\subseteq[2d]^{n_{\pi}}, and by reindexing and Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §count with aa constantly 11 and M=1M=1,

∑ξ∈Iπ1=∑v∈Bπ1≤(2d)nπ≤(2d)k−m(12)t,\sum_{\xi\in I_{\pi}}1=\sum_{v\in B_{\pi}}1\le(2d)^{n_{\pi}}\le(2d)^{k-m}\Bigl(\frac12\Bigr)^{t},

the last inequality being trivial if t=0t=0 and following for t≥1t\ge1 from (2d)k−m=(2d)nπ(2d)t(2d)^{k-m}=(2d)^{n_{\pi}}(2d)^{t} (claim 1 of Properties of Natural Number Powers in a Field, as in (E2)), (2d)t≥2t(2d)^{t}\ge2^{t} (claim 5 of that lemma) and 2t(12)t=12^{t}(\frac12)^{t}=1 (claims 3 and 2). Splitting I(u,k)I(u,k) over the nonempty IπI_{\pi} and then enlarging to all π∈Π\pi\in\Pi (nonnegative terms), and writing the sum over Π\Pi as an iterated sum whose summand depends only on tt, with (E6),

N(u,k)≤2⋅k⋅2⋅2⋅(2d)k−m∑t∈T(12)t≤8k(2d)k−m⋅2,N(u,k)\le2\cdot k\cdot2\cdot2\cdot(2d)^{k-m}\sum_{t\in T}\Bigl(\frac12\Bigr)^{t}\le8k(2d)^{k-m}\cdot2,

since ∑t∈T(12)t≤1+∑t=1k(12)t≤1+1\sum_{t\in T}(\frac12)^{t}\le1+\sum_{t=1}^{k}(\frac12)^{t}\le1+1 by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric (the sum of the halving series is 11) and Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates. This is (6.1).

Step 7 (claim 3). Let μ,ν∈Ld\mu,\nu\in\mathcal{L}_{d}, and let Δv=μ(v)−ν(v)\Delta_{v}=\mu(v)-\nu(v) and R(w)R(w) be as in (3c). By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §sobolev, x:=ιd(μ)−ιd(ν)∈Edx:=\iota_{d}(\mu)-\iota_{d}(\nu)\in E_{d} is the map v↦Δvv\mapsto\Delta_{v} on Wd∘W^{\circ}_{d}, the map v↦cv∣v∣∣Δv∣2v\mapsto c_{v}|v||\Delta_{v}|^{2} is summable, and ∥x∥d,1\lVert x\rVert_{d,1} is defined. By claim 1 and The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §space, h:=Θ^μ−Θ^ν∈Edh:=\widehat{\Theta}\mu-\widehat{\Theta}\nu\in E_{d}, and it is the map v↦Θμ(v)−Θν(v)v\mapsto\Theta\mu(v)-\Theta\nu(v) on Wd∘W^{\circ}_{d}. For k∈Nk\in\mathbb{N} put

Xk=∑w∈Wd,k∘cw∣Δw∣2≥0,Pk=∑w∈Wd,k∘cwRe⁡(Δw‾ h(w)).X_{k}=\sum_{w\in W^{\circ}_{d,k}}c_{w}|\Delta_{w}|^{2}\ge0,\qquad P_{k}=\sum_{w\in W^{\circ}_{d,k}}c_{w}\operatorname{Re}\bigl(\overline{\Delta_{w}}\,h(w)\bigr).

(7a) The two gauges and the inner product. The kk-th block sum of v↦cv∣v∣∣Δv∣2v\mapsto c_{v}|v||\Delta_{v}|^{2} is kXkkX_{k}, since ∣w∣=k|w|=k on Wd,k∘W^{\circ}_{d,k} (homogeneity). As Δ∅=0\Delta_{\varnothing}=0, The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §norm, The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §sobolev and The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §sums give, both series converging,

∥x∥d2=∑k=1∞Xk,∥x∥d,12=∑k=1∞kXk.\lVert x\rVert_{d}^{2}=\sum_{k=1}^{\infty}X_{k},\qquad\lVert x\rVert_{d,1}^{2}=\sum_{k=1}^{\infty}kX_{k}.

By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §expansion, the series ∑k=1∞Pk\sum_{k=1}^{\infty}P_{k} converges and ⟨x,h⟩d=Re⁡(Δ∅‾ h(∅))+∑k=1∞Pk=∑k=1∞Pk\langle x,h\rangle_{d}=\operatorname{Re}(\overline{\Delta_{\varnothing}}\,h(\varnothing))+\sum_{k=1}^{\infty}P_{k}=\sum_{k=1}^{\infty}P_{k}, since 0‾ h(∅)=0\overline{0}\,h(\varnothing)=0 and Re⁡0=0\operatorname{Re}0=0.

(7b) Pointwise bound. Let k∈Nk\in\mathbb{N} and w∈Wd,k∘w\in W^{\circ}_{d,k}, and put β(w)=h(w)+k2Δw\beta(w)=h(w)+\frac{k}{2}\Delta_{w}, so that ∣β(w)∣≤R(w)|\beta(w)|\le R(w) by (3.1). Then Δw‾ h(w)=−k2∣Δw∣2+Δw‾ β(w)\overline{\Delta_{w}}\,h(w)=-\frac{k}{2}|\Delta_{w}|^{2}+\overline{\Delta_{w}}\,\beta(w) by claim 3 of Properties of Complex Conjugation and Modulus. The real part of a real number is that number, and real parts add, by the uniqueness of the canonical form in Real and Imaginary Parts of a Complex Number; with claims 6, 4 and 3 of Properties of Complex Conjugation and Modulus,

Re⁡(Δw‾ h(w))=−k2∣Δw∣2+Re⁡(Δw‾ β(w))≤−k2∣Δw∣2+∣Δw∣ R(w).\operatorname{Re}\bigl(\overline{\Delta_{w}}\,h(w)\bigr)=-\frac{k}{2}|\Delta_{w}|^{2}+\operatorname{Re}\bigl(\overline{\Delta_{w}}\,\beta(w)\bigr)\le-\frac{k}{2}|\Delta_{w}|^{2}+|\Delta_{w}|\,R(w).

(7c) Splitting the cross terms. For (p,q)∈P(w)(p,q)\in P(w) and ι∈{in,out}\iota\in\{\mathrm{in},\mathrm{out}\} let n=k−∣upqι(w)∣∈Nn=k-|u^{\iota}_{pq}(w)|\in\mathbb{N} (by (3a)) and apply (E4) with α=∣Δw∣\alpha=|\Delta_{w}|, β=∣Δupqι(w)∣\beta=|\Delta_{u^{\iota}_{pq}(w)}|, a=ρna=\rho^{n} and a′=24na'=24^{n}, which are positive with aa′=(ρ⋅24)n=1n=1aa'=(\rho\cdot24)^{n}=1^{n}=1 by claims 3 and 2 of Properties of Natural Number Powers in a Field. Summing over (p,q)∈P(w)(p,q)\in P(w) and ι\iota, and using the definition of S(w)S(w) in Step 5 and (5.1),

∣Δw∣ R(w)≤12S(w)∣Δw∣2+12∑(p,q)∈P(w) ∑ι∈{in,out}24 k−∣upqι(w)∣ ∣Δupqι(w)∣2≤47k552∣Δw∣2+12 G(w),|\Delta_{w}|\,R(w)\le\frac12S(w)|\Delta_{w}|^{2}+\frac12\sum_{(p,q)\in P(w)}\ \sum_{\iota\in\{\mathrm{in},\mathrm{out}\}}24^{\,k-|u^{\iota}_{pq}(w)|}\,\bigl|\Delta_{u^{\iota}_{pq}(w)}\bigr|^{2}\le\frac{47k}{552}|\Delta_{w}|^{2}+\frac12\,G(w),

where G(w)G(w) denotes the double sum. Multiplying (7b) and this bound by cw=γk>0c_{w}=\gamma_{k}>0 and summing over w∈Wd,k∘w\in W^{\circ}_{d,k} gives, for every k∈Nk\in\mathbb{N},

Pk≤(−12+47552)kXk+12Gk,Gk:=∑w∈Wd,k∘γkG(w).(7.1)P_{k}\le\Bigl(-\frac12+\frac{47}{552}\Bigr)kX_{k}+\frac12G_{k},\qquad G_{k}:=\sum_{w\in W^{\circ}_{d,k}}\gamma_{k}G(w).\tag{7.1}

(7d) Resumming the second part by the stripped word. Let J∈NJ\in\mathbb{N} (a new letter, unrelated to the KK of Step 4). We show

∑k=1JGk≤1663∑m=1Jm Xm.(7.2)\sum_{k=1}^{J}G_{k}\le\frac{16}{63}\sum_{m=1}^{J}m\,X_{m}.\tag{7.2}

Let TJT_{J} be the set of triples ξ=(w,(p,q),ι)\xi=(w,(p,q),\iota) with w∈Wd∘w\in W^{\circ}_{d}, 1≤∣w∣≤J1\le|w|\le J, (p,q)∈P(w)(p,q)\in P(w) and ι∈{in,out}\iota\in\{\mathrm{in},\mathrm{out}\}, and for such ξ\xi with ∣w∣=k|w|=k put f(ξ)=γk24 k−∣upqι(w)∣∣Δupqι(w)∣2≥0f(\xi)=\gamma_{k}24^{\,k-|u^{\iota}_{pq}(w)|}|\Delta_{u^{\iota}_{pq}(w)}|^{2}\ge0. Writing iterated sums as sums over dependent pairs, and noting that words with P(w)P(w) empty contribute 00, ∑k=1JGk=∑ξ∈TJf(ξ)\sum_{k=1}^{J}G_{k}=\sum_{\xi\in T_{J}}f(\xi) (both sides 00 if TJT_{J} is empty). Terms with upqι(w)=∅u^{\iota}_{pq}(w)=\varnothing vanish, as Δ∅=0\Delta_{\varnothing}=0. For every other ξ∈TJ\xi\in T_{J} the word u=upqι(w)u=u^{\iota}_{pq}(w) lies in Wd,m∘W^{\circ}_{d,m} for some m∈Nm\in\mathbb{N} with m≤∣w∣−1≤J−1m\le|w|-1\le J-1 by (3a); so these ξ\xi form the disjoint union of the sets I(u,k)I(u,k) of Step 6 over m∈[J−1]m\in[J-1], u∈Wd,m∘u\in W^{\circ}_{d,m} and k∈Nk\in\mathbb{N} with m<k≤Jm<k\le J, and on I(u,k)I(u,k) the function ff has the constant value γk24k−m∣Δu∣2\gamma_{k}24^{k-m}|\Delta_{u}|^{2}. Hence (the right side being 00 if J=1J=1)

∑ξ∈TJf(ξ)=∑m=1J−1 ∑u∈Wd,m∘∣Δu∣2∑k=m+1JN(u,k) γk24k−m.\sum_{\xi\in T_{J}}f(\xi)=\sum_{m=1}^{J-1}\ \sum_{u\in W^{\circ}_{d,m}}|\Delta_{u}|^{2}\sum_{k=m+1}^{J}N(u,k)\,\gamma_{k}24^{k-m}.

By (6.1), by (E7)(ii) in the form γk≤θk−mγm=θk−mcu\gamma_{k}\le\theta^{k-m}\gamma_{m}=\theta^{k-m}c_{u} (here m<km<k and u∈Wd,m∘u\in W^{\circ}_{d,m}), and by claim 3 of Properties of Natural Number Powers in a Field,

N(u,k) γk24k−m≤16k cu (2d⋅24⋅θ)k−m=16k cu(1128)k−m,N(u,k)\,\gamma_{k}24^{k-m}\le16k\,c_{u}\,(2d\cdot24\cdot\theta)^{k-m}=16k\,c_{u}\Bigl(\frac{1}{128}\Bigr)^{k-m},

because 2d⋅24⋅θ=48d6144d=11282d\cdot24\cdot\theta=\frac{48d}{6144d}=\frac{1}{128}. Reindex by r=k−mr=k-m, which runs over [J−m][J-m] with J−m∈NJ-m\in\mathbb{N}. By (E3), k=m+r≤m 2rk=m+r\le m\,2^{r}, and 2r(1128)r=(164)r2^{r}(\frac{1}{128})^{r}=(\frac{1}{64})^{r} by claim 3 of Properties of Natural Number Powers in a Field; applying (E5) with s=164s=\frac{1}{64}, for which s1−s=163\frac{s}{1-s}=\frac{1}{63},

∑k=m+1Jk(1128)k−m=∑r=1J−m(m+r)(1128)r≤m∑r=1J−m(164)r≤m63.\sum_{k=m+1}^{J}k\Bigl(\frac{1}{128}\Bigr)^{k-m}=\sum_{r=1}^{J-m}(m+r)\Bigl(\frac{1}{128}\Bigr)^{r}\le m\sum_{r=1}^{J-m}\Bigl(\frac{1}{64}\Bigr)^{r}\le\frac{m}{63}.

Therefore ∑ξ∈TJf(ξ)≤16∑m=1J−1m63∑u∈Wd,m∘cu∣Δu∣2=1663∑m=1J−1m Xm\sum_{\xi\in T_{J}}f(\xi)\le16\sum_{m=1}^{J-1}\frac{m}{63}\sum_{u\in W^{\circ}_{d,m}}c_{u}|\Delta_{u}|^{2}=\frac{16}{63}\sum_{m=1}^{J-1}m\,X_{m}, and adding the nonnegative term with m=Jm=J gives (7.2). (No term independent of mm arises here: this is where m≥1m\ge1, that is Δ∅=0\Delta_{\varnothing}=0, is used to absorb the factor k=m+rk=m+r into m 2rm\,2^{r}.)

(7e) Truncated inequality. Let J∈NJ\in\mathbb{N}. Summing (7.1) over k∈[J]k\in[J] and using (7.2),

∑k=1JPk≤∑k=1J(−12+47552+863)kXk≤∑k=1J(−14)kXk,\sum_{k=1}^{J}P_{k}\le\sum_{k=1}^{J}\Bigl(-\frac12+\frac{47}{552}+\frac{8}{63}\Bigr)kX_{k}\le\sum_{k=1}^{J}\Bigl(-\frac{1}{4}\Bigr)kX_{k},

the last step because kXk≥0kX_{k}\ge0 and, with the common denominator 552⋅63=34776552\cdot63=34776,

−12+47552+863=−17388+2961+441634776=−1001134776≤−869434776=−14.-\frac12+\frac{47}{552}+\frac{8}{63}=\frac{-17388+2961+4416}{34776}=-\frac{10011}{34776}\le-\frac{8694}{34776}=-\frac14.

(7f) Passage to the limit. By (7a), ∑k=1∞kXk\sum_{k=1}^{\infty}kX_{k} converges with sum ∥x∥d,12\lVert x\rVert_{d,1}^{2}, so by Elementary Properties of Series of Real Numbers §linearity the series with terms −14kXk-\frac14kX_{k} converges with sum −14∥x∥d,12-\frac14\lVert x\rVert_{d,1}^{2}. By Series of Real Numbers §convergent, the two sides of the inequality in (7e) are the JJ-th partial sums of convergent series, which converge to ⟨x,h⟩d\langle x,h\rangle_{d} (by (7a)) and to −14∥x∥d,12-\frac14\lVert x\rVert_{d,1}^{2} respectively. Claim 1 of Order Properties of Limits of Real Sequences gives

⟨ιd(μ)−ιd(ν), Θ^μ−Θ^ν⟩d≤−14∥ιd(μ)−ιd(ν)∥d,12.\bigl\langle\iota_{d}(\mu)-\iota_{d}(\nu),\,\widehat{\Theta}\mu-\widehat{\Theta}\nu\bigr\rangle_{d}\le-\frac14\lVert\iota_{d}(\mu)-\iota_{d}(\nu)\rVert_{d,1}^{2}.

This proves claim 3.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…