Each result cited is universally quantified over the data in its own statement.
We prove the estimate with the explicit trilinear constant K = 128 K=128 K = 128 .
Conventions. Fix μ , ν ∈ L d \mu,\nu\in\mathcal{L}_{d} μ , ν ∈ L d and n ∈ N n\in\mathbb{N} n ∈ N , and write θ = θ d = 1 6144 d \theta=\theta_{d}=\frac{1}{6144\,d} θ = θ d = 6144 d 1 as in The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §weights . Since nothing beyond μ , ν ∈ L d \mu,\nu\in\mathcal{L}_{d} μ , ν ∈ L d and n ∈ N n\in\mathbb{N} n ∈ N is used, proving the displayed inequality with K = 128 K=128 K = 128 for these data proves the lemma. For z ∈ W 2 d z\in W_{2d} z ∈ W 2 d put Δ ( z ) = μ ( z ) − ν ( z ) ∈ C \Delta(z)=\mu(z)-\nu(z)\in\mathbb{C} Δ ( z ) = μ ( z ) − ν ( z ) ∈ C ; then Δ ( ∅ ) = 1 − 1 = 0 \Delta(\varnothing)=1-1=0 Δ ( ∅ ) = 1 − 1 = 0 by Laws of d-Tuples of Unitaries §normalised . By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §sobolev , x = ι d ( μ ) − ι d ( ν ) x=\iota_{d}(\mu)-\iota_{d}(\nu) x = ι d ( μ ) − ι d ( ν ) is the map w ↦ Δ ( w ) w\mapsto\Delta(w) w ↦ Δ ( w ) on W d ∘ W^{\circ}_{d} W d ∘ , the map w ↦ c w ∣ w ∣ ∣ x ( w ) ∣ 2 w\mapsto c_{w}|w||x(w)|^{2} w ↦ c w ∣ w ∣∣ x ( w ) ∣ 2 is summable, and ∥ x ∥ d , 1 \lVert x\rVert_{d,1} ∥ x ∥ d , 1 is defined. For k ∈ N k\in\mathbb{N} k ∈ N put X k = ∑ w ∈ W d , k ∘ c w ∣ x ( w ) ∣ 2 X_{k}=\sum_{w\in W^{\circ}_{d,k}}c_{w}|x(w)|^{2} X k = ∑ w ∈ W d , k ∘ c w ∣ x ( w ) ∣ 2 , which is nonnegative by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §nonnegative since c w > 0 c_{w}>0 c w > 0 (The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §weights ). The empty word is cyclically reduced (Reduced and Cyclically Reduced Words in Unitary Letters §reduced , Reduced and Cyclically Reduced Words in Unitary Letters §cyclically-reduced ) and x ( ∅ ) = Δ ( ∅ ) = 0 x(\varnothing)=\Delta(\varnothing)=0 x ( ∅ ) = Δ ( ∅ ) = 0 ; the k k k -th block sum of w ↦ c w ∣ w ∣ ∣ x ( w ) ∣ 2 w\mapsto c_{w}|w||x(w)|^{2} w ↦ c w ∣ w ∣∣ x ( w ) ∣ 2 is k X k kX_{k} k X k , because ∣ w ∣ = k |w|=k ∣ w ∣ = k on W d , k ∘ W^{\circ}_{d,k} W d , k ∘ . Hence The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §sums , 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 §sobolev give, the square of a nonnegative square root being the number itself, ∥ x ∥ d 2 = ∑ k = 1 ∞ X k \lVert x\rVert_{d}^{2}=\sum_{k=1}^{\infty}X_{k} ∥ x ∥ d 2 = ∑ k = 1 ∞ X k and ∥ x ∥ d , 1 2 = ∑ k = 1 ∞ k X k \lVert x\rVert_{d,1}^{2}=\sum_{k=1}^{\infty}kX_{k} ∥ x ∥ d , 1 2 = ∑ k = 1 ∞ k X k , both series converging and having nonnegative terms; so by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates , for every M ∈ N M\in\mathbb{N} M ∈ N ,
∑ k = 1 M X k ≤ ∥ x ∥ d 2 , ∑ k = 1 M k X k ≤ ∥ x ∥ d , 1 2 . (0.1) \sum_{k=1}^{M}X_{k}\le\lVert x\rVert_{d}^{2},\qquad\sum_{k=1}^{M}k\,X_{k}\le\lVert x\rVert_{d,1}^{2}.\tag{0.1} k = 1 ∑ M X k ≤ ∥ x ∥ d 2 , k = 1 ∑ M k X k ≤ ∥ x ∥ d , 1 2 . ( 0.1 )
Real arithmetic. Rearrangements of real equalities and inequalities (adding, multiplying by a nonnegative number, dividing by a positive number, taking reciprocals of positive numbers) are those of Elementary Arithmetic in an Ordered Field , Elementary Order Arithmetic in an Ordered Field and Order Reversal under Reciprocals, and Summability of the Reciprocals of the Squares §reciprocal . For a nonnegative real r r r , r \sqrt{r} r is its nonnegative square root (Existence and Uniqueness of the Nonnegative Square Root ). For nonnegative reals r , r ′ r,r' r , r ′ one has r ≤ r ′ r\le r' r ≤ r ′ if and only if r 2 ≤ r ′ 2 r^{2}\le r'^{2} r 2 ≤ r ′ 2 , and r = r ′ r=r' r = r ′ if and only if r 2 = r ′ 2 r^{2}=r'^{2} r 2 = r ′ 2 , by the weak and equality forms of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field ; in particular r r ′ = r r ′ \sqrt{rr'}=\sqrt{r}\sqrt{r'} r r ′ = r r ′ and r 2 = r \sqrt{r^{2}}=r r 2 = r . Put ϑ = θ \vartheta=\sqrt{\theta} ϑ = θ . For a real r r r and N ∈ N N\in\mathbb{N} N ∈ N , r N r^{N} r N is the natural power, and r 0 = 1 r^{0}=1 r 0 = 1 ; then r N + N ′ = r N r N ′ r^{N+N'}=r^{N}r^{N'} r N + N ′ = r N r N ′ for N , N ′ ∈ { 0 } ∪ N N,N'\in\{0\}\cup\mathbb{N} N , N ′ ∈ { 0 } ∪ N , by Addition of Exponents for Natural Number Powers in a Field when both are in N \mathbb{N} N and trivially otherwise; ( r r ′ ) N = r N r ′ N (rr')^{N}=r^{N}r'^{N} ( r r ′ ) N = r N r ′ N and 1 N = 1 1^{N}=1 1 N = 1 , and 0 ≤ r ≤ r ′ 0\le r\le r' 0 ≤ r ≤ r ′ implies 0 ≤ r N ≤ r ′ N 0\le r^{N}\le r'^{N} 0 ≤ r N ≤ r ′ N , by claims 3, 2 and 5 of Properties of Natural Number Powers in a Field (trivially for N = 0 N=0 N = 0 ). In particular ( ϑ N ) 2 = ( ϑ ϑ ) N = θ N (\vartheta^{N})^{2}=(\vartheta\vartheta)^{N}=\theta^{N} ( ϑ N ) 2 = ( ϑϑ ) N = θ N .
Finite sums. A sum over an empty index set is 0 0 0 . For sums of real numbers over finite sets: termwise inequalities pass to sums by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §comparison ; sums of nonnegative terms are nonnegative and do 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 agree with numerical ranges, may be reindexed along bijections, and are additive and homogeneous, by claims 1, 2, 3 and 4 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 ; terms equal to 0 0 0 may be dropped by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing ; and a product of two sums is the sum over the Cartesian product, by The Product of Two Sums over Finite Index Sets is a Sum over the Cartesian Product . Finite unions of finite sets are finite by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §finite-union , subsets and images of finite sets are finite by claims 3 and 4 of Basic Properties of Finite Sets , and finite products of finite sets are finite by claim 1 of Finiteness of Cartesian Products, Tuple Sets, and Permutation Sets . For complex numbers we use the rules for conjugation, the modulus of a conjugate, multiplicativity of the modulus, the triangle inequality and the modulus of a nonnegative real, claims 1, 3, 4, 7 and 8 of Properties of Complex Conjugation and Modulus , and the modulus of a finite sum, Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §modulus . In particular ∣ i ∣ = 1 |\mathrm{i}|=1 ∣ i ∣ = 1 (as ∣ i ∣ 2 = i i ‾ = 1 |\mathrm{i}|^{2}=\mathrm{i}\,\overline{\mathrm{i}}=1 ∣ i ∣ 2 = i i = 1 ) and ∣ ε ( l ) ∣ = 1 |\varepsilon(l)|=1 ∣ ε ( l ) ∣ = 1 for every letter l l l .
Words. For L ∈ N L\in\mathbb{N} L ∈ N let [ 2 d ] L [2d]^{L} [ 2 d ] L be the set of words of length L L L , nonempty and finite by Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §finite , and put [ 2 d ] 0 = { ∅ } [2d]^{0}=\{\varnothing\} [ 2 d ] 0 = { ∅ } . For M ∈ { 0 } ∪ N M\in\{0\}\cup\mathbb{N} M ∈ { 0 } ∪ N let U M U_{M} U M be the union of the sets [ 2 d ] L [2d]^{L} [ 2 d ] L , L ∈ { 0 , 1 , … , M } L\in\{0,1,\dots,M\} L ∈ { 0 , 1 , … , M } , a finite set containing every word of length at most M M M . By Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §concatenation and Basic Properties of Words: Associativity, Reversal, Finitely Many Factorisations, and Countability §monoid , if u u u has length L L L and v v v length L ′ L' L ′ then u v uv uv has length L + L ′ L+L' L + L ′ , its first L L L letters are those of u u u and its last L ′ L' L ′ letters those of v v v , and concatenation is associative. Consequently, for fixed L ∈ { 0 } ∪ N L\in\{0\}\cup\mathbb{N} L ∈ { 0 } ∪ N , the maps ( u , v ) ↦ u v (u,v)\mapsto uv ( u , v ) ↦ uv on [ 2 d ] L × W 2 d [2d]^{L}\times W_{2d} [ 2 d ] L × W 2 d and on W 2 d × [ 2 d ] L W_{2d}\times[2d]^{L} W 2 d × [ 2 d ] L are injective, and for L ≤ L ′ L\le L' L ≤ L ′ the map ( u , v ) ↦ u v (u,v)\mapsto uv ( u , v ) ↦ uv is a bijection from [ 2 d ] L × [ 2 d ] L ′ − L [2d]^{L}\times[2d]^{L'-L} [ 2 d ] L × [ 2 d ] L ′ − L onto [ 2 d ] L ′ [2d]^{L'} [ 2 d ] L ′ (split a word after its L L L -th letter).
Step 0 (elementary inequalities).
(C1) Finite Cauchy--Schwarz. Let F F F be a nonempty finite set and a , b : F → R a,b:F\to\mathbb{R} a , b : F → R with a , b ≥ 0 a,b\ge0 a , b ≥ 0 . Then
∑ y ∈ F a ( y ) b ( y ) ≤ ∑ y ∈ F a ( y ) 2 ∑ y ∈ F b ( y ) 2 , \sum_{y\in F}a(y)b(y)\le\sqrt{\textstyle\sum_{y\in F}a(y)^{2}}\;\sqrt{\textstyle\sum_{y\in F}b(y)^{2}}, y ∈ F ∑ a ( y ) b ( y ) ≤ ∑ y ∈ F a ( y ) 2 ∑ y ∈ F b ( y ) 2 ,
and, with b b b constantly 1 1 1 and N F = ∑ y ∈ F 1 N_{F}=\sum_{y\in F}1 N F = ∑ y ∈ F 1 , ( ∑ y ∈ F a ( y ) ) 2 ≤ N F ∑ y ∈ F a ( y ) 2 \bigl(\sum_{y\in F}a(y)\bigr)^{2}\le N_{F}\sum_{y\in F}a(y)^{2} ( ∑ y ∈ F a ( y ) ) 2 ≤ N F ∑ y ∈ F a ( y ) 2 . (This is The Cauchy-Schwarz Inequality in a Real Inner Product Space for the standard pairing on R F \mathbb{R}^{F} R F ; we give the direct proof.) Put A = ∑ a 2 A=\sqrt{\sum a^{2}} A = ∑ a 2 and B = ∑ b 2 B=\sqrt{\sum b^{2}} B = ∑ b 2 . If A = 0 A=0 A = 0 , then ∑ a 2 = 0 \sum a^{2}=0 ∑ a 2 = 0 , and since each nonnegative term a ( y ) 2 a(y)^{2} a ( y ) 2 is at most the sum (monotonicity in the index set), a ( y ) 2 = 0 a(y)^{2}=0 a ( y ) 2 = 0 , so a ( y ) = 0 a(y)=0 a ( y ) = 0 for all y y y and the left side is 0 0 0 ; likewise if B = 0 B=0 B = 0 . Otherwise, for each y y y , 0 ≤ ( a ( y ) A − b ( y ) B ) 2 0\le\bigl(\frac{a(y)}{A}-\frac{b(y)}{B}\bigr)^{2} 0 ≤ ( A a ( y ) − B b ( y ) ) 2 by Nonnegativity of Squares in an Ordered Field , that is 2 a ( y ) b ( y ) A B ≤ a ( y ) 2 A 2 + b ( y ) 2 B 2 2\frac{a(y)b(y)}{AB}\le\frac{a(y)^{2}}{A^{2}}+\frac{b(y)^{2}}{B^{2}} 2 A B a ( y ) b ( y ) ≤ A 2 a ( y ) 2 + B 2 b ( y ) 2 ; summing gives 2 A B ∑ a b ≤ 1 + 1 \frac{2}{AB}\sum ab\le1+1 A B 2 ∑ ab ≤ 1 + 1 . The second form follows by squaring the nonnegative sides.
(C2) Numerical facts. 0 < θ ≤ 1 6144 0<\theta\le\frac{1}{6144} 0 < θ ≤ 6144 1 , because 6144 ≤ 6144 d 6144\le6144\,d 6144 ≤ 6144 d and reciprocals reverse the order; hence 0 ≤ 2 θ ≤ 1 0\le2\theta\le1 0 ≤ 2 θ ≤ 1 and θ ≤ 1 4 \theta\le\frac14 θ ≤ 4 1 , so ϑ 2 ≤ ( 1 2 ) 2 \vartheta^{2}\le(\frac12)^{2} ϑ 2 ≤ ( 2 1 ) 2 and 0 ≤ ϑ ≤ 1 2 0\le\vartheta\le\frac12 0 ≤ ϑ ≤ 2 1 . Moreover 2 d θ = 1 3072 2d\,\theta=\frac{1}{3072} 2 d θ = 3072 1 , as in The Embedding of Unitary Laws into the Word Gauge Space and the Gauge Distance §embedding .
(C3) For every N ∈ N N\in\mathbb{N} N ∈ N , N ≤ 2 N N\le2^{N} N ≤ 2 N and hence N θ N ≤ 1 N\theta^{N}\le1 N θ N ≤ 1 . Indeed 1 ≤ 2 = 2 1 1\le2=2^{1} 1 ≤ 2 = 2 1 , and if N ≤ 2 N N\le2^{N} N ≤ 2 N then, as 1 = 1 N ≤ 2 N 1=1^{N}\le2^{N} 1 = 1 N ≤ 2 N , N + 1 ≤ 2 N + 2 N = 2 N + 1 N+1\le2^{N}+2^{N}=2^{N+1} N + 1 ≤ 2 N + 2 N = 2 N + 1 by claim 1 of Properties of Natural Number Powers in a Field ; so N θ N ≤ 2 N θ N = ( 2 θ ) N ≤ 1 N = 1 N\theta^{N}\le2^{N}\theta^{N}=(2\theta)^{N}\le1^{N}=1 N θ N ≤ 2 N θ N = ( 2 θ ) N ≤ 1 N = 1 by (C2).
(C4) For k ∈ { 0 } ∪ N k\in\{0\}\cup\mathbb{N} k ∈ { 0 } ∪ N , ∑ j = 0 k ϑ j ≤ 2 \sum_{j=0}^{k}\vartheta^{j}\le2 ∑ j = 0 k ϑ j ≤ 2 ; and ∑ t = 0 n ( 2 d θ ) t ≤ 2 \sum_{t=0}^{n}(2d\,\theta)^{t}\le2 ∑ t = 0 n ( 2 d θ ) t ≤ 2 . For the first, the case k = 0 k=0 k = 0 reads 1 ≤ 2 1\le2 1 ≤ 2 ; for k ≥ 1 k\ge1 k ≥ 1 , ϑ j ≤ ( 1 2 ) j \vartheta^{j}\le(\frac12)^{j} ϑ j ≤ ( 2 1 ) j by (C2), and ∑ j = 1 k ( 1 2 ) j ≤ 1 \sum_{j=1}^{k}(\frac12)^{j}\le1 ∑ j = 1 k ( 2 1 ) j ≤ 1 by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric (the halving series has sum 1 1 1 ) and Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates . For the second, with r = 1 3072 r=\frac{1}{3072} r = 3072 1 , Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric gives ∑ t = 1 n r t = r − r n + 1 1 − r ≤ r 1 − r = 1 3071 ≤ 1 \sum_{t=1}^{n}r^{t}=\frac{r-r^{n+1}}{1-r}\le\frac{r}{1-r}=\frac{1}{3071}\le1 ∑ t = 1 n r t = 1 − r r − r n + 1 ≤ 1 − r r = 3071 1 ≤ 1 .
(C5) Let N ∈ { 0 } ∪ N N\in\{0\}\cup\mathbb{N} N ∈ { 0 } ∪ N and k , k ′ , p ∈ N k,k',p\in\mathbb{N} k , k ′ , p ∈ N with k ≤ k ′ k\le k' k ≤ k ′ and k + k ′ = p + 2 N k+k'=p+2N k + k ′ = p + 2 N . Then θ N k ′ ≤ 4 p \theta^{N}k'\le4p θ N k ′ ≤ 4 p . Indeed, if k ′ ≤ 4 p k'\le4p k ′ ≤ 4 p , then θ N ≤ 1 N = 1 \theta^{N}\le1^{N}=1 θ N ≤ 1 N = 1 by (C2), so θ N k ′ ≤ k ′ ≤ 4 p \theta^{N}k'\le k'\le4p θ N k ′ ≤ k ′ ≤ 4 p . Otherwise 4 p + 1 ≤ k ′ 4p+1\le k' 4 p + 1 ≤ k ′ ; since 1 ≤ k 1\le k 1 ≤ k , 2 N = k + k ′ − p ≥ k ′ − p 2N=k+k'-p\ge k'-p 2 N = k + k ′ − p ≥ k ′ − p , so 8 N ≥ 4 k ′ − 4 p ≥ 4 k ′ − ( k ′ − 1 ) = 3 k ′ + 1 8N\ge4k'-4p\ge4k'-(k'-1)=3k'+1 8 N ≥ 4 k ′ − 4 p ≥ 4 k ′ − ( k ′ − 1 ) = 3 k ′ + 1 . Hence N ≥ 1 N\ge1 N ≥ 1 and 3 k ′ < 8 N ≤ 9 N 3k'<8N\le9N 3 k ′ < 8 N ≤ 9 N , so k ′ ≤ 3 N k'\le3N k ′ ≤ 3 N , and by (C3) θ N k ′ ≤ 3 N θ N ≤ 3 ≤ 4 p \theta^{N}k'\le3N\theta^{N}\le3\le4p θ N k ′ ≤ 3 N θ N ≤ 3 ≤ 4 p .
Step 1 (facts about words).
(W1) For every letter l ∈ [ 2 d ] l\in[2d] l ∈ [ 2 d ] , ( l − 1 ) − 1 = l (l^{-1})^{-1}=l ( l − 1 ) − 1 = l . This is read off Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters : if l ≤ d l\le d l ≤ d then l − 1 = d + l l^{-1}=d+l l − 1 = d + l , whose inverse letter is the unique j j j with d + l = d + j d+l=d+j d + l = d + j , namely l l l ; if d < l d<l d < l then l = d + j l=d+j l = d + j with j ∈ [ d ] j\in[d] j ∈ [ d ] , l − 1 = j l^{-1}=j l − 1 = j and j − 1 = d + j = l j^{-1}=d+j=l j − 1 = d + j = l .
(W2) Let z ∈ W 2 d z\in W_{2d} z ∈ W 2 d have length L ∈ N L\in\mathbb{N} L ∈ N . Then z ∗ z^{*} z ∗ has length L L L and ( z ∗ ) i = ( z i ′ ) − 1 (z^{*})_{i}=(z_{i'})^{-1} ( z ∗ ) i = ( z i ′ ) − 1 where i + i ′ = L + 1 i+i'=L+1 i + i ′ = L + 1 , by Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §adjoint and Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §reversal ; for a letter l l l , l ∗ = l − 1 l^{*}=l^{-1} l ∗ = l − 1 (the case L = 1 L=1 L = 1 ); ( z ∗ ) ∗ = z (z^{*})^{*}=z ( z ∗ ) ∗ = z for every z ∈ W 2 d z\in W_{2d} z ∈ W 2 d , as noted in Word Polynomials in Unitary Letters: the Normed *-Algebra of Finitely Supported Word Functions §adjoint , so z ↦ z ∗ z\mapsto z^{*} z ↦ z ∗ is injective; and ( u v ) ∗ = v ∗ u ∗ (uv)^{*}=v^{*}u^{*} ( uv ) ∗ = v ∗ u ∗ for all u , v ∈ W 2 d u,v\in W_{2d} u , v ∈ W 2 d . For the last, Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra gives e ( u v ) ∗ = ( e u v ) ∗ = ( e u e v ) ∗ = ( e v ) ∗ ( e u ) ∗ = e v ∗ e u ∗ = e v ∗ u ∗ e_{(uv)^{*}}=(e_{uv})^{*}=(e_{u}e_{v})^{*}=(e_{v})^{*}(e_{u})^{*}=e_{v^{*}}e_{u^{*}}=e_{v^{*}u^{*}} e ( uv ) ∗ = ( e uv ) ∗ = ( e u e v ) ∗ = ( e v ) ∗ ( e u ) ∗ = e v ∗ e u ∗ = e v ∗ u ∗ , and evaluating at ( u v ) ∗ (uv)^{*} ( uv ) ∗ , where the left side takes the value 1 1 1 , gives ( u v ) ∗ = v ∗ u ∗ (uv)^{*}=v^{*}u^{*} ( uv ) ∗ = v ∗ u ∗ .
(W3) If z z z is reduced , then so is every word formed by consecutive letters of z z z (its consecutive letters are consecutive letters of z z z ), and so is z ∗ z^{*} z ∗ . For the latter let z z z have length L L L , i ∈ [ L ] i\in[L] i ∈ [ L ] with i < L i<L i < L , and i ′ i' i ′ with i + i ′ = L + 1 i+i'=L+1 i + i ′ = L + 1 , so that ( z ∗ ) i = ( z i ′ ) − 1 (z^{*})_{i}=(z_{i'})^{-1} ( z ∗ ) i = ( z i ′ ) − 1 and ( z ∗ ) i + 1 = ( z i ′ − 1 ) − 1 (z^{*})_{i+1}=(z_{i'-1})^{-1} ( z ∗ ) i + 1 = ( z i ′ − 1 ) − 1 by (W2). Since ( ( z ∗ ) i ) − 1 = z i ′ ((z^{*})_{i})^{-1}=z_{i'} (( z ∗ ) i ) − 1 = z i ′ by (W1), an equality ( z ∗ ) i + 1 = ( ( z ∗ ) i ) − 1 (z^{*})_{i+1}=((z^{*})_{i})^{-1} ( z ∗ ) i + 1 = (( z ∗ ) i ) − 1 would read z i ′ = ( z i ′ − 1 ) − 1 z_{i'}=(z_{i'-1})^{-1} z i ′ = ( z i ′ − 1 ) − 1 , contradicting reducedness of z z z at position i ′ − 1 < L i'-1<L i ′ − 1 < L (note 1 < i ′ 1<i' 1 < i ′ as i < L i<L i < L ).
(W4) Let k ∈ N k\in\mathbb{N} k ∈ N , w ∈ W d , k ∘ w\in W^{\circ}_{d,k} w ∈ W d , k ∘ and m ∈ [ k ] m\in[k] m ∈ [ k ] . Put q = m q=m q = m if ε ( w m ) = 1 \varepsilon(w_{m})=1 ε ( w m ) = 1 ; q = m + 1 q=m+1 q = m + 1 if ε ( w m ) = − 1 \varepsilon(w_{m})=-1 ε ( w m ) = − 1 and m < k m<k m < k ; and q = 1 q=1 q = 1 if ε ( w m ) = − 1 \varepsilon(w_{m})=-1 ε ( w m ) = − 1 and m = k m=k m = k . By Rotations and Cyclic Derivatives of Words in Unitary Letters §rotations , r m ( w ) = w r_{m}(w)=w r m ( w ) = w if q = 1 q=1 q = 1 , and otherwise r m ( w ) r_{m}(w) r m ( w ) is the concatenation of the word w q w q + 1 ⋯ w k w_{q}w_{q+1}\cdots w_{k} w q w q + 1 ⋯ w k and the word w 1 ⋯ w q − 1 w_{1}\cdots w_{q-1} w 1 ⋯ w q − 1 . Hence r m ( w ) r_{m}(w) r m ( w ) has length k k k , its i i i -th letter is w q − 1 + i w_{q-1+i} w q − 1 + i for i ≤ k − q + 1 i\le k-q+1 i ≤ k − q + 1 and its ( k − q + 1 + i ) (k-q+1+i) ( k − q + 1 + i ) -th letter is w i w_{i} w i for i ≤ q − 1 i\le q-1 i ≤ q − 1 ; so w w w is determined by r m ( w ) r_{m}(w) r m ( w ) together with m m m and ε ( w m ) \varepsilon(w_{m}) ε ( w m ) . Moreover r m ( w ) r_{m}(w) r m ( w ) is reduced: two consecutive letters of r m ( w ) r_{m}(w) r m ( w ) are either consecutive letters w i , w i + 1 w_{i},w_{i+1} w i , w i + 1 of w w w , which satisfy w i + 1 ≠ ( w i ) − 1 w_{i+1}\ne(w_{i})^{-1} w i + 1 = ( w i ) − 1 , or (when 1 < q 1<q 1 < q , so 1 < k 1<k 1 < k ) the letters w k , w 1 w_{k},w_{1} w k , w 1 , and w 1 = ( w k ) − 1 w_{1}=(w_{k})^{-1} w 1 = ( w k ) − 1 would give ( w 1 ) − 1 = w k (w_{1})^{-1}=w_{k} ( w 1 ) − 1 = w k by (W1), contradicting Reduced and Cyclically Reduced Words in Unitary Letters §cyclically-reduced .
(W5) For all a , c , b ∈ W 2 d a,c,b\in W_{2d} a , c , b ∈ W 2 d and λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d , λ ( a c c ∗ b ) = λ ( a b ) \lambda(a\,c\,c^{*}\,b)=\lambda(ab) λ ( a c c ∗ b ) = λ ( ab ) . By induction on the length of c c c : for c = ∅ c=\varnothing c = ∅ there is nothing to prove, as ∅ ∗ = ∅ \varnothing^{*}=\varnothing ∅ ∗ = ∅ . If c c c has length j + 1 j+1 j + 1 , then by Basic Properties of Words: Associativity, Reversal, Finitely Many Factorisations, and Countability §last-letter c = c ′ l c=c'l c = c ′ l with a letter l l l and c ′ c' c ′ of length j j j (or c ′ = ∅ c'=\varnothing c ′ = ∅ if j = 0 j=0 j = 0 ), so c ∗ = l − 1 c ′ ∗ c^{*}=l^{-1}c'^{*} c ∗ = l − 1 c ′ ∗ by (W2), and a c c ∗ b = ( a c ′ ) l l − 1 ( c ′ ∗ b ) a\,c\,c^{*}\,b=(ac')\,l\,l^{-1}\,(c'^{*}b) a c c ∗ b = ( a c ′ ) l l − 1 ( c ′ ∗ b ) ; by Laws of d-Tuples of Unitaries §cancellation and the induction hypothesis, λ ( a c c ∗ b ) = λ ( a c ′ c ′ ∗ b ) = λ ( a b ) \lambda(acc^{*}b)=\lambda(ac'c'^{*}b)=\lambda(ab) λ ( a c c ∗ b ) = λ ( a c ′ c ′ ∗ b ) = λ ( ab ) .
Step 2 (maximal cancellation and stripping). Let v , v ′ v,v' v , v ′ be reduced words of lengths k , k ′ ∈ N k,k'\in\mathbb{N} k , k ′ ∈ N . Let J J J be the set of j ∈ { 0 , 1 , … , min ( k , k ′ ) } j\in\{0,1,\dots,\min(k,k')\} j ∈ { 0 , 1 , … , min ( k , k ′ )} such that v i ′ = ( v i ′ ) − 1 v'_{i}=(v_{i'})^{-1} v i ′ = ( v i ′ ) − 1 whenever i ∈ [ j ] i\in[j] i ∈ [ j ] and i + i ′ = k + 1 i+i'=k+1 i + i ′ = k + 1 ; it contains 0 0 0 and is finite, and we let j ( v , v ′ ) j(v,v') j ( v , v ′ ) be its largest element. With j = j ( v , v ′ ) j=j(v,v') j = j ( v , v ′ ) , let a a a be the word of length k − j k-j k − j with a i = v i a_{i}=v_{i} a i = v i , c c c the word of length j j j with c i = v k − j + i c_{i}=v_{k-j+i} c i = v k − j + i , and b b b the word of length k ′ − j k'-j k ′ − j with b i = v j + i ′ b_{i}=v'_{j+i} b i = v j + i ′ (each read as ∅ \varnothing ∅ when its length is 0 0 0 ). Then:
(2a) v = a c v=ac v = a c and v ′ = c ∗ b v'=c^{*}b v ′ = c ∗ b . The first is the letter description of concatenation. For the second, by (W2) the i i i -th letter of c ∗ c^{*} c ∗ , i ∈ [ j ] i\in[j] i ∈ [ j ] , is ( c j + 1 − i ) − 1 = ( v k + 1 − i ) − 1 = v i ′ (c_{j+1-i})^{-1}=(v_{k+1-i})^{-1}=v'_{i} ( c j + 1 − i ) − 1 = ( v k + 1 − i ) − 1 = v i ′ by the definition of J J J ; the remaining letters of v ′ v' v ′ are those of b b b .
(2b) a b ab ab is reduced, of length k + k ′ − 2 j k+k'-2j k + k ′ − 2 j . The words a a a and b b b are reduced by (W3). If one of them is empty, a b ab ab is the other. If both are nonempty, then j < k j<k j < k and j < k ′ j<k' j < k ′ , so j + 1 ≤ min ( k , k ′ ) j+1\le\min(k,k') j + 1 ≤ min ( k , k ′ ) and j + 1 ∉ J j+1\notin J j + 1 ∈ / J by maximality; since the defining condition of J J J holds for all i ∈ [ j ] i\in[j] i ∈ [ j ] , it fails at i = j + 1 i=j+1 i = j + 1 , that is v j + 1 ′ ≠ ( v k − j ) − 1 v'_{j+1}\ne(v_{k-j})^{-1} v j + 1 ′ = ( v k − j ) − 1 . The consecutive letters of a b ab ab are consecutive letters of a a a , consecutive letters of b b b , or the pair a k − j = v k − j a_{k-j}=v_{k-j} a k − j = v k − j , b 1 = v j + 1 ′ b_{1}=v'_{j+1} b 1 = v j + 1 ′ ; so a b ab ab is reduced.
(2c) Δ ( v v ′ ) = Δ ( a b ) \Delta(vv')=\Delta(ab) Δ ( v v ′ ) = Δ ( ab ) , by (2a), associativity and (W5) applied to μ \mu μ and to ν \nu ν .
We write ρ ( v , v ′ ) = a b \rho(v,v')=ab ρ ( v , v ′ ) = ab . Since ∣ a b ∣ = k + k ′ − 2 j ≤ k + k ′ |ab|=k+k'-2j\le k+k' ∣ ab ∣ = k + k ′ − 2 j ≤ k + k ′ , ρ ( v , v ′ ) ∈ U 2 n \rho(v,v')\in U_{2n} ρ ( v , v ′ ) ∈ U 2 n whenever k , k ′ ≤ n k,k'\le n k , k ′ ≤ n .
For every reduced z ∈ U 2 n z\in U_{2n} z ∈ U 2 n fix, by Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §strip-exists , words q z ∈ W 2 d q_{z}\in W_{2d} q z ∈ W 2 d and u z ∈ W d ∘ u_{z}\in W^{\circ}_{d} u z ∈ W d ∘ with z = q z u z q z ∗ z=q_{z}\,u_{z}\,q_{z}^{*} z = q z u z q z ∗ and ∣ z ∣ = 2 ∣ q z ∣ + ∣ u z ∣ |z|=2|q_{z}|+|u_{z}| ∣ z ∣ = 2∣ q z ∣ + ∣ u z ∣ , and put t z = ∣ q z ∣ ∈ { 0 } ∪ N t_{z}=|q_{z}|\in\{0\}\cup\mathbb{N} t z = ∣ q z ∣ ∈ { 0 } ∪ N . By Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §strip-invariance , applied to μ \mu μ and to ν \nu ν ,
Δ ( z ) = Δ ( u z ) , (2.1) \Delta(z)=\Delta(u_{z}),\tag{2.1} Δ ( z ) = Δ ( u z ) , ( 2.1 )
which is 0 0 0 if u z = ∅ u_{z}=\varnothing u z = ∅ and equals x ( u z ) x(u_{z}) x ( u z ) otherwise (Conventions).
Weights. For k ∈ N k\in\mathbb{N} k ∈ N and w ∈ W d , k ∘ w\in W^{\circ}_{d,k} w ∈ W d , k ∘ put c ^ w = ϑ k ( k + 1 ) 2 ≥ 0 \hat c_{w}=\frac{\vartheta^{k}}{(k+1)^{2}}\ge0 c ^ w = ( k + 1 ) 2 ϑ k ≥ 0 ; then c ^ w 2 = θ k ( k + 1 ) 4 = c w \hat c_{w}^{2}=\frac{\theta^{k}}{(k+1)^{4}}=c_{w} c ^ w 2 = ( k + 1 ) 4 θ k = c w by The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §weights . For p ∈ N p\in\mathbb{N} p ∈ N and u ∈ W d , p ∘ u\in W^{\circ}_{d,p} u ∈ W d , p ∘ put Y ( u ) = p c ^ u ∣ x ( u ) ∣ ≥ 0 Y(u)=\sqrt{p}\,\hat c_{u}\,|x(u)|\ge0 Y ( u ) = p c ^ u ∣ x ( u ) ∣ ≥ 0 , so that Y ( u ) 2 = p c u ∣ x ( u ) ∣ 2 Y(u)^{2}=p\,c_{u}|x(u)|^{2} Y ( u ) 2 = p c u ∣ x ( u ) ∣ 2 . Define h : W 2 d → R h:W_{2d}\to\mathbb{R} h : W 2 d → R by h ( z ) = ϑ t z Y ( u z ) h(z)=\vartheta^{t_{z}}\,Y(u_{z}) h ( z ) = ϑ t z Y ( u z ) if z ∈ U 2 n z\in U_{2n} z ∈ U 2 n is reduced and u z ≠ ∅ u_{z}\ne\varnothing u z = ∅ , and h ( z ) = 0 h(z)=0 h ( z ) = 0 otherwise; h ≥ 0 h\ge0 h ≥ 0 .
Step 3 (slots, and expansion of the left side). Let W ≤ n ∘ W^{\circ}_{\le n} W ≤ n ∘ be as in The Truncated Cyclic Gradient of a Gauge Vector §gradient ; it is finite and nonempty, as it contains W d , 1 ∘ W^{\circ}_{d,1} W d , 1 ∘ (Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §finite ). Let S \mathcal{S} S be the set of slots α = ( w , m , s ) \alpha=(w,m,s) α = ( w , m , s ) with w ∈ W ≤ n ∘ w\in W^{\circ}_{\le n} w ∈ W ≤ n ∘ , m ∈ [ ∣ w ∣ ] m\in[|w|] m ∈ [ ∣ w ∣ ] and s ∈ { 1 , − 1 } s\in\{1,-1\} s ∈ { 1 , − 1 } ; as a set of dependent pairs ( w , ( m , s ) ) (w,(m,s)) ( w , ( m , s )) it is nonempty and finite. For such α \alpha α write w α = w w_{\alpha}=w w α = w , k α = ∣ w ∣ k_{\alpha}=|w| k α = ∣ w ∣ , and put
v α = r m ( w ) if s = 1 , v α = r m ( w ) ∗ if s = − 1 ; σ α = c w ∣ x ( w ) ∣ ; v_{\alpha}=r_{m}(w)\ \text{if }s=1,\qquad v_{\alpha}=r_{m}(w)^{*}\ \text{if }s=-1;\qquad\sigma_{\alpha}=c_{w}\,|x(w)|; v α = r m ( w ) if s = 1 , v α = r m ( w ) ∗ if s = − 1 ; σ α = c w ∣ x ( w ) ∣ ;
κ α = 1 2 c w x ( w ) ‾ i ε ( w m ) if s = 1 , κ α = 1 2 c w x ( w ) ‾ i ε ( w m ) ‾ if s = − 1. \kappa_{\alpha}=\tfrac12\,c_{w}\,\overline{x(w)}\;\mathrm{i}\,\varepsilon(w_{m})\ \text{if }s=1,\qquad\kappa_{\alpha}=\overline{\tfrac12\,c_{w}\,\overline{x(w)}\;\mathrm{i}\,\varepsilon(w_{m})}\ \text{if }s=-1. κ α = 2 1 c w x ( w ) i ε ( w m ) if s = 1 , κ α = 2 1 c w x ( w ) i ε ( w m ) if s = − 1.
By (W4), (W3) and (W2), v α v_{\alpha} v α is a reduced word of length k α ∈ [ n ] k_{\alpha}\in[n] k α ∈ [ n ] , so v α ∈ U n v_{\alpha}\in U_{n} v α ∈ U n ; and ∣ κ α ∣ = 1 2 σ α |\kappa_{\alpha}|=\frac12\sigma_{\alpha} ∣ κ α ∣ = 2 1 σ α , since c w > 0 c_{w}>0 c w > 0 , ∣ x ( w ) ‾ ∣ = ∣ x ( w ) ∣ |\overline{x(w)}|=|x(w)| ∣ x ( w ) ∣ = ∣ x ( w ) ∣ and ∣ i ∣ = ∣ ε ( w m ) ∣ = 1 |\mathrm{i}|=|\varepsilon(w_{m})|=1 ∣ i ∣ = ∣ ε ( w m ) ∣ = 1 . For α = ( w , m , s ) \alpha=(w,m,s) α = ( w , m , s ) put α ˉ = ( w , m , − s ) ∈ S \bar\alpha=(w,m,-s)\in\mathcal{S} α ˉ = ( w , m , − s ) ∈ S ; then v α ˉ = v α ∗ v_{\bar\alpha}=v_{\alpha}^{*} v α ˉ = v α ∗ (using ( r ∗ ) ∗ = r (r^{*})^{*}=r ( r ∗ ) ∗ = r from (W2)), κ α ˉ = κ α ‾ \kappa_{\bar\alpha}=\overline{\kappa_{\alpha}} κ α ˉ = κ α , σ α ˉ = σ α \sigma_{\bar\alpha}=\sigma_{\alpha} σ α ˉ = σ α and k α ˉ = k α k_{\bar\alpha}=k_{\alpha} k α ˉ = k α . For i ∈ [ d ] i\in[d] i ∈ [ d ] let S i \mathcal{S}_{i} S i be the set of ( w , m , s ) ∈ S (w,m,s)\in\mathcal{S} ( w , m , s ) ∈ S with g ( w m ) = i g(w_{m})=i g ( w m ) = i ; since g ( w m ) ∈ [ d ] g(w_{m})\in[d] g ( w m ) ∈ [ d ] is unique (Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters ), the sets S i \mathcal{S}_{i} S i are pairwise disjoint with union S \mathcal{S} S , and α ∈ S i \alpha\in\mathcal{S}_{i} α ∈ S i if and only if α ˉ ∈ S i \bar\alpha\in\mathcal{S}_{i} α ˉ ∈ S i .
(3a) Let i ∈ [ d ] i\in[d] i ∈ [ d ] and λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d , and write Ξ i = Ξ x , n i \Xi^{i}=\Xi^{i}_{x,n} Ξ i = Ξ x , n i , Z i = Z x , n i Z^{i}=Z^{i}_{x,n} Z i = Z x , n i . If S i \mathcal{S}_{i} S i is empty then λ ( Ξ i Ξ i ) = 0 \lambda(\Xi^{i}\Xi^{i})=0 λ ( Ξ i Ξ i ) = 0 ; otherwise
λ ( Ξ i Ξ i ) = ∑ ( α , β ) ∈ S i × S i κ α κ β λ ( v α v β ) . \lambda\bigl(\Xi^{i}\Xi^{i}\bigr)=\sum_{(\alpha,\beta)\in\mathcal{S}_{i}\times\mathcal{S}_{i}}\kappa_{\alpha}\kappa_{\beta}\,\lambda\bigl(v_{\alpha}v_{\beta}\bigr). λ ( Ξ i Ξ i ) = ( α , β ) ∈ S i × S i ∑ κ α κ β λ ( v α v β ) .
Proof. Let S i + \mathcal{S}^{+}_{i} S i + and S i − \mathcal{S}^{-}_{i} S i − be the slots in S i \mathcal{S}_{i} S i with s = 1 s=1 s = 1 and s = − 1 s=-1 s = − 1 ; α ↦ α ˉ \alpha\mapsto\bar\alpha α ↦ α ˉ is a bijection from S i + \mathcal{S}^{+}_{i} S i + onto S i − \mathcal{S}^{-}_{i} S i − . By The Truncated Cyclic Gradient of a Gauge Vector §gradient and Rotations and Cyclic Derivatives of Words in Unitary Letters §derivative , and since sums and complex multiples of word polynomials are pointwise (Word Polynomials in Unitary Letters: the Normed *-Algebra of Finitely Supported Word Functions §polynomials ), Z i Z^{i} Z i is the sum over w ∈ W ≤ n ∘ w\in W^{\circ}_{\le n} w ∈ W ≤ n ∘ of c w x ( w ) ‾ D w i c_{w}\overline{x(w)}\,D^{i}_{w} c w x ( w ) D w i , and D w i D^{i}_{w} D w i is a sum over the m ∈ [ ∣ w ∣ ] m\in[|w|] m ∈ [ ∣ w ∣ ] with g ( w m ) = i g(w_{m})=i g ( w m ) = i . A word w w w with no such m m m has D w i = 0 D^{i}_{w}=0 D w i = 0 , and its term is dropped by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing . If every word is dropped, then S i + \mathcal{S}^{+}_{i} S i + is empty and Z i = 0 Z^{i}=0 Z i = 0 by the same clause, which is the sum over the empty index set S i + \mathcal{S}^{+}_{i} S i + . Otherwise the remaining words have nonempty sets of such m m m , so iterated sums over dependent pairs (Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs ), reindexed along ( w , m ) ↦ ( w , m , 1 ) (w,m)\mapsto(w,m,1) ( w , m ) ↦ ( w , m , 1 ) , and homogeneity give Z i = ∑ α ∈ S i + 2 κ α e v α Z^{i}=\sum_{\alpha\in\mathcal{S}^{+}_{i}}2\kappa_{\alpha}\,e_{v_{\alpha}} Z i = ∑ α ∈ S i + 2 κ α e v α . By Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra (adjoints of sums and of complex multiples, ( e u ) ∗ = e u ∗ (e_{u})^{*}=e_{u^{*}} ( e u ) ∗ = e u ∗ ), applied finitely often, and reindexing along α ↦ α ˉ \alpha\mapsto\bar\alpha α ↦ α ˉ , ( Z i ) ∗ = ∑ α ∈ S i + 2 κ α ‾ e v α ∗ = ∑ α ∈ S i − 2 κ α e v α (Z^{i})^{*}=\sum_{\alpha\in\mathcal{S}^{+}_{i}}\overline{2\kappa_{\alpha}}\,e_{v_{\alpha}^{*}}=\sum_{\alpha\in\mathcal{S}^{-}_{i}}2\kappa_{\alpha}\,e_{v_{\alpha}} ( Z i ) ∗ = ∑ α ∈ S i + 2 κ α e v α ∗ = ∑ α ∈ S i − 2 κ α e v α . Hence Ξ i = 1 2 ( Z i + ( Z i ) ∗ ) = ∑ α ∈ S i κ α e v α \Xi^{i}=\frac12\bigl(Z^{i}+(Z^{i})^{*}\bigr)=\sum_{\alpha\in\mathcal{S}_{i}}\kappa_{\alpha}e_{v_{\alpha}} Ξ i = 2 1 ( Z i + ( Z i ) ∗ ) = ∑ α ∈ S i κ α e v α by the disjoint union S i = S i + ∪ S i − \mathcal{S}_{i}=\mathcal{S}^{+}_{i}\cup\mathcal{S}^{-}_{i} S i = S i + ∪ S i − . If S i \mathcal{S}_{i} S i is empty, all these sums are the zero word polynomial, so Ξ i = 0 \Xi^{i}=0 Ξ i = 0 , Ξ i Ξ i = 0 \Xi^{i}\Xi^{i}=0 Ξ i Ξ i = 0 by Word Polynomials in Unitary Letters: the Normed *-Algebra of Finitely Supported Word Functions §product , and λ ( 0 ) = 0 \lambda(0)=0 λ ( 0 ) = 0 by Evaluation of Word Polynomials by a Unitary Law §evaluation . Otherwise distributivity, compatibility with complex multiples and e u e u ′ = e u u ′ e_{u}e_{u'}=e_{uu'} e u e u ′ = e u u ′ from Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra , applied finitely often, give Ξ i Ξ i = ∑ ( α , β ) ∈ S i × S i κ α κ β e v α v β \Xi^{i}\Xi^{i}=\sum_{(\alpha,\beta)\in\mathcal{S}_{i}\times\mathcal{S}_{i}}\kappa_{\alpha}\kappa_{\beta}\,e_{v_{\alpha}v_{\beta}} Ξ i Ξ i = ∑ ( α , β ) ∈ S i × S i κ α κ β e v α v β , and Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §linear together with λ ( e z ) = λ ( z ) \lambda(e_{z})=\lambda(z) λ ( e z ) = λ ( z ) (Evaluation of Word Polynomials by a Unitary Law §evaluation ) gives the formula.
(3b) Put
S = ∑ ( α , β ) ∈ S × S σ α σ β ∣ Δ ( v α v β ) ∣ ≥ 0. S=\sum_{(\alpha,\beta)\in\mathcal{S}\times\mathcal{S}}\sigma_{\alpha}\sigma_{\beta}\,\bigl|\Delta(v_{\alpha}v_{\beta})\bigr|\ \ge0. S = ( α , β ) ∈ S × S ∑ σ α σ β Δ ( v α v β ) ≥ 0.
Then
∣ ∑ i ∈ [ d ] ( μ ( Ξ i Ξ i ) − ν ( Ξ i Ξ i ) ) ∣ ≤ 1 4 S . (3.1) \Bigl|\sum_{i\in[d]}\Bigl(\mu\bigl(\Xi^{i}\Xi^{i}\bigr)-\nu\bigl(\Xi^{i}\Xi^{i}\bigr)\Bigr)\Bigr|\le\frac14\,S.\tag{3.1} i ∈ [ d ] ∑ ( μ ( Ξ i Ξ i ) − ν ( Ξ i Ξ i ) ) ≤ 4 1 S . ( 3.1 )
Indeed, by (3a) and additivity, the i i i -th term is 0 0 0 if S i \mathcal{S}_{i} S i is empty and otherwise equals ∑ S i × S i κ α κ β Δ ( v α v β ) \sum_{\mathcal{S}_{i}\times\mathcal{S}_{i}}\kappa_{\alpha}\kappa_{\beta}\Delta(v_{\alpha}v_{\beta}) ∑ S i × S i κ α κ β Δ ( v α v β ) , whose modulus is at most ∑ S i × S i 1 4 σ α σ β ∣ Δ ( v α v β ) ∣ \sum_{\mathcal{S}_{i}\times\mathcal{S}_{i}}\frac14\sigma_{\alpha}\sigma_{\beta}|\Delta(v_{\alpha}v_{\beta})| ∑ S i × S i 4 1 σ α σ β ∣Δ ( v α v β ) ∣ . The modulus of the sum over i ∈ [ d ] i\in[d] i ∈ [ d ] is at most the sum of the moduli, and the sets S i × S i \mathcal{S}_{i}\times\mathcal{S}_{i} S i × S i with S i \mathcal{S}_{i} S i nonempty are pairwise disjoint subsets of S × S \mathcal{S}\times\mathcal{S} S × S ; as all terms are nonnegative, splitting over the disjoint union and enlarging the index set give (3.1).
Step 4 (the short word may be placed on the left). Let P 1 \mathcal{P}_{1} P 1 be the set of ( α , β ) ∈ S × S (\alpha,\beta)\in\mathcal{S}\times\mathcal{S} ( α , β ) ∈ S × S with k α ≤ k β k_{\alpha}\le k_{\beta} k α ≤ k β (nonempty, as it contains the pairs ( α , α ) (\alpha,\alpha) ( α , α ) ), P 2 \mathcal{P}_{2} P 2 its complement in S × S \mathcal{S}\times\mathcal{S} S × S , and S 1 S_{1} S 1 , S 2 S_{2} S 2 the sums defining S S S restricted to P 1 \mathcal{P}_{1} P 1 , P 2 \mathcal{P}_{2} P 2 (S 2 = 0 S_{2}=0 S 2 = 0 if P 2 \mathcal{P}_{2} P 2 is empty), so S = S 1 + S 2 S=S_{1}+S_{2} S = S 1 + S 2 . The map ( α , β ) ↦ ( β ˉ , α ˉ ) (\alpha,\beta)\mapsto(\bar\beta,\bar\alpha) ( α , β ) ↦ ( β ˉ , α ˉ ) is injective (it is its own inverse on S × S \mathcal{S}\times\mathcal{S} S × S ) and maps P 2 \mathcal{P}_{2} P 2 into P 1 \mathcal{P}_{1} P 1 , since k β ˉ = k β < k α = k α ˉ k_{\bar\beta}=k_{\beta}<k_{\alpha}=k_{\bar\alpha} k β ˉ = k β < k α = k α ˉ . By Step 3 and (W2), v β ˉ v α ˉ = v β ∗ v α ∗ = ( v α v β ) ∗ v_{\bar\beta}v_{\bar\alpha}=v_{\beta}^{*}v_{\alpha}^{*}=(v_{\alpha}v_{\beta})^{*} v β ˉ v α ˉ = v β ∗ v α ∗ = ( v α v β ) ∗ , so by Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §adjoint for μ \mu μ and ν \nu ν and the rules for conjugation, Δ ( v β ˉ v α ˉ ) = Δ ( v α v β ) ‾ \Delta(v_{\bar\beta}v_{\bar\alpha})=\overline{\Delta(v_{\alpha}v_{\beta})} Δ ( v β ˉ v α ˉ ) = Δ ( v α v β ) , which has the same modulus; also σ β ˉ σ α ˉ = σ α σ β \sigma_{\bar\beta}\sigma_{\bar\alpha}=\sigma_{\alpha}\sigma_{\beta} σ β ˉ σ α ˉ = σ α σ β . Reindexing S 2 S_{2} S 2 along this map and enlarging the index set to P 1 \mathcal{P}_{1} P 1 (nonnegative terms) gives S 2 ≤ S 1 S_{2}\le S_{1} S 2 ≤ S 1 , hence
S ≤ 2 S 1 . (4.1) S\le2S_{1}.\tag{4.1} S ≤ 2 S 1 . ( 4.1 )
Step 5 (one term). For a slot α \alpha α with w = w α w=w_{\alpha} w = w α and k = k α k=k_{\alpha} k = k α put
ϕ α = c ^ w ∣ x ( w ) ∣ ( k + 1 ) 2 , ψ α = c ^ w ∣ x ( w ) ∣ k , \phi_{\alpha}=\frac{\hat c_{w}\,|x(w)|}{(k+1)^{2}},\qquad\psi_{\alpha}=\frac{\hat c_{w}\,|x(w)|}{\sqrt{k}}, ϕ α = ( k + 1 ) 2 c ^ w ∣ x ( w ) ∣ , ψ α = k c ^ w ∣ x ( w ) ∣ ,
both nonnegative, with ϕ α 2 = c w ∣ x ( w ) ∣ 2 ( k + 1 ) 4 \phi_{\alpha}^{2}=\frac{c_{w}|x(w)|^{2}}{(k+1)^{4}} ϕ α 2 = ( k + 1 ) 4 c w ∣ x ( w ) ∣ 2 and ψ α 2 = c w ∣ x ( w ) ∣ 2 k \psi_{\alpha}^{2}=\frac{c_{w}|x(w)|^{2}}{k} ψ α 2 = k c w ∣ x ( w ) ∣ 2 . We show: for every ( α , β ) ∈ P 1 (\alpha,\beta)\in\mathcal{P}_{1} ( α , β ) ∈ P 1 ,
σ α σ β ∣ Δ ( v α v β ) ∣ ≤ 8 ϑ j ( v α , v β ) ϕ α ψ β h ( ρ ( v α , v β ) ) . (5.1) \sigma_{\alpha}\sigma_{\beta}\,\bigl|\Delta(v_{\alpha}v_{\beta})\bigr|\le8\,\vartheta^{\,j(v_{\alpha},v_{\beta})}\,\phi_{\alpha}\,\psi_{\beta}\,h\bigl(\rho(v_{\alpha},v_{\beta})\bigr).\tag{5.1} σ α σ β Δ ( v α v β ) ≤ 8 ϑ j ( v α , v β ) ϕ α ψ β h ( ρ ( v α , v β ) ) . ( 5.1 )
Write w = w α w=w_{\alpha} w = w α , w ′ = w β w'=w_{\beta} w ′ = w β , k = k α ≤ k ′ = k β k=k_{\alpha}\le k'=k_{\beta} k = k α ≤ k ′ = k β , v = v α v=v_{\alpha} v = v α , v ′ = v β v'=v_{\beta} v ′ = v β , j = j ( v , v ′ ) j=j(v,v') j = j ( v , v ′ ) and z = ρ ( v , v ′ ) z=\rho(v,v') z = ρ ( v , v ′ ) , a reduced word in U 2 n U_{2n} U 2 n of length k + k ′ − 2 j k+k'-2j k + k ′ − 2 j by (2b). By (2c) and (2.1), Δ ( v v ′ ) = Δ ( u z ) \Delta(vv')=\Delta(u_{z}) Δ ( v v ′ ) = Δ ( u z ) . If u z = ∅ u_{z}=\varnothing u z = ∅ , the left side of (5.1) is 0 0 0 , and so is the right side. Otherwise put t = t z t=t_{z} t = t z , p = ∣ u z ∣ ∈ N p=|u_{z}|\in\mathbb{N} p = ∣ u z ∣ ∈ N , u = u z ∈ W d , p ∘ u=u_{z}\in W^{\circ}_{d,p} u = u z ∈ W d , p ∘ and N = j + t N=j+t N = j + t ; then ∣ Δ ( v v ′ ) ∣ = ∣ x ( u ) ∣ |\Delta(vv')|=|x(u)| ∣Δ ( v v ′ ) ∣ = ∣ x ( u ) ∣ , h ( z ) = ϑ t Y ( u ) h(z)=\vartheta^{t}Y(u) h ( z ) = ϑ t Y ( u ) , and k + k ′ = ∣ z ∣ + 2 j = p + 2 N k+k'=|z|+2j=p+2N k + k ′ = ∣ z ∣ + 2 j = p + 2 N . Since c w = c ^ w ϑ k ( k + 1 ) 2 c_{w}=\hat c_{w}\frac{\vartheta^{k}}{(k+1)^{2}} c w = c ^ w ( k + 1 ) 2 ϑ k , c w ′ = c ^ w ′ ϑ k ′ ( k ′ + 1 ) 2 c_{w'}=\hat c_{w'}\frac{\vartheta^{k'}}{(k'+1)^{2}} c w ′ = c ^ w ′ ( k ′ + 1 ) 2 ϑ k ′ and ϑ k ϑ k ′ = ϑ k + k ′ = ϑ 2 N ϑ p \vartheta^{k}\vartheta^{k'}=\vartheta^{k+k'}=\vartheta^{2N}\vartheta^{p} ϑ k ϑ k ′ = ϑ k + k ′ = ϑ 2 N ϑ p , and since ϑ p ∣ x ( u ) ∣ = ( p + 1 ) 2 c ^ u ∣ x ( u ) ∣ = ( p + 1 ) 2 p Y ( u ) \vartheta^{p}|x(u)|=(p+1)^{2}\hat c_{u}|x(u)|=\frac{(p+1)^{2}}{\sqrt{p}}Y(u) ϑ p ∣ x ( u ) ∣ = ( p + 1 ) 2 c ^ u ∣ x ( u ) ∣ = p ( p + 1 ) 2 Y ( u ) ,
σ α σ β ∣ x ( u ) ∣ = c ^ w ∣ x ( w ) ∣ c ^ w ′ ∣ x ( w ′ ) ∣ ϑ 2 N ( p + 1 ) 2 ( k + 1 ) 2 ( k ′ + 1 ) 2 p Y ( u ) . \sigma_{\alpha}\sigma_{\beta}|x(u)|=\hat c_{w}|x(w)|\;\hat c_{w'}|x(w')|\;\vartheta^{2N}\;\frac{(p+1)^{2}}{(k+1)^{2}(k'+1)^{2}\sqrt{p}}\;Y(u). σ α σ β ∣ x ( u ) ∣ = c ^ w ∣ x ( w ) ∣ c ^ w ′ ∣ x ( w ′ ) ∣ ϑ 2 N ( k + 1 ) 2 ( k ′ + 1 ) 2 p ( p + 1 ) 2 Y ( u ) .
Now p ≤ k + k ′ ≤ 2 k ′ p\le k+k'\le2k' p ≤ k + k ′ ≤ 2 k ′ , so 0 < p + 1 ≤ 2 ( k ′ + 1 ) 0<p+1\le2(k'+1) 0 < p + 1 ≤ 2 ( k ′ + 1 ) and ( p + 1 ) 2 ≤ 4 ( k ′ + 1 ) 2 (p+1)^{2}\le4(k'+1)^{2} ( p + 1 ) 2 ≤ 4 ( k ′ + 1 ) 2 ; hence the right side is at most 4 ϕ α c ^ w ′ ∣ x ( w ′ ) ∣ ϑ N Y ( u ) ⋅ ϑ N p 4\,\phi_{\alpha}\;\hat c_{w'}|x(w')|\;\vartheta^{N}Y(u)\cdot\frac{\vartheta^{N}}{\sqrt{p}} 4 ϕ α c ^ w ′ ∣ x ( w ′ ) ∣ ϑ N Y ( u ) ⋅ p ϑ N . By (C5) (with N = j + t N=j+t N = j + t ), ( ϑ N k ′ ) 2 = θ N k ′ ≤ 4 p = ( 2 p ) 2 (\vartheta^{N}\sqrt{k'})^{2}=\theta^{N}k'\le4p=(2\sqrt{p})^{2} ( ϑ N k ′ ) 2 = θ N k ′ ≤ 4 p = ( 2 p ) 2 , so ϑ N k ′ ≤ 2 p \vartheta^{N}\sqrt{k'}\le2\sqrt{p} ϑ N k ′ ≤ 2 p , that is ϑ N p ≤ 2 k ′ \frac{\vartheta^{N}}{\sqrt{p}}\le\frac{2}{\sqrt{k'}} p ϑ N ≤ k ′ 2 . Therefore
σ α σ β ∣ Δ ( v v ′ ) ∣ ≤ 8 ϕ α c ^ w ′ ∣ x ( w ′ ) ∣ k ′ ϑ j ϑ t Y ( u ) = 8 ϑ j ϕ α ψ β h ( z ) , \sigma_{\alpha}\sigma_{\beta}|\Delta(vv')|\le8\,\phi_{\alpha}\,\frac{\hat c_{w'}|x(w')|}{\sqrt{k'}}\,\vartheta^{j}\,\vartheta^{t}Y(u)=8\,\vartheta^{j}\phi_{\alpha}\psi_{\beta}\,h(z), σ α σ β ∣Δ ( v v ′ ) ∣ ≤ 8 ϕ α k ′ c ^ w ′ ∣ x ( w ′ ) ∣ ϑ j ϑ t Y ( u ) = 8 ϑ j ϕ α ψ β h ( z ) ,
which is (5.1).
Step 6 (grouping the slots by their words). For v ∈ W 2 d v\in W_{2d} v ∈ W 2 d let S ( v ) \mathcal{S}(v) S ( v ) be the set of slots α \alpha α with v α = v v_{\alpha}=v v α = v , and put f ( v ) = ∑ α ∈ S ( v ) ϕ α f(v)=\sum_{\alpha\in\mathcal{S}(v)}\phi_{\alpha} f ( v ) = ∑ α ∈ S ( v ) ϕ α and ℓ ( v ) = ∑ α ∈ S ( v ) ψ α \ell(v)=\sum_{\alpha\in\mathcal{S}(v)}\psi_{\alpha} ℓ ( v ) = ∑ α ∈ S ( v ) ψ α (both 0 0 0 if S ( v ) \mathcal{S}(v) S ( v ) is empty); f , ℓ ≥ 0 f,\ell\ge0 f , ℓ ≥ 0 . Let V = { v α : α ∈ S } ⊆ U n V=\{v_{\alpha}:\alpha\in\mathcal{S}\}\subseteq U_{n} V = { v α : α ∈ S } ⊆ U n , nonempty and finite; f f f and ℓ \ell ℓ vanish outside V V V , and the sets S ( v ) \mathcal{S}(v) S ( v ) , v ∈ V v\in V v ∈ V , are nonempty, pairwise disjoint, with union S \mathcal{S} S . For k ∈ [ n ] k\in[n] k ∈ [ n ] let S ( k ) \mathcal{S}_{(k)} S ( k ) be the set of slots with k α = k k_{\alpha}=k k α = k ; it is the set of dependent pairs ( w , ( m , s ) ) (w,(m,s)) ( w , ( m , s )) with w ∈ W d , k ∘ w\in W^{\circ}_{d,k} w ∈ W d , k ∘ and ( m , s ) ∈ [ k ] × { 1 , − 1 } (m,s)\in[k]\times\{1,-1\} ( m , s ) ∈ [ k ] × { 1 , − 1 } , a set with 2 k 2k 2 k elements for each w w w (by The Product of Two Sums over Finite Index Sets is a Sum over the Cartesian Product and claim 1 of Properties of a Sum over a Finite Index Set ), so that for every map χ \chi χ on W d , k ∘ W^{\circ}_{d,k} W d , k ∘
∑ α ∈ S ( k ) χ ( w α ) = ∑ w ∈ W d , k ∘ 2 k χ ( w ) . (6.1) \sum_{\alpha\in\mathcal{S}_{(k)}}\chi(w_{\alpha})=\sum_{w\in W^{\circ}_{d,k}}2k\,\chi(w).\tag{6.1} α ∈ S ( k ) ∑ χ ( w α ) = w ∈ W d , k ∘ ∑ 2 k χ ( w ) . ( 6.1 )
(6a) If v ∈ V v\in V v ∈ V has length k k k , then ∑ α ∈ S ( v ) 1 ≤ 4 k \sum_{\alpha\in\mathcal{S}(v)}1\le4k ∑ α ∈ S ( v ) 1 ≤ 4 k . Indeed, the map α = ( w , m , s ) ↦ ( m , s , ε ( w m ) ) \alpha=(w,m,s)\mapsto(m,s,\varepsilon(w_{m})) α = ( w , m , s ) ↦ ( m , s , ε ( w m )) from S ( v ) \mathcal{S}(v) S ( v ) to [ k ] × { 1 , − 1 } × { 1 , − 1 } [k]\times\{1,-1\}\times\{1,-1\} [ k ] × { 1 , − 1 } × { 1 , − 1 } (note k α = ∣ v ∣ = k k_{\alpha}=|v|=k k α = ∣ v ∣ = k ) is injective: given ( m , s , e ) (m,s,e) ( m , s , e ) , the word r m ( w ) r_{m}(w) r m ( w ) equals v v v if s = 1 s=1 s = 1 and v ∗ v^{*} v ∗ if s = − 1 s=-1 s = − 1 (by (W2)), and by (W4) w w w is determined by r m ( w ) r_{m}(w) r m ( w ) , m m m and e = ε ( w m ) e=\varepsilon(w_{m}) e = ε ( w m ) . Reindexing onto the image and enlarging to the product set, which has 4 k 4k 4 k elements, gives the claim.
(6b) For k ∈ [ n ] k\in[n] k ∈ [ n ] , Φ k : = ∑ v ∈ [ 2 d ] k f ( v ) 2 ≤ 8 k X k \Phi_{k}:=\sqrt{\sum_{v\in[2d]^{k}}f(v)^{2}}\le\frac{\sqrt{8}}{k}\sqrt{X_{k}} Φ k := ∑ v ∈ [ 2 d ] k f ( v ) 2 ≤ k 8 X k . Indeed, for v ∈ V v\in V v ∈ V of length k k k , (C1) and (6a) give f ( v ) 2 ≤ 4 k ∑ α ∈ S ( v ) ϕ α 2 = 4 k ( k + 1 ) 4 ∑ α ∈ S ( v ) c w α ∣ x ( w α ) ∣ 2 f(v)^{2}\le4k\sum_{\alpha\in\mathcal{S}(v)}\phi_{\alpha}^{2}=\frac{4k}{(k+1)^{4}}\sum_{\alpha\in\mathcal{S}(v)}c_{w_{\alpha}}|x(w_{\alpha})|^{2} f ( v ) 2 ≤ 4 k ∑ α ∈ S ( v ) ϕ α 2 = ( k + 1 ) 4 4 k ∑ α ∈ S ( v ) c w α ∣ x ( w α ) ∣ 2 , and f ( v ) = 0 f(v)=0 f ( v ) = 0 for v ∈ [ 2 d ] k ∖ V v\in[2d]^{k}\setminus V v ∈ [ 2 d ] k ∖ V . The sets S ( v ) \mathcal{S}(v) S ( v ) , v ∈ V ∩ [ 2 d ] k v\in V\cap[2d]^{k} v ∈ V ∩ [ 2 d ] k , are pairwise disjoint with union S ( k ) \mathcal{S}_{(k)} S ( k ) (as ∣ v α ∣ = k α |v_{\alpha}|=k_{\alpha} ∣ v α ∣ = k α ). Summing and using (6.1),
∑ v ∈ [ 2 d ] k f ( v ) 2 ≤ 4 k ( k + 1 ) 4 ∑ α ∈ S ( k ) c w α ∣ x ( w α ) ∣ 2 = 8 k 2 ( k + 1 ) 4 X k ≤ 8 k 2 X k , \sum_{v\in[2d]^{k}}f(v)^{2}\le\frac{4k}{(k+1)^{4}}\sum_{\alpha\in\mathcal{S}_{(k)}}c_{w_{\alpha}}|x(w_{\alpha})|^{2}=\frac{8k^{2}}{(k+1)^{4}}X_{k}\le\frac{8}{k^{2}}X_{k}, v ∈ [ 2 d ] k ∑ f ( v ) 2 ≤ ( k + 1 ) 4 4 k α ∈ S ( k ) ∑ c w α ∣ x ( w α ) ∣ 2 = ( k + 1 ) 4 8 k 2 X k ≤ k 2 8 X k ,
the last step because k 4 ≤ ( k + 1 ) 4 k^{4}\le(k+1)^{4} k 4 ≤ ( k + 1 ) 4 . Taking square roots gives (6b).
(6c) Λ : = ∑ z ∈ U 2 n ℓ ( z ) 2 ≤ 8 ∥ x ∥ d , 1 \Lambda:=\sqrt{\sum_{z\in U_{2n}}\ell(z)^{2}}\le\sqrt{8}\,\lVert x\rVert_{d,1} Λ := ∑ z ∈ U 2 n ℓ ( z ) 2 ≤ 8 ∥ x ∥ d , 1 . Indeed, for v ∈ V v\in V v ∈ V of length k k k , (C1) and (6a) give ℓ ( v ) 2 ≤ 4 k ∑ α ∈ S ( v ) ψ α 2 = 4 ∑ α ∈ S ( v ) c w α ∣ x ( w α ) ∣ 2 \ell(v)^{2}\le4k\sum_{\alpha\in\mathcal{S}(v)}\psi_{\alpha}^{2}=4\sum_{\alpha\in\mathcal{S}(v)}c_{w_{\alpha}}|x(w_{\alpha})|^{2} ℓ ( v ) 2 ≤ 4 k ∑ α ∈ S ( v ) ψ α 2 = 4 ∑ α ∈ S ( v ) c w α ∣ x ( w α ) ∣ 2 , and ℓ \ell ℓ vanishes on U 2 n ∖ V U_{2n}\setminus V U 2 n ∖ V . Since the S ( v ) \mathcal{S}(v) S ( v ) , v ∈ V v\in V v ∈ V , partition S \mathcal{S} S , and S \mathcal{S} S is the disjoint union of the S ( k ) \mathcal{S}_{(k)} S ( k ) , k ∈ [ n ] k\in[n] k ∈ [ n ] , (6.1) and (0.1) give
∑ z ∈ U 2 n ℓ ( z ) 2 ≤ 4 ∑ α ∈ S c w α ∣ x ( w α ) ∣ 2 = 4 ∑ k = 1 n 2 k X k ≤ 8 ∥ x ∥ d , 1 2 . \sum_{z\in U_{2n}}\ell(z)^{2}\le4\sum_{\alpha\in\mathcal{S}}c_{w_{\alpha}}|x(w_{\alpha})|^{2}=4\sum_{k=1}^{n}2k\,X_{k}\le8\,\lVert x\rVert_{d,1}^{2}. z ∈ U 2 n ∑ ℓ ( z ) 2 ≤ 4 α ∈ S ∑ c w α ∣ x ( w α ) ∣ 2 = 4 k = 1 ∑ n 2 k X k ≤ 8 ∥ x ∥ d , 1 2 .
(6d) Regrouping. For every map Ψ : V × V → R \Psi:V\times V\to\mathbb{R} Ψ : V × V → R ,
∑ ( α , β ) ∈ S × S Ψ ( v α , v β ) ϕ α ψ β = ∑ ( v , v ′ ) ∈ V × V Ψ ( v , v ′ ) f ( v ) ℓ ( v ′ ) , \sum_{(\alpha,\beta)\in\mathcal{S}\times\mathcal{S}}\Psi(v_{\alpha},v_{\beta})\,\phi_{\alpha}\psi_{\beta}=\sum_{(v,v')\in V\times V}\Psi(v,v')\,f(v)\,\ell(v'), ( α , β ) ∈ S × S ∑ Ψ ( v α , v β ) ϕ α ψ β = ( v , v ′ ) ∈ V × V ∑ Ψ ( v , v ′ ) f ( v ) ℓ ( v ′ ) ,
because S × S \mathcal{S}\times\mathcal{S} S × S is the disjoint union of the sets S ( v ) × S ( v ′ ) \mathcal{S}(v)\times\mathcal{S}(v') S ( v ) × S ( v ′ ) , ( v , v ′ ) ∈ V × V (v,v')\in V\times V ( v , v ′ ) ∈ V × V , on each of which Ψ ( v α , v β ) = Ψ ( v , v ′ ) \Psi(v_{\alpha},v_{\beta})=\Psi(v,v') Ψ ( v α , v β ) = Ψ ( v , v ′ ) , and ∑ S ( v ) × S ( v ′ ) ϕ α ψ β = f ( v ) ℓ ( v ′ ) \sum_{\mathcal{S}(v)\times\mathcal{S}(v')}\phi_{\alpha}\psi_{\beta}=f(v)\ell(v') ∑ S ( v ) × S ( v ′ ) ϕ α ψ β = f ( v ) ℓ ( v ′ ) by The Product of Two Sums over Finite Index Sets is a Sum over the Cartesian Product .
Step 7 (the Cauchy--Schwarz argument over cancellations). Put Ω = ∑ z ∈ U 2 n h ( z ) 2 \Omega=\sqrt{\sum_{z\in U_{2n}}h(z)^{2}} Ω = ∑ z ∈ U 2 n h ( z ) 2 . For k ∈ [ n ] k\in[n] k ∈ [ n ] and j ∈ { 0 , 1 , … , k } j\in\{0,1,\dots,k\} j ∈ { 0 , 1 , … , k } let Q k , j Q_{k,j} Q k , j be the set of ( v , v ′ ) ∈ V × V (v,v')\in V\times V ( v , v ′ ) ∈ V × V with ∣ v ∣ = k |v|=k ∣ v ∣ = k and j ( v , v ′ ) = j j(v,v')=j j ( v , v ′ ) = j . We show
∑ ( v , v ′ ) ∈ Q k , j f ( v ) ℓ ( v ′ ) h ( ρ ( v , v ′ ) ) ≤ Φ k Λ Ω (7.1) \sum_{(v,v')\in Q_{k,j}}f(v)\,\ell(v')\,h\bigl(\rho(v,v')\bigr)\le\Phi_{k}\,\Lambda\,\Omega\tag{7.1} ( v , v ′ ) ∈ Q k , j ∑ f ( v ) ℓ ( v ′ ) h ( ρ ( v , v ′ ) ) ≤ Φ k Λ Ω ( 7.1 )
(the left side being 0 0 0 if Q k , j Q_{k,j} Q k , j is empty). Put A = [ 2 d ] k − j \mathcal{A}=[2d]^{k-j} A = [ 2 d ] k − j , C = [ 2 d ] j \mathcal{C}=[2d]^{j} C = [ 2 d ] j and B = U n \mathcal{B}=U_{n} B = U n . By Step 2, each ( v , v ′ ) ∈ Q k , j (v,v')\in Q_{k,j} ( v , v ′ ) ∈ Q k , j determines words a ∈ A a\in\mathcal{A} a ∈ A , c ∈ C c\in\mathcal{C} c ∈ C , b ∈ B b\in\mathcal{B} b ∈ B (indeed ∣ b ∣ = ∣ v ′ ∣ − j ≤ n |b|=|v'|-j\le n ∣ b ∣ = ∣ v ′ ∣ − j ≤ n ) with v = a c v=ac v = a c , v ′ = c ∗ b v'=c^{*}b v ′ = c ∗ b and ρ ( v , v ′ ) = a b \rho(v,v')=ab ρ ( v , v ′ ) = ab , and ( v , v ′ ) ↦ ( a , c , b ) (v,v')\mapsto(a,c,b) ( v , v ′ ) ↦ ( a , c , b ) is injective since ( v , v ′ ) = ( a c , c ∗ b ) (v,v')=(ac,c^{*}b) ( v , v ′ ) = ( a c , c ∗ b ) . Reindexing onto the image and enlarging to C × ( A × B ) \mathcal{C}\times(\mathcal{A}\times\mathcal{B}) C × ( A × B ) (nonnegative terms; iterated sums as sums over pairs), the left side of (7.1) is at most
∑ c ∈ C ∑ ( a , b ) ∈ A × B f ( a c ) ℓ ( c ∗ b ) h ( a b ) . \sum_{c\in\mathcal{C}}\ \sum_{(a,b)\in\mathcal{A}\times\mathcal{B}}f(ac)\,\ell(c^{*}b)\,h(ab). c ∈ C ∑ ( a , b ) ∈ A × B ∑ f ( a c ) ℓ ( c ∗ b ) h ( ab ) .
Fix c ∈ C c\in\mathcal{C} c ∈ C and put η ( c ) = ∑ a ∈ A f ( a c ) 2 \eta(c)=\sqrt{\sum_{a\in\mathcal{A}}f(ac)^{2}} η ( c ) = ∑ a ∈ A f ( a c ) 2 and ζ ( c ) = ∑ b ∈ B ℓ ( c ∗ b ) 2 \zeta(c)=\sqrt{\sum_{b\in\mathcal{B}}\ell(c^{*}b)^{2}} ζ ( c ) = ∑ b ∈ B ℓ ( c ∗ b ) 2 . By (C1) on A × B \mathcal{A}\times\mathcal{B} A × B and The Product of Two Sums over Finite Index Sets is a Sum over the Cartesian Product ,
∑ ( a , b ) f ( a c ) ℓ ( c ∗ b ) h ( a b ) ≤ ∑ ( a , b ) f ( a c ) 2 ℓ ( c ∗ b ) 2 ∑ ( a , b ) h ( a b ) 2 = η ( c ) ζ ( c ) ∑ ( a , b ) h ( a b ) 2 . \sum_{(a,b)}f(ac)\ell(c^{*}b)h(ab)\le\sqrt{\textstyle\sum_{(a,b)}f(ac)^{2}\ell(c^{*}b)^{2}}\;\sqrt{\textstyle\sum_{(a,b)}h(ab)^{2}}=\eta(c)\,\zeta(c)\,\sqrt{\textstyle\sum_{(a,b)}h(ab)^{2}}. ( a , b ) ∑ f ( a c ) ℓ ( c ∗ b ) h ( ab ) ≤ ∑ ( a , b ) f ( a c ) 2 ℓ ( c ∗ b ) 2 ∑ ( a , b ) h ( ab ) 2 = η ( c ) ζ ( c ) ∑ ( a , b ) h ( ab ) 2 .
The map ( a , b ) ↦ a b (a,b)\mapsto ab ( a , b ) ↦ ab is injective on A × B \mathcal{A}\times\mathcal{B} A × B (the length of a a a being fixed), with values in U 2 n U_{2n} U 2 n since ∣ a b ∣ ≤ ( k − j ) + n ≤ 2 n |ab|\le(k-j)+n\le2n ∣ ab ∣ ≤ ( k − j ) + n ≤ 2 n ; so ∑ ( a , b ) h ( a b ) 2 ≤ Ω 2 \sum_{(a,b)}h(ab)^{2}\le\Omega^{2} ∑ ( a , b ) h ( ab ) 2 ≤ Ω 2 by reindexing and enlarging the index set, and the inner sum is at most η ( c ) ζ ( c ) Ω \eta(c)\zeta(c)\Omega η ( c ) ζ ( c ) Ω . By (C1) on C \mathcal{C} C , ∑ c η ( c ) ζ ( c ) ≤ ∑ c η ( c ) 2 ∑ c ζ ( c ) 2 \sum_{c}\eta(c)\zeta(c)\le\sqrt{\sum_{c}\eta(c)^{2}}\sqrt{\sum_{c}\zeta(c)^{2}} ∑ c η ( c ) ζ ( c ) ≤ ∑ c η ( c ) 2 ∑ c ζ ( c ) 2 . Here ∑ c η ( c ) 2 = ∑ ( a , c ) ∈ A × C f ( a c ) 2 = ∑ v ∈ [ 2 d ] k f ( v ) 2 = Φ k 2 \sum_{c}\eta(c)^{2}=\sum_{(a,c)\in\mathcal{A}\times\mathcal{C}}f(ac)^{2}=\sum_{v\in[2d]^{k}}f(v)^{2}=\Phi_{k}^{2} ∑ c η ( c ) 2 = ∑ ( a , c ) ∈ A × C f ( a c ) 2 = ∑ v ∈ [ 2 d ] k f ( v ) 2 = Φ k 2 , because ( a , c ) ↦ a c (a,c)\mapsto ac ( a , c ) ↦ a c is a bijection from A × C \mathcal{A}\times\mathcal{C} A × C onto [ 2 d ] k [2d]^{k} [ 2 d ] k (Conventions); and ∑ c ζ ( c ) 2 = ∑ ( c , b ) ∈ C × B ℓ ( c ∗ b ) 2 ≤ Λ 2 \sum_{c}\zeta(c)^{2}=\sum_{(c,b)\in\mathcal{C}\times\mathcal{B}}\ell(c^{*}b)^{2}\le\Lambda^{2} ∑ c ζ ( c ) 2 = ∑ ( c , b ) ∈ C × B ℓ ( c ∗ b ) 2 ≤ Λ 2 , because ( c , b ) ↦ c ∗ b (c,b)\mapsto c^{*}b ( c , b ) ↦ c ∗ b is injective on C × B \mathcal{C}\times\mathcal{B} C × B (the first j j j letters of c ∗ b c^{*}b c ∗ b form c ∗ c^{*} c ∗ , which determines c = ( c ∗ ) ∗ c=(c^{*})^{*} c = ( c ∗ ) ∗ by (W2), and the rest is b b b ) with values in U 2 n U_{2n} U 2 n , as ∣ c ∗ b ∣ ≤ j + n ≤ 2 n |c^{*}b|\le j+n\le2n ∣ c ∗ b ∣ ≤ j + n ≤ 2 n . Combining the three estimates gives (7.1).
Step 8 (the norm of h h h ). We show Ω 2 ≤ 2 ∥ x ∥ d , 1 2 \Omega^{2}\le2\,\lVert x\rVert_{d,1}^{2} Ω 2 ≤ 2 ∥ x ∥ d , 1 2 . Let Z Z Z be the set of reduced z ∈ U 2 n z\in U_{2n} z ∈ U 2 n with u z ≠ ∅ u_{z}\ne\varnothing u z = ∅ ; h h h vanishes on U 2 n ∖ Z U_{2n}\setminus Z U 2 n ∖ Z , so Ω 2 = ∑ z ∈ Z h ( z ) 2 \Omega^{2}=\sum_{z\in Z}h(z)^{2} Ω 2 = ∑ z ∈ Z h ( z ) 2 , which is 0 0 0 if Z Z Z is empty. For z ∈ Z z\in Z z ∈ Z , h ( z ) 2 = θ t z Y ( u z ) 2 h(z)^{2}=\theta^{t_{z}}Y(u_{z})^{2} h ( z ) 2 = θ t z Y ( u z ) 2 , u z u_{z} u z lies in the finite set U = ⋃ p ∈ [ 2 n ] W d , p ∘ \mathcal{U}=\bigcup_{p\in[2n]}W^{\circ}_{d,p} U = ⋃ p ∈ [ 2 n ] W d , p ∘ (as 1 ≤ ∣ u z ∣ ≤ ∣ z ∣ ≤ 2 n 1\le|u_{z}|\le|z|\le2n 1 ≤ ∣ u z ∣ ≤ ∣ z ∣ ≤ 2 n ), and t z ∈ { 0 , 1 , … , n } t_{z}\in\{0,1,\dots,n\} t z ∈ { 0 , 1 , … , n } (as 2 t z ≤ ∣ z ∣ ≤ 2 n 2t_{z}\le|z|\le2n 2 t z ≤ ∣ z ∣ ≤ 2 n ). For t ∈ { 0 , … , n } t\in\{0,\dots,n\} t ∈ { 0 , … , n } and u ∈ U u\in\mathcal{U} u ∈ U let Z t , u Z_{t,u} Z t , u be the set of z ∈ Z z\in Z z ∈ Z with t z = t t_{z}=t t z = t and u z = u u_{z}=u u z = u ; these sets are pairwise disjoint with union Z Z Z . On Z t , u Z_{t,u} Z t , u the map z ↦ q z ∈ [ 2 d ] t z\mapsto q_{z}\in[2d]^{t} z ↦ q z ∈ [ 2 d ] t is injective, since z = q z u q z ∗ z=q_{z}\,u\,q_{z}^{*} z = q z u q z ∗ ; hence ∑ z ∈ Z t , u 1 ≤ 1 \sum_{z\in Z_{t,u}}1\le1 ∑ z ∈ Z t , u 1 ≤ 1 if t = 0 t=0 t = 0 , and ∑ z ∈ Z t , u 1 ≤ ( 2 d ) t \sum_{z\in Z_{t,u}}1\le(2d)^{t} ∑ z ∈ Z t , u 1 ≤ ( 2 d ) t if t ≥ 1 t\ge1 t ≥ 1 and Z t , u Z_{t,u} Z t , u is nonempty, by Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §count applied to the image with a a a constantly 1 1 1 and M = 1 M=1 M = 1 . Since θ t ( 2 d ) t = ( 2 d θ ) t \theta^{t}(2d)^{t}=(2d\,\theta)^{t} θ t ( 2 d ) t = ( 2 d θ ) t , we get ∑ z ∈ Z t , u h ( z ) 2 ≤ ( 2 d θ ) t Y ( u ) 2 \sum_{z\in Z_{t,u}}h(z)^{2}\le(2d\,\theta)^{t}\,Y(u)^{2} ∑ z ∈ Z t , u h ( z ) 2 ≤ ( 2 d θ ) t Y ( u ) 2 in all cases, and summing over ( t , u ) (t,u) ( t , u ) with (C4),
Ω 2 ≤ ∑ u ∈ U Y ( u ) 2 ∑ t = 0 n ( 2 d θ ) t ≤ 2 ∑ u ∈ U Y ( u ) 2 = 2 ∑ p = 1 2 n ∑ u ∈ W d , p ∘ p c u ∣ x ( u ) ∣ 2 = 2 ∑ p = 1 2 n p X p ≤ 2 ∥ x ∥ d , 1 2 , \Omega^{2}\le\sum_{u\in\mathcal{U}}Y(u)^{2}\sum_{t=0}^{n}(2d\,\theta)^{t}\le2\sum_{u\in\mathcal{U}}Y(u)^{2}=2\sum_{p=1}^{2n}\ \sum_{u\in W^{\circ}_{d,p}}p\,c_{u}|x(u)|^{2}=2\sum_{p=1}^{2n}p\,X_{p}\le2\,\lVert x\rVert_{d,1}^{2}, Ω 2 ≤ u ∈ U ∑ Y ( u ) 2 t = 0 ∑ n ( 2 d θ ) t ≤ 2 u ∈ U ∑ Y ( u ) 2 = 2 p = 1 ∑ 2 n u ∈ W d , p ∘ ∑ p c u ∣ x ( u ) ∣ 2 = 2 p = 1 ∑ 2 n p X p ≤ 2 ∥ x ∥ d , 1 2 ,
using that U \mathcal{U} U is the disjoint union of the W d , p ∘ W^{\circ}_{d,p} W d , p ∘ , p ∈ [ 2 n ] p\in[2n] p ∈ [ 2 n ] , and (0.1). Hence Ω ≤ 2 ∥ x ∥ d , 1 \Omega\le\sqrt{2}\,\lVert x\rVert_{d,1} Ω ≤ 2 ∥ x ∥ d , 1 .
Step 9 (conclusion). By (5.1), enlarging the index set from P 1 \mathcal{P}_{1} P 1 to S × S \mathcal{S}\times\mathcal{S} S × S (the right sides of (5.1) being nonnegative), and (6d) with Ψ ( v , v ′ ) = 8 ϑ j ( v , v ′ ) h ( ρ ( v , v ′ ) ) \Psi(v,v')=8\,\vartheta^{\,j(v,v')}h(\rho(v,v')) Ψ ( v , v ′ ) = 8 ϑ j ( v , v ′ ) h ( ρ ( v , v ′ )) ,
S 1 ≤ 8 ∑ ( v , v ′ ) ∈ V × V ϑ j ( v , v ′ ) f ( v ) ℓ ( v ′ ) h ( ρ ( v , v ′ ) ) . S_{1}\le8\sum_{(v,v')\in V\times V}\vartheta^{\,j(v,v')}f(v)\,\ell(v')\,h\bigl(\rho(v,v')\bigr). S 1 ≤ 8 ( v , v ′ ) ∈ V × V ∑ ϑ j ( v , v ′ ) f ( v ) ℓ ( v ′ ) h ( ρ ( v , v ′ ) ) .
Every ( v , v ′ ) ∈ V × V (v,v')\in V\times V ( v , v ′ ) ∈ V × V has ∣ v ∣ = k ∈ [ n ] |v|=k\in[n] ∣ v ∣ = k ∈ [ n ] and j ( v , v ′ ) ∈ { 0 , … , min ( k , ∣ v ′ ∣ ) } ⊆ { 0 , … , k } j(v,v')\in\{0,\dots,\min(k,|v'|)\}\subseteq\{0,\dots,k\} j ( v , v ′ ) ∈ { 0 , … , min ( k , ∣ v ′ ∣ )} ⊆ { 0 , … , k } , so V × V V\times V V × V is the disjoint union of the sets Q k , j Q_{k,j} Q k , j , k ∈ [ n ] k\in[n] k ∈ [ n ] , j ∈ { 0 , … , k } j\in\{0,\dots,k\} j ∈ { 0 , … , k } , and ϑ j ( v , v ′ ) = ϑ j \vartheta^{\,j(v,v')}=\vartheta^{j} ϑ j ( v , v ′ ) = ϑ j on Q k , j Q_{k,j} Q k , j . Splitting over the nonempty Q k , j Q_{k,j} Q k , j , applying (7.1), adding the nonnegative bounds for the empty ones, and using (C4),
S 1 ≤ 8 Λ Ω ∑ k = 1 n Φ k ∑ j = 0 k ϑ j ≤ 16 Λ Ω ∑ k = 1 n Φ k . S_{1}\le8\,\Lambda\,\Omega\sum_{k=1}^{n}\Phi_{k}\sum_{j=0}^{k}\vartheta^{j}\le16\,\Lambda\,\Omega\sum_{k=1}^{n}\Phi_{k}. S 1 ≤ 8 Λ Ω k = 1 ∑ n Φ k j = 0 ∑ k ϑ j ≤ 16 Λ Ω k = 1 ∑ n Φ k .
By (6b), (C1) on [ n ] [n] [ n ] , Order Reversal under Reciprocals, and Summability of the Reciprocals of the Squares §partial (which gives ∑ k = 1 n 1 k 2 ≤ 2 − 1 n ≤ 2 \sum_{k=1}^{n}\frac{1}{k^{2}}\le2-\frac1n\le2 ∑ k = 1 n k 2 1 ≤ 2 − n 1 ≤ 2 ) and (0.1),
∑ k = 1 n Φ k ≤ 8 ∑ k = 1 n 1 k X k ≤ 8 ∑ k = 1 n 1 k 2 ∑ k = 1 n X k ≤ 8 2 ∥ x ∥ d = 4 ∥ x ∥ d . \sum_{k=1}^{n}\Phi_{k}\le\sqrt{8}\sum_{k=1}^{n}\frac1k\sqrt{X_{k}}\le\sqrt{8}\,\sqrt{\textstyle\sum_{k=1}^{n}\frac{1}{k^{2}}}\,\sqrt{\textstyle\sum_{k=1}^{n}X_{k}}\le\sqrt{8}\,\sqrt{2}\,\lVert x\rVert_{d}=4\,\lVert x\rVert_{d}. k = 1 ∑ n Φ k ≤ 8 k = 1 ∑ n k 1 X k ≤ 8 ∑ k = 1 n k 2 1 ∑ k = 1 n X k ≤ 8 2 ∥ x ∥ d = 4 ∥ x ∥ d .
With (6c) and Step 8, Λ Ω ≤ 8 2 ∥ x ∥ d , 1 2 = 4 ∥ x ∥ d , 1 2 \Lambda\,\Omega\le\sqrt{8}\,\sqrt{2}\,\lVert x\rVert_{d,1}^{2}=4\,\lVert x\rVert_{d,1}^{2} Λ Ω ≤ 8 2 ∥ x ∥ d , 1 2 = 4 ∥ x ∥ d , 1 2 , so S 1 ≤ 16 ⋅ 4 ⋅ 4 ∥ x ∥ d ∥ x ∥ d , 1 2 = 256 ∥ x ∥ d ∥ x ∥ d , 1 2 S_{1}\le16\cdot4\cdot4\,\lVert x\rVert_{d}\lVert x\rVert_{d,1}^{2}=256\,\lVert x\rVert_{d}\lVert x\rVert_{d,1}^{2} S 1 ≤ 16 ⋅ 4 ⋅ 4 ∥ x ∥ d ∥ x ∥ d , 1 2 = 256 ∥ x ∥ d ∥ x ∥ d , 1 2 . By (3.1) and (4.1),
∣ ∑ i ∈ [ d ] ( μ ( Ξ x , n i Ξ x , n i ) − ν ( Ξ x , n i Ξ x , n i ) ) ∣ ≤ 1 4 S ≤ 1 2 S 1 ≤ 128 ∥ x ∥ d ∥ x ∥ d , 1 2 . \Bigl|\sum_{i\in[d]}\Bigl(\mu\bigl(\Xi^{i}_{x,n}\Xi^{i}_{x,n}\bigr)-\nu\bigl(\Xi^{i}_{x,n}\Xi^{i}_{x,n}\bigr)\Bigr)\Bigr|\le\frac14\,S\le\frac12\,S_{1}\le128\,\lVert x\rVert_{d}\,\lVert x\rVert_{d,1}^{2}. i ∈ [ d ] ∑ ( μ ( Ξ x , n i Ξ x , n i ) − ν ( Ξ x , n i Ξ x , n i ) ) ≤ 4 1 S ≤ 2 1 S 1 ≤ 128 ∥ x ∥ d ∥ x ∥ d , 1 2 .
As μ , ν ∈ L d \mu,\nu\in\mathcal{L}_{d} μ , ν ∈ L d and n ∈ N n\in\mathbb{N} n ∈ N were arbitrary, K = 128 K=128 K = 128 is a trilinear constant.