Each result cited is universally quantified over the data in its own statement.
Notation and conventions. W d ∘ W^{\circ}_{d} W d ∘ , the sets W d , k ∘ W^{\circ}_{d,k} W d , k ∘ (k ∈ N k\in\mathbb{N} k ∈ N ), the weights c w c_{w} c w and the lengths ∣ w ∣ |w| ∣ w ∣ are as in The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge ; every c w c_{w} c w 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 . For n ∈ N n\in\mathbb{N} n ∈ N , W ≤ n ∘ W^{\circ}_{\le n} W ≤ n ∘ is the finite set of The Truncated Cyclic Gradient of a Gauge Vector §gradient , the union of the W d , k ∘ W^{\circ}_{d,k} W d , k ∘ with k ∈ [ n ] k\in[n] k ∈ [ n ] . For m , n ∈ N m,n\in\mathbb{N} m , n ∈ N with m < n m<n m < n let V m , n V_{m,n} V m , n be the set of w ∈ W d ∘ w\in W^{\circ}_{d} w ∈ W d ∘ whose length k k k satisfies m < k ≤ n m<k\le n m < k ≤ n . Words of different lengths are different, so W ≤ n ∘ W^{\circ}_{\le n} W ≤ n ∘ is the disjoint union of W ≤ m ∘ W^{\circ}_{\le m} W ≤ m ∘ and V m , n V_{m,n} V m , n , and V m , n V_{m,n} V m , n (respectively W ≤ n ∘ W^{\circ}_{\le n} W ≤ n ∘ ) is the disjoint union of the sets W d , k ∘ W^{\circ}_{d,k} W d , k ∘ with m < k ≤ n m<k\le n m < k ≤ n (respectively k ∈ [ n ] k\in[n] k ∈ [ n ] ), each nonempty and finite by Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §finite .
By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §inner-product , E d E_{d} E d is a real inner product space with norm ∥ ⋅ ∥ d \lVert\cdot\rVert_{d} ∥ ⋅ ∥ d whose zero vector is the zero map, so The Cauchy-Schwarz Inequality in a Real Inner Product Space , Elementary Identities in a Real Inner Product Space and The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity apply to it. Absolute values of real numbers are handled by Properties of the Absolute Value in an Ordered Field (claim 2: ∣ − x ∣ = ∣ x ∣ |-x|=|x| ∣ − x ∣ = ∣ x ∣ ; claim 3: − ∣ x ∣ ≤ x ≤ ∣ x ∣ -|x|\le x\le|x| − ∣ x ∣ ≤ x ≤ ∣ x ∣ ; claim 5: the triangle inequality; claim 6: ∣ x ∣ ≤ c |x|\le c ∣ x ∣ ≤ c if and only if − c ≤ x ≤ c -c\le x\le c − c ≤ x ≤ c ); by its claim 8 the absolute value of a real number is its modulus as a complex number. Order manipulations are those of Elementary Order Arithmetic in an Ordered Field and Elementary Arithmetic in an Ordered Field . Sums over nonempty finite index sets are those of Sum over a Finite Index Set ; they are additive and homogeneous by claims 3 and 4 of Properties of a Sum over a Finite Index Set , split over disjoint unions by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §disjoint-union , compare termwise by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §comparison , and satisfy ∣ ∑ x ∈ F f ( x ) ∣ ≤ ∑ x ∈ F ∣ f ( x ) ∣ \bigl|\sum_{x\in F}f(x)\bigr|\le\sum_{x\in F}|f(x)| ∑ x ∈ F f ( x ) ≤ ∑ x ∈ F ∣ f ( x ) ∣ by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §modulus . A sum over [ d ] [d] [ d ] of real terms each at most a real number c c c is at most d c d\,c d c : by the comparison just cited, claim 1 of Properties of a Sum over a Finite Index Set and the recursion in claim 1 of Properties of Finite Sums , it is at most c c c added d d d times, which is d c d\,c d c with d d d read in R \mathbb{R} R .
Sums and complex multiples of word polynomials are pointwise, and so are finite sums of word polynomials (Rotations and Cyclic Derivatives of Words in Unitary Letters ); we write − X = ( − 1 ) X -X=(-1)X − X = ( − 1 ) X and X − Y = X + ( − 1 ) Y X-Y=X+(-1)Y X − Y = X + ( − 1 ) Y . From the triangle inequality in Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §l1 , by induction on the number of terms along the recursive definition of finite sums,
∥ ∑ j ∈ J X j ∥ 1 ≤ ∑ j ∈ J ∥ X j ∥ 1 (N) \Bigl\lVert\sum_{j\in J}X_{j}\Bigr\rVert_{1}\le\sum_{j\in J}\lVert X_{j}\rVert_{1}\tag{N} j ∈ J ∑ X j 1 ≤ j ∈ J ∑ ∥ X j ∥ 1 ( N )
for word polynomials X j X_{j} X j indexed by a nonempty finite set J J J . Since Y − X = ( − 1 ) ( X − Y ) Y-X=(-1)(X-Y) Y − X = ( − 1 ) ( X − Y ) pointwise and ∥ ( − 1 ) Z ∥ 1 = ∥ Z ∥ 1 \lVert(-1)Z\rVert_{1}=\lVert Z\rVert_{1} ∥( − 1 ) Z ∥ 1 = ∥ Z ∥ 1 by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §l1 , ∥ Y − X ∥ 1 = ∥ X − Y ∥ 1 \lVert Y-X\rVert_{1}=\lVert X-Y\rVert_{1} ∥ Y − X ∥ 1 = ∥ X − Y ∥ 1 ; and X − X X-X X − X is the zero word polynomial, whose ℓ 1 \ell^{1} ℓ 1 norm is 0 0 0 by Word Polynomials in Unitary Letters: the Normed *-Algebra of Finitely Supported Word Functions §norm .
Convergence of real sequences is that of Limit of a Sequence of Real Numbers , which agrees with convergence in ( R , d R ) (\mathbb{R},d_{\mathbb{R}}) ( R , d R ) by The Real Numbers: Standing Notation and Background §sequences . We use the limit laws Arithmetic of Limits of Real Sequences , the order properties Order Properties of Limits of Real Sequences (claim 1: comparison; claim 3: domination by a null sequence; claim 4: absolute values), uniqueness of limits (claim 1 of Uniqueness of Limits and Boundedness of Convergent Real Sequences ), the fact that a constant sequence converges to its value, and the following consequence of Limit of a Sequence of Real Numbers : (T) if ( a n ) (a_{n}) ( a n ) converges to L L L and m ∈ N m\in\mathbb{N} m ∈ N , then n ↦ a n + m n\mapsto a_{n+m} n ↦ a n + m converges to L L L , since n + m ≥ n n+m\ge n n + m ≥ n ; hence claim 1 of Order Properties of Limits of Real Sequences applies to inequalities a n ≤ b n a_{n}\le b_{n} a n ≤ b n known only for n ≥ m n\ge m n ≥ m (apply it to the shifted sequences).
For λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d , p ∈ E d p\in E_{d} p ∈ E d and a ∈ P d a\in\mathcal{P}_{d} a ∈ P d write v a ( λ , p ) = − ⟨ p , b a ( λ ) ⟩ d − 1 2 ∥ a ∥ λ 2 v_{a}(\lambda,p)=-\langle p,b_{a}(\lambda)\rangle_{d}-\frac12\lVert a\rVert_{\lambda}^{2} v a ( λ , p ) = − ⟨ p , b a ( λ ) ⟩ d − 2 1 ∥ a ∥ λ 2 , so that H Q ( λ , p ) H_{Q}(\lambda,p) H Q ( λ , p ) is the supremum of { v a ( λ , p ) : a ∈ P d } \{v_{a}(\lambda,p):a\in\mathcal{P}_{d}\} { v a ( λ , p ) : a ∈ P d } by The Quadratic Control Hamiltonian on Unitary Laws §hamiltonian . By Upper Bound and Least Upper Bound : (S1) v a ( λ , p ) ≤ H Q ( λ , p ) v_{a}(\lambda,p)\le H_{Q}(\lambda,p) v a ( λ , p ) ≤ H Q ( λ , p ) for every a ∈ P d a\in\mathcal{P}_{d} a ∈ P d ; (S2) H Q ( λ , p ) ≤ c H_{Q}(\lambda,p)\le c H Q ( λ , p ) ≤ c for every real c c c with v a ( λ , p ) ≤ c v_{a}(\lambda,p)\le c v a ( λ , p ) ≤ c for all a ∈ P d a\in\mathcal{P}_{d} a ∈ P d .
Step 1 (length vectors). Let p ∈ E d p\in E_{d} p ∈ E d . For k ∈ N k\in\mathbb{N} k ∈ N put
β k ( p ) = ∑ w ∈ W d , k ∘ c w ∣ p ( w ) ∣ k , K k = ∑ w ∈ W d , k ∘ c w k 2 , s n = ∑ k = 1 n K k ( n ∈ N ) , \beta_{k}(p)=\sum_{w\in W^{\circ}_{d,k}}c_{w}\,|p(w)|\,k,\qquad K_{k}=\sum_{w\in W^{\circ}_{d,k}}c_{w}\,k^{2},\qquad s_{n}=\sum_{k=1}^{n}K_{k}\ (n\in\mathbb{N}), β k ( p ) = w ∈ W d , k ∘ ∑ c w ∣ p ( w ) ∣ k , K k = w ∈ W d , k ∘ ∑ c w k 2 , s n = k = 1 ∑ n K k ( n ∈ N ) ,
nonnegative reals by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §nonnegative . By Feedback Drifts: the Energy Bound in the Word Gauge, the Momentum Pairing and Two-Sided Tangency §constant , w ↦ c w ∣ w ∣ 2 w\mapsto c_{w}|w|^{2} w ↦ c w ∣ w ∣ 2 is summable; its block sums are the K k K_{k} K k , since ∣ w ∣ = k |w|=k ∣ w ∣ = k on W d , k ∘ W^{\circ}_{d,k} W d , k ∘ , and since ∣ ∅ ∣ = 0 |\varnothing|=0 ∣ ∅ ∣ = 0 its sum is ∑ k = 1 ∞ K k = κ d 2 \sum_{k=1}^{\infty}K_{k}=\kappa_{d}^{2} ∑ k = 1 ∞ K k = κ d 2 (The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §sums ). Thus s n → κ d 2 s_{n}\to\kappa_{d}^{2} s n → κ d 2 by Series of Real Numbers §convergent .
Define maps W d ∘ → C W^{\circ}_{d}\to\mathbb{C} W d ∘ → C with nonnegative real values by p ^ ( w ) = ∣ p ( w ) ∣ \widehat{p}(w)=|p(w)| p ( w ) = ∣ p ( w ) ∣ , ℓ ( w ) = ∣ w ∣ \ell(w)=|w| ℓ ( w ) = ∣ w ∣ and, for m ∈ N m\in\mathbb{N} m ∈ N , ℓ m ( w ) = ∣ w ∣ \ell_{m}(w)=|w| ℓ m ( w ) = ∣ w ∣ if ∣ w ∣ > m |w|>m ∣ w ∣ > m and ℓ m ( w ) = 0 \ell_{m}(w)=0 ℓ m ( w ) = 0 otherwise; so ℓ ( ∅ ) = ℓ m ( ∅ ) = 0 \ell(\varnothing)=\ell_{m}(\varnothing)=0 ℓ ( ∅ ) = ℓ m ( ∅ ) = 0 . By claim 8 of Properties of Complex Conjugation and Modulus the modulus of a nonnegative real is the number itself, so ∣ p ^ ( w ) ∣ 2 = ∣ p ( w ) ∣ 2 |\widehat{p}(w)|^{2}=|p(w)|^{2} ∣ p ( w ) ∣ 2 = ∣ p ( w ) ∣ 2 , ∣ ℓ ( w ) ∣ 2 = ∣ w ∣ 2 |\ell(w)|^{2}=|w|^{2} ∣ ℓ ( w ) ∣ 2 = ∣ w ∣ 2 and ∣ ℓ m ( w ) ∣ 2 |\ell_{m}(w)|^{2} ∣ ℓ m ( w ) ∣ 2 is ∣ w ∣ 2 |w|^{2} ∣ w ∣ 2 or 0 0 0 . Hence w ↦ c w ∣ p ^ ( w ) ∣ 2 w\mapsto c_{w}|\widehat{p}(w)|^{2} w ↦ c w ∣ p ( w ) ∣ 2 is the summable map w ↦ c w ∣ p ( w ) ∣ 2 w\mapsto c_{w}|p(w)|^{2} w ↦ c w ∣ p ( w ) ∣ 2 , so p ^ ∈ E d \widehat{p}\in E_{d} p ∈ E d and ∥ p ^ ∥ d = ∥ p ∥ d \lVert\widehat{p}\rVert_{d}=\lVert p\rVert_{d} ∥ p ∥ d = ∥ p ∥ d (The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §space , The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §norm ); w ↦ c w ∣ ℓ ( w ) ∣ 2 w\mapsto c_{w}|\ell(w)|^{2} w ↦ c w ∣ ℓ ( w ) ∣ 2 is w ↦ c w ∣ w ∣ 2 w\mapsto c_{w}|w|^{2} w ↦ c w ∣ w ∣ 2 , so ℓ ∈ E d \ell\in E_{d} ℓ ∈ E d with ∥ ℓ ∥ d 2 = κ d 2 \lVert\ell\rVert_{d}^{2}=\kappa_{d}^{2} ∥ ℓ ∥ d 2 = κ d 2 , whence ∥ ℓ ∥ d = κ d \lVert\ell\rVert_{d}=\kappa_{d} ∥ ℓ ∥ d = κ d by the uniqueness in Existence and Uniqueness of the Nonnegative Square Root ; and the block sums of w ↦ c w ∣ ℓ m ( w ) ∣ 2 w\mapsto c_{w}|\ell_{m}(w)|^{2} w ↦ c w ∣ ℓ m ( w ) ∣ 2 are Q k ( m ) = 0 Q^{(m)}_{k}=0 Q k ( m ) = 0 for k ≤ m k\le m k ≤ m and Q k ( m ) = K k Q^{(m)}_{k}=K_{k} Q k ( m ) = K k for k > m k>m k > m , so 0 ≤ Q k ( m ) ≤ K k 0\le Q^{(m)}_{k}\le K_{k} 0 ≤ Q k ( m ) ≤ K k and ℓ m ∈ E d \ell_{m}\in E_{d} ℓ m ∈ E d by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §comparison . Put τ m = ∥ ℓ m ∥ d ≥ 0 \tau_{m}=\lVert\ell_{m}\rVert_{d}\ge0 τ m = ∥ ℓ m ∥ d ≥ 0 ; it does not depend on p p p .
The numbers τ m \tau_{m} τ m tend to 0 0 0 . Let m ∈ N m\in\mathbb{N} m ∈ N . For n > m n>m n > m , splitting the finite sum at m m m (claim 1 of Properties of Finite Sums , by induction on n n n ) gives ∑ k = 1 n Q k ( m ) = s n − s m \sum_{k=1}^{n}Q^{(m)}_{k}=s_{n}-s_{m} ∑ k = 1 n Q k ( m ) = s n − s m , which converges to κ d 2 − s m \kappa_{d}^{2}-s_{m} κ d 2 − s m as n → ∞ n\to\infty n → ∞ by claim 3 of Arithmetic of Limits of Real Sequences and (T). As c ∅ ∣ ℓ m ( ∅ ) ∣ 2 = 0 c_{\varnothing}|\ell_{m}(\varnothing)|^{2}=0 c ∅ ∣ ℓ m ( ∅ ) ∣ 2 = 0 , uniqueness of limits gives τ m 2 = κ d 2 − s m \tau_{m}^{2}=\kappa_{d}^{2}-s_{m} τ m 2 = κ d 2 − s m , and so τ m 2 → 0 \tau_{m}^{2}\to0 τ m 2 → 0 as m → ∞ m\to\infty m → ∞ by the limit laws. Given a real ε > 0 \varepsilon>0 ε > 0 , ε 2 > 0 \varepsilon^{2}>0 ε 2 > 0 by claim 5 of Elementary Order Arithmetic in an Ordered Field ; choose N N N with τ m 2 = ∣ τ m 2 − 0 ∣ < ε 2 \tau_{m}^{2}=|\tau_{m}^{2}-0|<\varepsilon^{2} τ m 2 = ∣ τ m 2 − 0∣ < ε 2 for m ≥ N m\ge N m ≥ N ; then τ m < ε \tau_{m}<\varepsilon τ m < ε by claim 1 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field . Hence
τ m → 0 ( m → ∞ ) . (1) \tau_{m}\to0\quad(m\to\infty).\tag{1} τ m → 0 ( m → ∞ ) . ( 1 )
Pairings. Apply The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §expansion to p ^ \widehat{p} p and ℓ \ell ℓ . The values being nonnegative reals, p ^ ( w ) ‾ = p ^ ( w ) \overline{\widehat{p}(w)}=\widehat{p}(w) p ( w ) = p ( w ) by claim 1 of Properties of Complex Conjugation and Modulus , and the real part of the real number ∣ p ( w ) ∣ ∣ w ∣ |p(w)|\,|w| ∣ p ( w ) ∣ ∣ w ∣ is itself by Real and Imaginary Parts of a Complex Number ; so the terms P k P_{k} P k there are the β k ( p ) \beta_{k}(p) β k ( p ) and the term at ∅ \varnothing ∅ is 0 0 0 . Thus ∑ k = 1 ∞ β k ( p ) \sum_{k=1}^{\infty}\beta_{k}(p) ∑ k = 1 ∞ β k ( p ) converges with sum ⟨ p ^ , ℓ ⟩ d \langle\widehat{p},\ell\rangle_{d} ⟨ p , ℓ ⟩ d . In the same way, for m ∈ N m\in\mathbb{N} m ∈ N , ⟨ p ^ , ℓ m ⟩ d \langle\widehat{p},\ell_{m}\rangle_{d} ⟨ p , ℓ m ⟩ d is the sum of the convergent series with nonnegative terms 0 0 0 for k ≤ m k\le m k ≤ m and β k ( p ) \beta_{k}(p) β k ( p ) for k > m k>m k > m , whose n n n -th partial sum is ∑ k = m + 1 n β k ( p ) \sum_{k=m+1}^{n}\beta_{k}(p) ∑ k = m + 1 n β k ( p ) for n > m n>m n > m . By Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates , x ≤ ∣ x ∣ x\le|x| x ≤ ∣ x ∣ and The Cauchy-Schwarz Inequality in a Real Inner Product Space , for all m , n ∈ N m,n\in\mathbb{N} m , n ∈ N with m < n m<n m < n ,
∑ k = 1 n β k ( p ) ≤ ⟨ p ^ , ℓ ⟩ d ≤ ∥ p ∥ d κ d , ∑ k = m + 1 n β k ( p ) ≤ ⟨ p ^ , ℓ m ⟩ d ≤ ∥ p ∥ d τ m . (2) \sum_{k=1}^{n}\beta_{k}(p)\le\langle\widehat{p},\ell\rangle_{d}\le\lVert p\rVert_{d}\,\kappa_{d},\qquad
\sum_{k=m+1}^{n}\beta_{k}(p)\le\langle\widehat{p},\ell_{m}\rangle_{d}\le\lVert p\rVert_{d}\,\tau_{m}.\tag{2} k = 1 ∑ n β k ( p ) ≤ ⟨ p , ℓ ⟩ d ≤ ∥ p ∥ d κ d , k = m + 1 ∑ n β k ( p ) ≤ ⟨ p , ℓ m ⟩ d ≤ ∥ p ∥ d τ m . ( 2 )
Step 2 (ℓ 1 \ell^{1} ℓ 1 bounds for the truncated gradients). First, ∥ D w i ∥ 1 ≤ ∣ w ∣ \lVert D^{i}_{w}\rVert_{1}\le|w| ∥ D w i ∥ 1 ≤ ∣ w ∣ for i ∈ [ d ] i\in[d] i ∈ [ d ] and every word w w w of length k ∈ N k\in\mathbb{N} k ∈ N . Let M M M be the set of m ∈ [ k ] m\in[k] m ∈ [ k ] with g ( w m ) = i g(w_{m})=i g ( w m ) = i (Rotations and Cyclic Derivatives of Words in Unitary Letters §derivative ). If M M M is empty, D w i D^{i}_{w} D w i is the zero word polynomial, of ℓ 1 \ell^{1} ℓ 1 norm 0 0 0 . Otherwise, by (N) and Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §l1 ,
∥ D w i ∥ 1 ≤ ∑ m ∈ M ∣ i ε ( w m ) ∣ ∥ e r m ( w ) ∥ 1 = ∑ m ∈ M 1 ≤ ∑ m ∈ [ k ] 1 = k , \lVert D^{i}_{w}\rVert_{1}\le\sum_{m\in M}\bigl|\mathrm{i}\,\varepsilon(w_{m})\bigr|\,\lVert e_{r_{m}(w)}\rVert_{1}=\sum_{m\in M}1\le\sum_{m\in[k]}1=k, ∥ D w i ∥ 1 ≤ m ∈ M ∑ i ε ( w m ) ∥ e r m ( w ) ∥ 1 = m ∈ M ∑ 1 ≤ m ∈ [ k ] ∑ 1 = k ,
because ∣ i ε ( w m ) ∣ = ∣ i ∣ ∣ ε ( w m ) ∣ = 1 |\mathrm{i}\,\varepsilon(w_{m})|=|\mathrm{i}|\,|\varepsilon(w_{m})|=1 ∣ i ε ( w m ) ∣ = ∣ i ∣ ∣ ε ( w m ) ∣ = 1 (claim 4 of Properties of Complex Conjugation and Modulus , ε ( w m ) \varepsilon(w_{m}) ε ( w m ) being 1 1 1 or − 1 -1 − 1 ), ∥ e u ∥ 1 = ∣ 1 ∣ = 1 \lVert e_{u}\rVert_{1}=|1|=1 ∥ e u ∥ 1 = ∣1∣ = 1 computed with the support set { u } \{u\} { u } (Word Polynomials in Unitary Letters: the Normed *-Algebra of Finitely Supported Word Functions §norm ), and by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §monotone and claim 1 of Properties of a Sum over a Finite Index Set .
Let p ∈ E d p\in E_{d} p ∈ E d , i ∈ [ d ] i\in[d] i ∈ [ d ] and m , n ∈ N m,n\in\mathbb{N} m , n ∈ N with m < n m<n m < n . Evaluating at each word and splitting W ≤ n ∘ W^{\circ}_{\le n} W ≤ n ∘ into W ≤ m ∘ W^{\circ}_{\le m} W ≤ m ∘ and V m , n V_{m,n} V m , n , Z p , n i = Z p , m i + ∑ w ∈ V m , n c w p ( w ) ‾ D w i Z^{i}_{p,n}=Z^{i}_{p,m}+\sum_{w\in V_{m,n}}c_{w}\overline{p(w)}D^{i}_{w} Z p , n i = Z p , m i + ∑ w ∈ V m , n c w p ( w ) D w i (The Truncated Cyclic Gradient of a Gauge Vector §gradient ), so Z p , n i − Z p , m i Z^{i}_{p,n}-Z^{i}_{p,m} Z p , n i − Z p , m i is this last sum. Since ∣ c w p ( w ) ‾ ∣ = c w ∣ p ( w ) ∣ |c_{w}\overline{p(w)}|=c_{w}|p(w)| ∣ c w p ( w ) ∣ = c w ∣ p ( w ) ∣ (claims 3, 4 and 8 of Properties of Complex Conjugation and Modulus , c w c_{w} c w being positive), (N), Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §l1 and the bound on ∥ D w i ∥ 1 \lVert D^{i}_{w}\rVert_{1} ∥ D w i ∥ 1 give
∥ Z p , n i − Z p , m i ∥ 1 ≤ ∑ w ∈ V m , n c w ∣ p ( w ) ∣ ∣ w ∣ = ∑ k = m + 1 n β k ( p ) , \lVert Z^{i}_{p,n}-Z^{i}_{p,m}\rVert_{1}\le\sum_{w\in V_{m,n}}c_{w}|p(w)|\,|w|=\sum_{k=m+1}^{n}\beta_{k}(p), ∥ Z p , n i − Z p , m i ∥ 1 ≤ w ∈ V m , n ∑ c w ∣ p ( w ) ∣ ∣ w ∣ = k = m + 1 ∑ n β k ( p ) ,
the equality by splitting V m , n V_{m,n} V m , n into the sets W d , k ∘ W^{\circ}_{d,k} W d , k ∘ and claim 1 of Properties of a Sum over a Finite Index Set ; in the same way ∥ Z p , n i ∥ 1 ≤ ∑ k = 1 n β k ( p ) \lVert Z^{i}_{p,n}\rVert_{1}\le\sum_{k=1}^{n}\beta_{k}(p) ∥ Z p , n i ∥ 1 ≤ ∑ k = 1 n β k ( p ) for every n ∈ N n\in\mathbb{N} n ∈ N . By Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra (adjoints are additive and ( z X ) ∗ = z ‾ X ∗ (zX)^{*}=\overline{z}X^{*} ( z X ) ∗ = z X ∗ ), Ξ p , n i − Ξ p , m i = 1 2 ( Z p , n i − Z p , m i ) + 1 2 ( Z p , n i − Z p , m i ) ∗ \Xi^{i}_{p,n}-\Xi^{i}_{p,m}=\frac12(Z^{i}_{p,n}-Z^{i}_{p,m})+\frac12(Z^{i}_{p,n}-Z^{i}_{p,m})^{*} Ξ p , n i − Ξ p , m i = 2 1 ( Z p , n i − Z p , m i ) + 2 1 ( Z p , n i − Z p , m i ) ∗ , and by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §l1 (triangle inequality, multiples, ∥ X ∗ ∥ 1 = ∥ X ∥ 1 \lVert X^{*}\rVert_{1}=\lVert X\rVert_{1} ∥ X ∗ ∥ 1 = ∥ X ∥ 1 ) ∥ Ξ p , n i − Ξ p , m i ∥ 1 ≤ ∥ Z p , n i − Z p , m i ∥ 1 \lVert\Xi^{i}_{p,n}-\Xi^{i}_{p,m}\rVert_{1}\le\lVert Z^{i}_{p,n}-Z^{i}_{p,m}\rVert_{1} ∥ Ξ p , n i − Ξ p , m i ∥ 1 ≤ ∥ Z p , n i − Z p , m i ∥ 1 ; likewise ∥ Ξ p , n i ∥ 1 ≤ ∥ Z p , n i ∥ 1 \lVert\Xi^{i}_{p,n}\rVert_{1}\le\lVert Z^{i}_{p,n}\rVert_{1} ∥ Ξ p , n i ∥ 1 ≤ ∥ Z p , n i ∥ 1 . With (2), and with Ξ p , n i − Ξ p , n i = 0 \Xi^{i}_{p,n}-\Xi^{i}_{p,n}=0 Ξ p , n i − Ξ p , n i = 0 for the case m = n m=n m = n ,
∥ Ξ p , n i ∥ 1 ≤ κ d ∥ p ∥ d ( n ∈ N ) , ∥ Ξ p , n i − Ξ p , m i ∥ 1 ≤ ∥ p ∥ d τ m ( m ≤ n ) . (3) \lVert\Xi^{i}_{p,n}\rVert_{1}\le\kappa_{d}\lVert p\rVert_{d}\quad(n\in\mathbb{N}),\qquad
\lVert\Xi^{i}_{p,n}-\Xi^{i}_{p,m}\rVert_{1}\le\lVert p\rVert_{d}\,\tau_{m}\quad(m\le n).\tag{3} ∥ Ξ p , n i ∥ 1 ≤ κ d ∥ p ∥ d ( n ∈ N ) , ∥ Ξ p , n i − Ξ p , m i ∥ 1 ≤ ∥ p ∥ d τ m ( m ≤ n ) . ( 3 )
Now let p , q ∈ E d p,q\in E_{d} p , q ∈ E d and n ∈ N n\in\mathbb{N} n ∈ N . Pointwise ( p − q ) ( w ) = p ( w ) − q ( w ) (p-q)(w)=p(w)-q(w) ( p − q ) ( w ) = p ( w ) − q ( w ) , and p ( w ) − q ( w ) ‾ = p ( w ) ‾ − q ( w ) ‾ \overline{p(w)-q(w)}=\overline{p(w)}-\overline{q(w)} p ( w ) − q ( w ) = p ( w ) − q ( w ) by claim 1 of Properties of Complex Conjugation and Modulus ; so, evaluating at each word and using additivity and homogeneity of finite sums, Z p − q , n i = Z p , n i − Z q , n i Z^{i}_{p-q,n}=Z^{i}_{p,n}-Z^{i}_{q,n} Z p − q , n i = Z p , n i − Z q , n i , and then Ξ p − q , n i = Ξ p , n i − Ξ q , n i \Xi^{i}_{p-q,n}=\Xi^{i}_{p,n}-\Xi^{i}_{q,n} Ξ p − q , n i = Ξ p , n i − Ξ q , n i by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra . By (3) applied to p − q p-q p − q ,
∥ Ξ p , n i − Ξ q , n i ∥ 1 ≤ κ d ∥ p − q ∥ d . (4) \lVert\Xi^{i}_{p,n}-\Xi^{i}_{q,n}\rVert_{1}\le\kappa_{d}\lVert p-q\rVert_{d}.\tag{4} ∥ Ξ p , n i − Ξ q , n i ∥ 1 ≤ κ d ∥ p − q ∥ d . ( 4 )
Step 3 (evaluation estimate). Each Ξ p , n i \Xi^{i}_{p,n} Ξ p , n i is self-adjoint, as recalled in the statement. For word polynomials X , X ′ , Y , Y ′ X,X',Y,Y' X , X ′ , Y , Y ′ and λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d , Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra gives ( X − X ′ ) Y = X Y − X ′ Y (X-X')Y=XY-X'Y ( X − X ′ ) Y = X Y − X ′ Y and X ′ ( Y − Y ′ ) = X ′ Y − X ′ Y ′ X'(Y-Y')=X'Y-X'Y' X ′ ( Y − Y ′ ) = X ′ Y − X ′ Y ′ , so by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §linear λ ( X Y ) − λ ( X ′ Y ′ ) = λ ( ( X − X ′ ) Y ) + λ ( X ′ ( Y − Y ′ ) ) \lambda(XY)-\lambda(X'Y')=\lambda\bigl((X-X')Y\bigr)+\lambda\bigl(X'(Y-Y')\bigr) λ ( X Y ) − λ ( X ′ Y ′ ) = λ ( ( X − X ′ ) Y ) + λ ( X ′ ( Y − Y ′ ) ) ; by the triangle inequality for the modulus (claim 7 of Properties of Complex Conjugation and Modulus ) and Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §l1 (∣ λ ( Z ) ∣ ≤ ∥ Z ∥ 1 |\lambda(Z)|\le\lVert Z\rVert_{1} ∣ λ ( Z ) ∣ ≤ ∥ Z ∥ 1 and ∥ Z Z ′ ∥ 1 ≤ ∥ Z ∥ 1 ∥ Z ′ ∥ 1 \lVert ZZ'\rVert_{1}\le\lVert Z\rVert_{1}\lVert Z'\rVert_{1} ∥ Z Z ′ ∥ 1 ≤ ∥ Z ∥ 1 ∥ Z ′ ∥ 1 ),
∣ λ ( X Y ) − λ ( X ′ Y ′ ) ∣ ≤ ∥ X − X ′ ∥ 1 ∥ Y ∥ 1 + ∥ X ′ ∥ 1 ∥ Y − Y ′ ∥ 1 . (5) |\lambda(XY)-\lambda(X'Y')|\le\lVert X-X'\rVert_{1}\lVert Y\rVert_{1}+\lVert X'\rVert_{1}\lVert Y-Y'\rVert_{1}.\tag{5} ∣ λ ( X Y ) − λ ( X ′ Y ′ ) ∣ ≤ ∥ X − X ′ ∥ 1 ∥ Y ∥ 1 + ∥ X ′ ∥ 1 ∥ Y − Y ′ ∥ 1 . ( 5 )
Step 4 (truncated Hamiltonians). For λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d , p ∈ E d p\in E_{d} p ∈ E d and n ∈ N n\in\mathbb{N} n ∈ N put h n ( λ , p ) = 1 2 ∑ i ∈ [ d ] λ ( Ξ p , n i Ξ p , n i ) h_{n}(\lambda,p)=\frac12\sum_{i\in[d]}\lambda\bigl(\Xi^{i}_{p,n}\Xi^{i}_{p,n}\bigr) h n ( λ , p ) = 2 1 ∑ i ∈ [ d ] λ ( Ξ p , n i Ξ p , n i ) , a real number, and C p = d κ d ∥ p ∥ d 2 ≥ 0 C_{p}=d\,\kappa_{d}\lVert p\rVert_{d}^{2}\ge0 C p = d κ d ∥ p ∥ d 2 ≥ 0 . For m ≤ n m\le n m ≤ n , (5) with X = Y = Ξ p , n i X=Y=\Xi^{i}_{p,n} X = Y = Ξ p , n i , X ′ = Y ′ = Ξ p , m i X'=Y'=\Xi^{i}_{p,m} X ′ = Y ′ = Ξ p , m i and (3) give ∣ λ ( Ξ p , n i Ξ p , n i ) − λ ( Ξ p , m i Ξ p , m i ) ∣ ≤ 2 κ d ∥ p ∥ d 2 τ m |\lambda(\Xi^{i}_{p,n}\Xi^{i}_{p,n})-\lambda(\Xi^{i}_{p,m}\Xi^{i}_{p,m})|\le2\kappa_{d}\lVert p\rVert_{d}^{2}\tau_{m} ∣ λ ( Ξ p , n i Ξ p , n i ) − λ ( Ξ p , m i Ξ p , m i ) ∣ ≤ 2 κ d ∥ p ∥ d 2 τ m ; summing over [ d ] [d] [ d ] (modulus of a sum at most the sum of moduli) and halving,
∣ h n ( λ , p ) − h m ( λ , p ) ∣ ≤ C p τ m ( m ≤ n ) . (6) |h_{n}(\lambda,p)-h_{m}(\lambda,p)|\le C_{p}\,\tau_{m}\qquad(m\le n).\tag{6} ∣ h n ( λ , p ) − h m ( λ , p ) ∣ ≤ C p τ m ( m ≤ n ) . ( 6 )
For p , q ∈ E d p,q\in E_{d} p , q ∈ E d and n ∈ N n\in\mathbb{N} n ∈ N , (5) with X = Y = Ξ p , n i X=Y=\Xi^{i}_{p,n} X = Y = Ξ p , n i , X ′ = Y ′ = Ξ q , n i X'=Y'=\Xi^{i}_{q,n} X ′ = Y ′ = Ξ q , n i , together with (3) and (4), gives ∣ λ ( Ξ p , n i Ξ p , n i ) − λ ( Ξ q , n i Ξ q , n i ) ∣ ≤ κ d 2 ( ∥ p ∥ d + ∥ q ∥ d ) ∥ p − q ∥ d |\lambda(\Xi^{i}_{p,n}\Xi^{i}_{p,n})-\lambda(\Xi^{i}_{q,n}\Xi^{i}_{q,n})|\le\kappa_{d}^{2}\bigl(\lVert p\rVert_{d}+\lVert q\rVert_{d}\bigr)\lVert p-q\rVert_{d} ∣ λ ( Ξ p , n i Ξ p , n i ) − λ ( Ξ q , n i Ξ q , n i ) ∣ ≤ κ d 2 ( ∥ p ∥ d + ∥ q ∥ d ) ∥ p − q ∥ d , hence
∣ h n ( λ , p ) − h n ( λ , q ) ∣ ≤ d 2 κ d 2 ( ∥ p ∥ d + ∥ q ∥ d ) ∥ p − q ∥ d . (7) |h_{n}(\lambda,p)-h_{n}(\lambda,q)|\le\tfrac{d}{2}\,\kappa_{d}^{2}\bigl(\lVert p\rVert_{d}+\lVert q\rVert_{d}\bigr)\lVert p-q\rVert_{d}.\tag{7} ∣ h n ( λ , p ) − h n ( λ , q ) ∣ ≤ 2 d κ d 2 ( ∥ p ∥ d + ∥ q ∥ d ) ∥ p − q ∥ d . ( 7 )
Clause 1 (Legendre formula). Let λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d and p ∈ E d p\in E_{d} p ∈ E d , and write h n = h n ( λ , p ) h_{n}=h_{n}(\lambda,p) h n = h n ( λ , p ) .
Convergence. By (1) and claim 3 of Arithmetic of Limits of Real Sequences , C p τ m → 0 C_{p}\tau_{m}\to0 C p τ m → 0 . Given a real ε > 0 \varepsilon>0 ε > 0 , choose N N N with C p τ m = ∣ C p τ m − 0 ∣ < ε C_{p}\tau_{m}=|C_{p}\tau_{m}-0|<\varepsilon C p τ m = ∣ C p τ m − 0∣ < ε for m ≥ N m\ge N m ≥ N . For m , n ≥ N m,n\ge N m , n ≥ N , (6) applied with the smaller of the two indices in the role of m m m (and ∣ x − y ∣ = ∣ y − x ∣ |x-y|=|y-x| ∣ x − y ∣ = ∣ y − x ∣ ) gives ∣ h n − h m ∣ < ε |h_{n}-h_{m}|<\varepsilon ∣ h n − h m ∣ < ε . So ( h n ) (h_{n}) ( h n ) is a Cauchy sequence and converges, by Every Cauchy Sequence of Real Numbers Converges , to a real number h h h . Fix m ∈ N m\in\mathbb{N} m ∈ N ; as n → ∞ n\to\infty n → ∞ , ∣ h n − h m ∣ → ∣ h − h m ∣ |h_{n}-h_{m}|\to|h-h_{m}| ∣ h n − h m ∣ → ∣ h − h m ∣ by claim 3 of Arithmetic of Limits of Real Sequences and claim 4 of Order Properties of Limits of Real Sequences , so (6), (T) and claim 1 of Order Properties of Limits of Real Sequences give
∣ h − h m ∣ ≤ C p τ m ( m ∈ N ) . (8) |h-h_{m}|\le C_{p}\,\tau_{m}\qquad(m\in\mathbb{N}).\tag{8} ∣ h − h m ∣ ≤ C p τ m ( m ∈ N ) . ( 8 )
Upper bound. Let a ∈ P d a\in\mathcal{P}_{d} a ∈ P d and n ∈ N n\in\mathbb{N} n ∈ N , and put π n = ∑ i ∈ [ d ] λ ( a i Ξ p , n i ) \pi_{n}=\sum_{i\in[d]}\lambda(a^{i}\Xi^{i}_{p,n}) π n = ∑ i ∈ [ d ] λ ( a i Ξ p , n i ) , real by Feedback Drifts: the Energy Bound in the Word Gauge, the Momentum Pairing and Two-Sided Tangency §pairing . For i ∈ [ d ] i\in[d] i ∈ [ d ] , A i = a i + Ξ p , n i A_{i}=a^{i}+\Xi^{i}_{p,n} A i = a i + Ξ p , n i is self-adjoint by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra , so 0 ≤ λ ( A i A i ) 0\le\lambda(A_{i}A_{i}) 0 ≤ λ ( A i A i ) by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §positive . By Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra and Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §linear , λ ( A i A i ) = λ ( a i a i ) + λ ( a i Ξ p , n i ) + λ ( Ξ p , n i a i ) + λ ( Ξ p , n i Ξ p , n i ) \lambda(A_{i}A_{i})=\lambda(a^{i}a^{i})+\lambda(a^{i}\Xi^{i}_{p,n})+\lambda(\Xi^{i}_{p,n}a^{i})+\lambda(\Xi^{i}_{p,n}\Xi^{i}_{p,n}) λ ( A i A i ) = λ ( a i a i ) + λ ( a i Ξ p , n i ) + λ ( Ξ p , n i a i ) + λ ( Ξ p , n i Ξ p , n i ) , and λ ( Ξ p , n i a i ) = λ ( a i Ξ p , n i ) \lambda(\Xi^{i}_{p,n}a^{i})=\lambda(a^{i}\Xi^{i}_{p,n}) λ ( Ξ p , n i a i ) = λ ( a i Ξ p , n i ) by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §tracial . Summing over [ d ] [d] [ d ] and using Polynomial Controls on Unitary Laws and Their Energy §energy and Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §nonnegative ,
0 ≤ ∑ i ∈ [ d ] λ ( A i A i ) = ∥ a ∥ λ 2 + 2 π n + 2 h n , so − π n − 1 2 ∥ a ∥ λ 2 ≤ h n . 0\le\sum_{i\in[d]}\lambda(A_{i}A_{i})=\lVert a\rVert_{\lambda}^{2}+2\pi_{n}+2h_{n},\qquad\text{so}\qquad-\pi_{n}-\tfrac12\lVert a\rVert_{\lambda}^{2}\le h_{n}. 0 ≤ i ∈ [ d ] ∑ λ ( A i A i ) = ∥ a ∥ λ 2 + 2 π n + 2 h n , so − π n − 2 1 ∥ a ∥ λ 2 ≤ h n .
By Feedback Drifts: the Energy Bound in the Word Gauge, the Momentum Pairing and Two-Sided Tangency §pairing , π n → ⟨ p , b a ( λ ) ⟩ d \pi_{n}\to\langle p,b_{a}(\lambda)\rangle_{d} π n → ⟨ p , b a ( λ ) ⟩ d , so the left side converges to v a ( λ , p ) v_{a}(\lambda,p) v a ( λ , p ) by the limit laws, and claim 1 of Order Properties of Limits of Real Sequences gives v a ( λ , p ) ≤ h v_{a}(\lambda,p)\le h v a ( λ , p ) ≤ h . As a a a was arbitrary, (S2) gives H Q ( λ , p ) ≤ h H_{Q}(\lambda,p)\le h H Q ( λ , p ) ≤ h .
Lower bound. For m , n ∈ N m,n\in\mathbb{N} m , n ∈ N put g m , n = ∑ i ∈ [ d ] λ ( Ξ p , m i Ξ p , n i ) g_{m,n}=\sum_{i\in[d]}\lambda(\Xi^{i}_{p,m}\Xi^{i}_{p,n}) g m , n = ∑ i ∈ [ d ] λ ( Ξ p , m i Ξ p , n i ) , real by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §positive . Fix m ∈ N m\in\mathbb{N} m ∈ N and let a m = ( − Ξ p , m 1 , … , − Ξ p , m d ) a_{m}=(-\Xi^{1}_{p,m},\dots,-\Xi^{d}_{p,m}) a m = ( − Ξ p , m 1 , … , − Ξ p , m d ) . By Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra , ( − X ) ∗ = ( − 1 ) ‾ X ∗ = − X (-X)^{*}=\overline{(-1)}X^{*}=-X ( − X ) ∗ = ( − 1 ) X ∗ = − X for self-adjoint X X X , so a m ∈ P d a_{m}\in\mathcal{P}_{d} a m ∈ P d (Polynomial Controls on Unitary Laws and Their Energy §controls ); ( − X ) ( − Y ) = ( − 1 ) ( − 1 ) ( X Y ) = X Y (-X)(-Y)=(-1)(-1)(XY)=XY ( − X ) ( − Y ) = ( − 1 ) ( − 1 ) ( X Y ) = X Y and ( − X ) Y = ( − 1 ) ( X Y ) (-X)Y=(-1)(XY) ( − X ) Y = ( − 1 ) ( X Y ) , so with Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §linear , ∥ a m ∥ λ 2 = 2 h m \lVert a_{m}\rVert_{\lambda}^{2}=2h_{m} ∥ a m ∥ λ 2 = 2 h m and ∑ i ∈ [ d ] λ ( ( − Ξ p , m i ) Ξ p , n i ) = − g m , n \sum_{i\in[d]}\lambda\bigl((-\Xi^{i}_{p,m})\Xi^{i}_{p,n}\bigr)=-g_{m,n} ∑ i ∈ [ d ] λ ( ( − Ξ p , m i ) Ξ p , n i ) = − g m , n . By Feedback Drifts: the Energy Bound in the Word Gauge, the Momentum Pairing and Two-Sided Tangency §pairing , − g m , n → ⟨ p , b a m ( λ ) ⟩ d -g_{m,n}\to\langle p,b_{a_{m}}(\lambda)\rangle_{d} − g m , n → ⟨ p , b a m ( λ ) ⟩ d as n → ∞ n\to\infty n → ∞ , so by claim 3 of Arithmetic of Limits of Real Sequences g m , n → g m g_{m,n}\to g_{m} g m , n → g m , where g m = − ⟨ p , b a m ( λ ) ⟩ d g_{m}=-\langle p,b_{a_{m}}(\lambda)\rangle_{d} g m = − ⟨ p , b a m ( λ ) ⟩ d , and v a m ( λ , p ) = g m − h m v_{a_{m}}(\lambda,p)=g_{m}-h_{m} v a m ( λ , p ) = g m − h m . For n ≥ m n\ge m n ≥ m , (5) with X = Ξ p , m i X=\Xi^{i}_{p,m} X = Ξ p , m i , X ′ = Y = Y ′ = Ξ p , n i X'=Y=Y'=\Xi^{i}_{p,n} X ′ = Y = Y ′ = Ξ p , n i and (3) give ∣ λ ( Ξ p , m i Ξ p , n i ) − λ ( Ξ p , n i Ξ p , n i ) ∣ ≤ ∥ p ∥ d τ m ⋅ κ d ∥ p ∥ d |\lambda(\Xi^{i}_{p,m}\Xi^{i}_{p,n})-\lambda(\Xi^{i}_{p,n}\Xi^{i}_{p,n})|\le\lVert p\rVert_{d}\tau_{m}\cdot\kappa_{d}\lVert p\rVert_{d} ∣ λ ( Ξ p , m i Ξ p , n i ) − λ ( Ξ p , n i Ξ p , n i ) ∣ ≤ ∥ p ∥ d τ m ⋅ κ d ∥ p ∥ d , since ∥ Ξ p , m i − Ξ p , n i ∥ 1 = ∥ Ξ p , n i − Ξ p , m i ∥ 1 \lVert\Xi^{i}_{p,m}-\Xi^{i}_{p,n}\rVert_{1}=\lVert\Xi^{i}_{p,n}-\Xi^{i}_{p,m}\rVert_{1} ∥ Ξ p , m i − Ξ p , n i ∥ 1 = ∥ Ξ p , n i − Ξ p , m i ∥ 1 and the second term of (5) vanishes, Y − Y ′ Y-Y' Y − Y ′ being the zero word polynomial; summing over [ d ] [d] [ d ] , ∣ g m , n − 2 h n ∣ ≤ C p τ m |g_{m,n}-2h_{n}|\le C_{p}\tau_{m} ∣ g m , n − 2 h n ∣ ≤ C p τ m for n ≥ m n\ge m n ≥ m . As n → ∞ n\to\infty n → ∞ , ∣ g m , n − 2 h n ∣ → ∣ g m − 2 h ∣ |g_{m,n}-2h_{n}|\to|g_{m}-2h| ∣ g m , n − 2 h n ∣ → ∣ g m − 2 h ∣ (limit laws, claim 4 of Order Properties of Limits of Real Sequences ), so ∣ g m − 2 h ∣ ≤ C p τ m |g_{m}-2h|\le C_{p}\tau_{m} ∣ g m − 2 h ∣ ≤ C p τ m by (T) and claim 1 of Order Properties of Limits of Real Sequences . With (8) and − ∣ x ∣ ≤ x -|x|\le x − ∣ x ∣ ≤ x ,
v a m ( λ , p ) = h + ( g m − 2 h ) + ( h − h m ) ≥ h − 2 C p τ m , v_{a_{m}}(\lambda,p)=h+(g_{m}-2h)+(h-h_{m})\ge h-2C_{p}\tau_{m}, v a m ( λ , p ) = h + ( g m − 2 h ) + ( h − h m ) ≥ h − 2 C p τ m ,
so h − 2 C p τ m ≤ H Q ( λ , p ) h-2C_{p}\tau_{m}\le H_{Q}(\lambda,p) h − 2 C p τ m ≤ H Q ( λ , p ) by (S1), for every m ∈ N m\in\mathbb{N} m ∈ N . By (1) and the limit laws h − 2 C p τ m → h h-2C_{p}\tau_{m}\to h h − 2 C p τ m → h , and comparison with the constant sequence H Q ( λ , p ) H_{Q}(\lambda,p) H Q ( λ , p ) (claim 1 of Order Properties of Limits of Real Sequences ) gives h ≤ H Q ( λ , p ) h\le H_{Q}(\lambda,p) h ≤ H Q ( λ , p ) .
Hence h = H Q ( λ , p ) h=H_{Q}(\lambda,p) h = H Q ( λ , p ) , which is clause 1, and (8) becomes
∣ H Q ( λ , p ) − h m ( λ , p ) ∣ ≤ C p τ m ( λ ∈ L d , p ∈ E d , m ∈ N ) . (9) |H_{Q}(\lambda,p)-h_{m}(\lambda,p)|\le C_{p}\,\tau_{m}\qquad(\lambda\in\mathcal{L}_{d},\ p\in E_{d},\ m\in\mathbb{N}).\tag{9} ∣ H Q ( λ , p ) − h m ( λ , p ) ∣ ≤ C p τ m ( λ ∈ L d , p ∈ E d , m ∈ N ) . ( 9 )
Clause 2 (bounds). Let λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d and p ∈ E d p\in E_{d} p ∈ E d . The upper bound holds because 1 2 κ d 2 ∥ p ∥ d 2 \frac12\kappa_{d}^{2}\lVert p\rVert_{d}^{2} 2 1 κ d 2 ∥ p ∥ d 2 is an upper bound of { v a ( λ , p ) : a ∈ P d } \{v_{a}(\lambda,p):a\in\mathcal{P}_{d}\} { v a ( λ , p ) : a ∈ P d } by The Quadratic Control Hamiltonian on Unitary Laws §hamiltonian , and (S2). For the lower bound let 0 0 0 also denote the tuple of zero word polynomials, which lies in P d \mathcal{P}_{d} P d by Polynomial Controls on Unitary Laws and Their Energy §controls . The zero word polynomial is 0 X 0X 0 X for any word polynomial X X X , so 0 Y = ( 0 X ) Y = 0 ( X Y ) 0\,Y=(0X)Y=0(XY) 0 Y = ( 0 X ) Y = 0 ( X Y ) is the zero word polynomial by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra , and its evaluation is 0 0 0 by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §linear . Hence every term of b 0 ( λ ) ( w ) = ∑ i ∈ [ d ] λ ( 0 D w i ) b_{0}(\lambda)(w)=\sum_{i\in[d]}\lambda(0\,D^{i}_{w}) b 0 ( λ ) ( w ) = ∑ i ∈ [ d ] λ ( 0 D w i ) (Feedback Drifts of Polynomial Controls on Unitary Laws §drift ) is 0 0 0 , and b 0 ( λ ) b_{0}(\lambda) b 0 ( λ ) is the zero map by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing , the zero vector of E d E_{d} E d ; so ⟨ p , b 0 ( λ ) ⟩ d = 0 \langle p,b_{0}(\lambda)\rangle_{d}=0 ⟨ p , b 0 ( λ ) ⟩ d = 0 by Elementary Identities in a Real Inner Product Space §zero . Likewise ∥ 0 ∥ λ 2 = ∑ i ∈ [ d ] λ ( 0 ⋅ 0 ) = 0 \lVert 0\rVert_{\lambda}^{2}=\sum_{i\in[d]}\lambda(0\cdot0)=0 ∥ 0 ∥ λ 2 = ∑ i ∈ [ d ] λ ( 0 ⋅ 0 ) = 0 . Thus v 0 ( λ , p ) = 0 v_{0}(\lambda,p)=0 v 0 ( λ , p ) = 0 , and 0 ≤ H Q ( λ , p ) 0\le H_{Q}(\lambda,p) 0 ≤ H Q ( λ , p ) by (S1).
Clause 3 (differences in the law). Let μ , ν ∈ L d \mu,\nu\in\mathcal{L}_{d} μ , ν ∈ L d and p ∈ E d p\in E_{d} p ∈ E d ; the word polynomials Ξ p , n i \Xi^{i}_{p,n} Ξ p , n i do not depend on the law. By claims 3 and 4 of Properties of a Sum over a Finite Index Set , − 1 2 ∑ i ∈ [ d ] ( μ ( Ξ p , n i Ξ p , n i ) − ν ( Ξ p , n i Ξ p , n i ) ) = h n ( ν , p ) − h n ( μ , p ) -\frac12\sum_{i\in[d]}\bigl(\mu(\Xi^{i}_{p,n}\Xi^{i}_{p,n})-\nu(\Xi^{i}_{p,n}\Xi^{i}_{p,n})\bigr)=h_{n}(\nu,p)-h_{n}(\mu,p) − 2 1 ∑ i ∈ [ d ] ( μ ( Ξ p , n i Ξ p , n i ) − ν ( Ξ p , n i Ξ p , n i ) ) = h n ( ν , p ) − h n ( μ , p ) . By clause 1 and claim 3 of Arithmetic of Limits of Real Sequences , h n ( ν , p ) − h n ( μ , p ) → H Q ( ν , p ) − H Q ( μ , p ) h_{n}(\nu,p)-h_{n}(\mu,p)\to H_{Q}(\nu,p)-H_{Q}(\mu,p) h n ( ν , p ) − h n ( μ , p ) → H Q ( ν , p ) − H Q ( μ , p ) , and ∑ i ∈ [ d ] ( μ ( ⋅ ) − ν ( ⋅ ) ) = 2 h n ( μ , p ) − 2 h n ( ν , p ) \sum_{i\in[d]}(\mu(\cdot)-\nu(\cdot))=2h_{n}(\mu,p)-2h_{n}(\nu,p) ∑ i ∈ [ d ] ( μ ( ⋅ ) − ν ( ⋅ )) = 2 h n ( μ , p ) − 2 h n ( ν , p ) converges to 2 H Q ( μ , p ) − 2 H Q ( ν , p ) 2H_{Q}(\mu,p)-2H_{Q}(\nu,p) 2 H Q ( μ , p ) − 2 H Q ( ν , p ) ; multiplying this limit by − 1 2 -\frac12 − 2 1 gives the displayed identity, the limit existing.
Clause 4 (admissibility). We check the two properties of Admissible Hamiltonians on Unitary Laws .
Dependence on the momentum. Let λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d and p , q ∈ E d p,q\in E_{d} p , q ∈ E d . By clause 1 and the limit laws, h n ( λ , p ) − h n ( λ , q ) → H Q ( λ , p ) − H Q ( λ , q ) h_{n}(\lambda,p)-h_{n}(\lambda,q)\to H_{Q}(\lambda,p)-H_{Q}(\lambda,q) h n ( λ , p ) − h n ( λ , q ) → H Q ( λ , p ) − H Q ( λ , q ) , so, by claim 4 and claim 1 of Order Properties of Limits of Real Sequences applied to (7) against a constant sequence,
∣ H Q ( λ , p ) − H Q ( λ , q ) ∣ ≤ d 2 κ d 2 ( ∥ p ∥ d + ∥ q ∥ d ) ∥ p − q ∥ d . (10) |H_{Q}(\lambda,p)-H_{Q}(\lambda,q)|\le\tfrac{d}{2}\,\kappa_{d}^{2}\bigl(\lVert p\rVert_{d}+\lVert q\rVert_{d}\bigr)\lVert p-q\rVert_{d}.\tag{10} ∣ H Q ( λ , p ) − H Q ( λ , q ) ∣ ≤ 2 d κ d 2 ( ∥ p ∥ d + ∥ q ∥ d ) ∥ p − q ∥ d . ( 10 )
Dependence on the law. Let μ , ν ∈ L d \mu,\nu\in\mathcal{L}_{d} μ , ν ∈ L d , p ∈ E d p\in E_{d} p ∈ E d and m ∈ N m\in\mathbb{N} m ∈ N . By the triangle inequality and (9) for μ \mu μ and for ν \nu ν ,
∣ H Q ( μ , p ) − H Q ( ν , p ) ∣ ≤ 2 C p τ m + ∣ h m ( μ , p ) − h m ( ν , p ) ∣ . (11) |H_{Q}(\mu,p)-H_{Q}(\nu,p)|\le2C_{p}\tau_{m}+|h_{m}(\mu,p)-h_{m}(\nu,p)|.\tag{11} ∣ H Q ( μ , p ) − H Q ( ν , p ) ∣ ≤ 2 C p τ m + ∣ h m ( μ , p ) − h m ( ν , p ) ∣. ( 11 )
Moreover, if ( λ j ) j ∈ N (\lambda_{j})_{j\in\mathbb{N}} ( λ j ) j ∈ N converges to λ 0 \lambda_{0} λ 0 in ( L d , d L ) (\mathcal{L}_{d},d_{\mathcal{L}}) ( L d , d L ) , then h m ( λ j , p ) → h m ( λ 0 , p ) h_{m}(\lambda_{j},p)\to h_{m}(\lambda_{0},p) h m ( λ j , p ) → h m ( λ 0 , p ) . Indeed, for i ∈ [ d ] i\in[d] i ∈ [ d ] let Y i = Ξ p , m i Ξ p , m i Y^{i}=\Xi^{i}_{p,m}\Xi^{i}_{p,m} Y i = Ξ p , m i Ξ p , m i , with a support set F i F_{i} F i . By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §convergence , for each w ∈ W 2 d w\in W_{2d} w ∈ W 2 d , λ j ( w ) → λ 0 ( w ) \lambda_{j}(w)\to\lambda_{0}(w) λ j ( w ) → λ 0 ( w ) in ( C , d C ) (\mathbb{C},d_{\mathbb{C}}) ( C , d C ) , that is (Convergent Sequence in a Metric Space ), the real sequence ∣ λ j ( w ) − λ 0 ( w ) ∣ |\lambda_{j}(w)-\lambda_{0}(w)| ∣ λ j ( w ) − λ 0 ( w ) ∣ converges to 0 0 0 . By Evaluation of Word Polynomials by a Unitary Law §evaluation , additivity and homogeneity of finite sums, the modulus bound for finite sums and claim 4 of Properties of Complex Conjugation and Modulus ,
∣ λ j ( Y i ) − λ 0 ( Y i ) ∣ = ∣ ∑ w ∈ F i Y i ( w ) ( λ j ( w ) − λ 0 ( w ) ) ∣ ≤ ∑ w ∈ F i ∣ Y i ( w ) ∣ ∣ λ j ( w ) − λ 0 ( w ) ∣ , |\lambda_{j}(Y^{i})-\lambda_{0}(Y^{i})|=\Bigl|\sum_{w\in F_{i}}Y^{i}(w)\bigl(\lambda_{j}(w)-\lambda_{0}(w)\bigr)\Bigr|\le\sum_{w\in F_{i}}|Y^{i}(w)|\,|\lambda_{j}(w)-\lambda_{0}(w)|, ∣ λ j ( Y i ) − λ 0 ( Y i ) ∣ = w ∈ F i ∑ Y i ( w ) ( λ j ( w ) − λ 0 ( w ) ) ≤ w ∈ F i ∑ ∣ Y i ( w ) ∣ ∣ λ j ( w ) − λ 0 ( w ) ∣ ,
and the right side converges to 0 0 0 as j → ∞ j\to\infty j → ∞ by claims 1 and 3 of Arithmetic of Limits of Real Sequences and induction on the number of terms. By claim 3 of Order Properties of Limits of Real Sequences , λ j ( Y i ) → λ 0 ( Y i ) \lambda_{j}(Y^{i})\to\lambda_{0}(Y^{i}) λ j ( Y i ) → λ 0 ( Y i ) (these are real numbers), and the limit laws give h m ( λ j , p ) → h m ( λ 0 , p ) h_{m}(\lambda_{j},p)\to h_{m}(\lambda_{0},p) h m ( λ j , p ) → h m ( λ 0 , p ) .
Continuity. Write d × d_{\times} d × for the product metric on L d × E d \mathcal{L}_{d}\times E_{d} L d × E d of Admissible Hamiltonians on Unitary Laws §continuous , built from the metric spaces ( L d , d L ) (\mathcal{L}_{d},d_{\mathcal{L}}) ( L d , d L ) (The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §metric ) and ( E d , ( x , y ) ↦ ∥ x − y ∥ d ) (E_{d},(x,y)\mapsto\lVert x-y\rVert_{d}) ( E d , ( x , y ) ↦ ∥ x − y ∥ d ) (The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity §metric ); it is a metric by claim 1 of The Product Metric is a Metric . Let ( λ 0 , p 0 ) ∈ L d × E d (\lambda_{0},p_{0})\in\mathcal{L}_{d}\times E_{d} ( λ 0 , p 0 ) ∈ L d × E d and let ( ( λ j , p j ) ) j ∈ N \bigl((\lambda_{j},p_{j})\bigr)_{j\in\mathbb{N}} ( ( λ j , p j ) ) j ∈ N converge to it in ( L d × E d , d × ) (\mathcal{L}_{d}\times E_{d},d_{\times}) ( L d × E d , d × ) . By claim 1 of Coordinatewise Convergence, Sequential Compactness and Density in a Product Metric Space , λ j → λ 0 \lambda_{j}\to\lambda_{0} λ j → λ 0 in ( L d , d L ) (\mathcal{L}_{d},d_{\mathcal{L}}) ( L d , d L ) and p j → p 0 p_{j}\to p_{0} p j → p 0 in E d E_{d} E d , that is, ∥ p j − p 0 ∥ d → 0 \lVert p_{j}-p_{0}\rVert_{d}\to0 ∥ p j − p 0 ∥ d → 0 ; and ∥ p j ∥ d → ∥ p 0 ∥ d \lVert p_{j}\rVert_{d}\to\lVert p_{0}\rVert_{d} ∥ p j ∥ d → ∥ p 0 ∥ d by The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity §continuity . So by the limit laws e j = d 2 κ d 2 ( ∥ p j ∥ d + ∥ p 0 ∥ d ) ∥ p j − p 0 ∥ d → 0 e_{j}=\frac{d}{2}\kappa_{d}^{2}\bigl(\lVert p_{j}\rVert_{d}+\lVert p_{0}\rVert_{d}\bigr)\lVert p_{j}-p_{0}\rVert_{d}\to0 e j = 2 d κ d 2 ( ∥ p j ∥ d + ∥ p 0 ∥ d ) ∥ p j − p 0 ∥ d → 0 . Let a real ε > 0 \varepsilon>0 ε > 0 be given. By (1) and the limit laws choose m ∈ N m\in\mathbb{N} m ∈ N with 2 C p 0 τ m < ε / 3 2C_{p_{0}}\tau_{m}<\varepsilon/3 2 C p 0 τ m < ε /3 ; then, by the preceding paragraph with p = p 0 p=p_{0} p = p 0 , choose J ∈ N J\in\mathbb{N} J ∈ N such that for j ≥ J j\ge J j ≥ J both e j < ε / 3 e_{j}<\varepsilon/3 e j < ε /3 and ∣ h m ( λ j , p 0 ) − h m ( λ 0 , p 0 ) ∣ < ε / 3 |h_{m}(\lambda_{j},p_{0})-h_{m}(\lambda_{0},p_{0})|<\varepsilon/3 ∣ h m ( λ j , p 0 ) − h m ( λ 0 , p 0 ) ∣ < ε /3 (take the larger of the two indices provided). For j ≥ J j\ge J j ≥ J , by the triangle inequality, (10) for λ j \lambda_{j} λ j , p j p_{j} p j , p 0 p_{0} p 0 , and (11) for λ j \lambda_{j} λ j , λ 0 \lambda_{0} λ 0 , p 0 p_{0} p 0 ,
∣ H Q ( λ j , p j ) − H Q ( λ 0 , p 0 ) ∣ ≤ ∣ H Q ( λ j , p j ) − H Q ( λ j , p 0 ) ∣ + ∣ H Q ( λ j , p 0 ) − H Q ( λ 0 , p 0 ) ∣ < ε 3 + ε 3 + ε 3 = ε . |H_{Q}(\lambda_{j},p_{j})-H_{Q}(\lambda_{0},p_{0})|\le|H_{Q}(\lambda_{j},p_{j})-H_{Q}(\lambda_{j},p_{0})|+|H_{Q}(\lambda_{j},p_{0})-H_{Q}(\lambda_{0},p_{0})|<\tfrac{\varepsilon}{3}+\tfrac{\varepsilon}{3}+\tfrac{\varepsilon}{3}=\varepsilon . ∣ H Q ( λ j , p j ) − H Q ( λ 0 , p 0 ) ∣ ≤ ∣ H Q ( λ j , p j ) − H Q ( λ j , p 0 ) ∣ + ∣ H Q ( λ j , p 0 ) − H Q ( λ 0 , p 0 ) ∣ < 3 ε + 3 ε + 3 ε = ε .
Thus H Q ( λ j , p j ) → H Q ( λ 0 , p 0 ) H_{Q}(\lambda_{j},p_{j})\to H_{Q}(\lambda_{0},p_{0}) H Q ( λ j , p j ) → H Q ( λ 0 , p 0 ) in ( R , d R ) (\mathbb{R},d_{\mathbb{R}}) ( R , d R ) . By Continuity Between Metric Spaces is Equivalent to Sequential Continuity §on-subset , applied with A A A the whole space L d × E d \mathcal{L}_{d}\times E_{d} L d × E d (on which the restricted metric is d × d_{\times} d × itself), H Q H_{Q} H Q is continuous on L d × E d \mathcal{L}_{d}\times E_{d} L d × E d ; this is property Admissible Hamiltonians on Unitary Laws §continuous .
Tangential invariance. Let λ ∈ L d \lambda\in\mathcal{L}_{d} λ ∈ L d and p , p ′ ∈ E d p,p'\in E_{d} p , p ′ ∈ E d with ⟨ p − p ′ , v ⟩ d = 0 \langle p-p',v\rangle_{d}=0 ⟨ p − p ′ , v ⟩ d = 0 for every v ∈ T λ ± v\in T^{\pm}_{\lambda} v ∈ T λ ± . Let a ∈ P d a\in\mathcal{P}_{d} a ∈ P d . Then b a ( λ ) ∈ T λ ± b_{a}(\lambda)\in T^{\pm}_{\lambda} b a ( λ ) ∈ T λ ± by Feedback Drifts: the Energy Bound in the Word Gauge, the Momentum Pairing and Two-Sided Tangency §tangent , so by Elementary Identities in a Real Inner Product Space §bilinear , ⟨ p , b a ( λ ) ⟩ d − ⟨ p ′ , b a ( λ ) ⟩ d = ⟨ p − p ′ , b a ( λ ) ⟩ d = 0 \langle p,b_{a}(\lambda)\rangle_{d}-\langle p',b_{a}(\lambda)\rangle_{d}=\langle p-p',b_{a}(\lambda)\rangle_{d}=0 ⟨ p , b a ( λ ) ⟩ d − ⟨ p ′ , b a ( λ ) ⟩ d = ⟨ p − p ′ , b a ( λ ) ⟩ d = 0 , and hence v a ( λ , p ) = v a ( λ , p ′ ) v_{a}(\lambda,p)=v_{a}(\lambda,p') v a ( λ , p ) = v a ( λ , p ′ ) . As a a a was arbitrary, the two sets whose suprema define H Q ( λ , p ) H_{Q}(\lambda,p) H Q ( λ , p ) and H Q ( λ , p ′ ) H_{Q}(\lambda,p') H Q ( λ , p ′ ) in The Quadratic Control Hamiltonian on Unitary Laws §hamiltonian coincide, and so do their suprema: H Q ( λ , p ) = H Q ( λ , p ′ ) H_{Q}(\lambda,p)=H_{Q}(\lambda,p') H Q ( λ , p ) = H Q ( λ , p ′ ) . This is property Admissible Hamiltonians on Unitary Laws §tangential , and H Q H_{Q} H Q is an admissible Hamiltonian.