TheoremBase

The drift is a sum over letters of values of the law on controls times rotated words, each bounded by the energy via Cauchy-Schwarz, which gives the gauge bound; the pairing follows by expanding the gauge inner product and symmetrising; and tangency follows by perturbing a realising unitary tuple by Cayley transforms of the operator controls and expanding the resulting laws word by word.

Proof

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

Word polynomials, the ewe_{w}, products, adjoints and evaluations are those of Word Polynomials in Unitary Letters: the Normed *-Algebra of Finitely Supported Word Functions and Evaluation of Word Polynomials by a Unitary Law. Finite sums are those of Sum over a Finite Index Set; reindexing, additivity and homogeneity refer to claims 2, 3 and 4 of Properties of a Sum over a Finite Index Set. By The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §weights, θd=16144 d\theta_{d}=\frac{1}{6144\,d} and 0<cw≤θd k0<c_{w}\le\theta_{d}^{\,k} for w∈Wd,k∘w\in W^{\circ}_{d,k}, while c∅=1c_{\varnothing}=1. Every set Wd,k∘W^{\circ}_{d,k} is nonempty and finite (Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §finite), and ∅∈Wd∘\varnothing\in W^{\circ}_{d} (Reduced and Cyclically Reduced Words in Unitary Letters §cyclically-reduced). For a word ww of length kk and m∈[k]m\in[k] we write jm=g(wm)j_{m}=g(w_{m}), εm=ε(wm)\varepsilon_{m}=\varepsilon(w_{m}) and rm=rm(w)r_{m}=r_{m}(w) (Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters, Rotations and Cyclic Derivatives of Words in Unitary Letters §rotations).

Step 0 (preliminaries). Let λ∈Ld\lambda\in\mathcal{L}_{d}.

(Z) For the zero word polynomial 00 and every word polynomial XX, X0=0X0=0 and λ(0)=0\lambda(0)=0. Every value of X0X0 is 00 or a sum of terms X(u)⋅0=0X(u)\cdot0=0, and λ(0)\lambda(0) is a sum of terms 0⋅λ(w)0\cdot\lambda(w); both vanish by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing.

(S) Let QQ be a nonempty finite set, XX and YqY_{q} (q∈Qq\in Q) word polynomials and zq∈Cz_{q}\in\mathbb{C}. Then λ(X∑q∈QzqYq)=∑q∈Qzqλ(XYq)\lambda\bigl(X\sum_{q\in Q}z_{q}Y_{q}\bigr)=\sum_{q\in Q}z_{q}\lambda(XY_{q}), the inner sum being pointwise (Rotations and Cyclic Derivatives of Words in Unitary Letters). This follows from 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 by induction on the number of elements of QQ.

(R) For every self-adjoint XX and every r∈W2dr\in W_{2d}, ∣λ(Xer)∣2≤λ(XX)|\lambda(Xe_{r})|^{2}\le\lambda(XX). Since X=X∗X=X^{*}, Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §cauchy-schwarz gives ∣λ(Xer)∣2=∣λ(X∗er)∣2≤λ(X∗X) λ(er∗er)|\lambda(Xe_{r})|^{2}=|\lambda(X^{*}e_{r})|^{2}\le\lambda(X^{*}X)\,\lambda(e_{r}^{*}e_{r}). By Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra, er∗er=er∗er=er∗re_{r}^{*}e_{r}=e_{r^{*}}e_{r}=e_{r^{*}r}, and λ(er∗r)=λ(r∗r)\lambda(e_{r^{*}r})=\lambda(r^{*}r) by Evaluation of Word Polynomials by a Unitary Law §evaluation. Put y=r∗y=r^{*}. Then y∗=ry^{*}=r, as recorded in Word Polynomials in Unitary Letters: the Normed *-Algebra of Finitely Supported Word Functions §adjoint, and r∗r=y ∅ y∗r^{*}r=y\,\varnothing\,y^{*} by Basic Properties of Words: Associativity, Reversal, Finitely Many Factorisations, and Countability §monoid. So λ(r∗r)=λ(∅)=1\lambda(r^{*}r)=\lambda(\varnothing)=1 by Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §strip-invariance and Laws of d-Tuples of Unitaries §normalised. Finally, λ(X∗X)=λ(XX)\lambda(X^{*}X)=\lambda(XX), and this number is real and nonnegative by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §positive.

(B1) Let a∈Pda\in\mathcal{P}_{d}. Then ba(λ)(∅)=0b_{a}(\lambda)(\varnothing)=0. For every w∈Wd∘w\in W^{\circ}_{d} of length k∈Nk\in\mathbb{N},

ba(λ)(w)=∑m∈[k]i εm λ(ajmerm).b_{a}(\lambda)(w)=\sum_{m\in[k]}\mathrm{i}\,\varepsilon_{m}\,\lambda\bigl(a^{j_{m}}e_{r_{m}}\bigr).

First, D∅i=0D^{i}_{\varnothing}=0 (Rotations and Cyclic Derivatives of Words in Unitary Letters §derivative), so every term of ba(λ)(∅)b_{a}(\lambda)(\varnothing) is 00 by (Z), and the sum is 00 by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing. For ww of length kk and i∈[d]i\in[d], put Mi={m∈[k]:jm=i}M_{i}=\{m\in[k]:j_{m}=i\}. If Mi=∅M_{i}=\emptyset, then Dwi=0D^{i}_{w}=0 and λ(aiDwi)=0\lambda(a^{i}D^{i}_{w})=0 by (Z). Otherwise (S) gives λ(aiDwi)=∑m∈Miiεmλ(aierm)\lambda(a^{i}D^{i}_{w})=\sum_{m\in M_{i}}\mathrm{i}\varepsilon_{m}\lambda(a^{i}e_{r_{m}}). The set I={i∈[d]:Mi≠∅}I=\{i\in[d]:M_{i}\neq\emptyset\} contains j1j_{1}. By Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing, ba(λ)(w)=∑i∈Iλ(aiDwi)b_{a}(\lambda)(w)=\sum_{i\in I}\lambda(a^{i}D^{i}_{w}) (Feedback Drifts of Polynomial Controls on Unitary Laws §drift). By Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs, this is the sum of iεmλ(ajmerm)\mathrm{i}\varepsilon_{m}\lambda(a^{j_{m}}e_{r_{m}}) over the pairs (i,m)(i,m) with i∈Ii\in I and m∈Mim\in M_{i}. The map (i,m)↦m(i,m)\mapsto m is a bijection of the set of these pairs onto [k][k], with inverse m↦(jm,m)m\mapsto(j_{m},m), and reindexing gives (B1).

(K) For k∈Nk\in\mathbb{N}, k≤2kk\le2^{k} and k2≤4kk^{2}\le4^{k} (natural powers in R\mathbb{R}). By induction, 1≤2=211\le2=2^{1}, and if k≤2kk\le2^{k} then k+1≤2k+2k=2k+1k+1\le2^{k}+2^{k}=2^{k+1}, since 1=1k≤2k1=1^{k}\le2^{k} (claims 1, 2 and 5 of Properties of Natural Number Powers in a Field). Then k2≤2k2k=4kk^{2}\le2^{k}2^{k}=4^{k} by claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field and claim 3 of Properties of Natural Number Powers in a Field.

Step 1 (clause 1, the constant). Let k∈Nk\in\mathbb{N} and w∈Wd,k∘w\in W^{\circ}_{d,k}. Then ∣w∣|w| is the image of kk (Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §length), so by (K) and claim 5 of Elementary Arithmetic in an Ordered Field, 0≤cw∣w∣2≤θd k4k0\le c_{w}|w|^{2}\le\theta_{d}^{\,k}4^{k}. By Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §count and claim 3 of Properties of Natural Number Powers in a Field, the kk-th block sum CkC_{k} of w↦cw∣w∣2w\mapsto c_{w}|w|^{2} satisfies

0≤Ck≤θd k4k(2d)k=(8d θd)k=(1768)k;0\le C_{k}\le\theta_{d}^{\,k}4^{k}(2d)^{k}=(8d\,\theta_{d})^{k}=\bigl(\tfrac{1}{768}\bigr)^{k};

here 0≤Ck0\le C_{k} holds by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §nonnegative. The geometric series of ratio 1768\frac1{768} converges by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric. By Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §comparison, ∑k=1∞Ck\sum_{k=1}^{\infty}C_{k} converges. The map w↦cw∣w∣2w\mapsto c_{w}|w|^{2} takes nonnegative values, its value at ∅\varnothing being 1⋅0=01\cdot0=0. It is therefore summable (The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §sums), with sum κd2=0+∑k=1∞Ck\kappa_{d}^{2}=0+\sum_{k=1}^{\infty}C_{k}. This sum is nonnegative by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates, as in The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §norm, and κd\kappa_{d} is its nonnegative square root (Existence and Uniqueness of the Nonnegative Square Root).

Step 2 (clause 2, the energy bound). Let a∈Pda\in\mathcal{P}_{d}, λ∈Ld\lambda\in\mathcal{L}_{d}, and write b=ba(λ)b=b_{a}(\lambda). Each aja^{j} is self-adjoint (Polynomial Controls on Unitary Laws and Their Energy §controls), so αj=λ(ajaj)\alpha_{j}=\lambda(a^{j}a^{j}) is a nonnegative real for every j∈[d]j\in[d] (Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §positive). By Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §monotone with E={j}E=\{j\} and by Polynomial Controls on Unitary Laws and Their Energy §energy, αj≤∑i∈[d]αi=∥a∥λ2\alpha_{j}\le\sum_{i\in[d]}\alpha_{i}=\lVert a\rVert_{\lambda}^{2}. Hence (R) gives ∣λ(ajer)∣2≤∥a∥λ2|\lambda(a^{j}e_{r})|^{2}\le\lVert a\rVert_{\lambda}^{2}, that is, ∣λ(ajer)∣≤∥a∥λ|\lambda(a^{j}e_{r})|\le\lVert a\rVert_{\lambda} for all j∈[d]j\in[d] and r∈W2dr\in W_{2d}, by claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field.

For w=∅w=\varnothing we have ∣b(∅)∣=0=∣∅∣ ∥a∥λ|b(\varnothing)|=0=|\varnothing|\,\lVert a\rVert_{\lambda} by (B1). Now let w∈Wd∘w\in W^{\circ}_{d} have length k∈Nk\in\mathbb{N}. Each εm∈{1,−1}\varepsilon_{m}\in\{1,-1\}, so ∣iεm∣=1|\mathrm{i}\varepsilon_{m}|=1 (claims 3, 4 and 8 of Properties of Complex Conjugation and Modulus). Then (B1), Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §modulus, claim 4 of Properties of Complex Conjugation and Modulus and Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §comparison give

∣b(w)∣≤∑m∈[k]∣λ(ajmerm)∣≤∑m∈[k]∥a∥λ=k ∥a∥λ=∣w∣ ∥a∥λ.|b(w)|\le\sum_{m\in[k]}\bigl|\lambda\bigl(a^{j_{m}}e_{r_{m}}\bigr)\bigr|\le\sum_{m\in[k]}\lVert a\rVert_{\lambda}=k\,\lVert a\rVert_{\lambda}=|w|\,\lVert a\rVert_{\lambda}.

Here the sum of a constant over [k][k] is kk times the constant, by claim 1 of Properties of a Sum over a Finite Index Set and induction on kk.

Consequently, for every w∈Wd∘w\in W^{\circ}_{d} we have cw∣b(w)∣2≤∥a∥λ2 cw∣w∣2c_{w}|b(w)|^{2}\le\lVert a\rVert_{\lambda}^{2}\,c_{w}|w|^{2}, by claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field and claim 5 of Elementary Arithmetic in an Ordered Field. So the kk-th block sum BkB_{k} of w↦cw∣b(w)∣2w\mapsto c_{w}|b(w)|^{2} satisfies 0≤Bk≤∥a∥λ2Ck0\le B_{k}\le\lVert a\rVert_{\lambda}^{2}C_{k}, by Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §nonnegative, Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §comparison and homogeneity. The series ∑k∥a∥λ2Ck\sum_{k}\lVert a\rVert_{\lambda}^{2}C_{k} converges to ∥a∥λ2κd2\lVert a\rVert_{\lambda}^{2}\kappa_{d}^{2} by Step 1 and Elementary Properties of Series of Real Numbers §linearity. By Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §comparison, ∑kBk\sum_{k}B_{k} converges, with sum at most ∥a∥λ2κd2\lVert a\rVert_{\lambda}^{2}\kappa_{d}^{2}. Therefore b∈Edb\in E_{d} (The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §space), and ∥b∥d2=c∅∣b(∅)∣2+∑k=1∞Bk≤(κd∥a∥λ)2\lVert b\rVert_{d}^{2}=c_{\varnothing}|b(\varnothing)|^{2}+\sum_{k=1}^{\infty}B_{k}\le(\kappa_{d}\lVert a\rVert_{\lambda})^{2} (The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §norm). Both ∥b∥d\lVert b\rVert_{d} and κd∥a∥λ\kappa_{d}\lVert a\rVert_{\lambda} are nonnegative, so ∥b∥d≤κd∥a∥λ\lVert b\rVert_{d}\le\kappa_{d}\lVert a\rVert_{\lambda} by claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field.

Step 3 (clause 3, the momentum pairing). Let a∈Pda\in\mathcal{P}_{d}, λ∈Ld\lambda\in\mathcal{L}_{d} and p∈Edp\in E_{d}, and write b=ba(λ)b=b_{a}(\lambda), which lies in EdE_{d} by Step 2.

(a) Reality. Let n∈Nn\in\mathbb{N} and i∈[d]i\in[d], and write A=aiA=a^{i} and Z=Zp,niZ=Z^{i}_{p,n}. 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 Ξp,ni)=12λ(AZ)+12λ(AZ∗)\lambda(A\,\Xi^{i}_{p,n})=\frac12\lambda(AZ)+\frac12\lambda(AZ^{*}) (The Truncated Cyclic Gradient of a Gauge Vector §gradient). We have λ(AZ∗)=λ(Z∗A)\lambda(AZ^{*})=\lambda(Z^{*}A) by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §tracial. Since A∗=AA^{*}=A, Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra gives Z∗A=Z∗A∗=(AZ)∗Z^{*}A=Z^{*}A^{*}=(AZ)^{*}, so λ(Z∗A)=λ(AZ)‾\lambda(Z^{*}A)=\overline{\lambda(AZ)} by Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §adjoint. By claim 2 of Properties of Complex Conjugation and Modulus,

λ(ai Ξp,ni)=12(λ(AZ)+λ(AZ)‾)=Re⁡λ(aiZp,ni),\lambda\bigl(a^{i}\,\Xi^{i}_{p,n}\bigr)=\tfrac12\bigl(\lambda(AZ)+\overline{\lambda(AZ)}\bigr)=\operatorname{Re}\lambda\bigl(a^{i}Z^{i}_{p,n}\bigr),

which is a real number.

(b) Partial sums. By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §expansion with x=px=p and y=by=b, the series ∑kPk\sum_{k}P_{k} with Pk=∑w∈Wd,k∘cwRe⁡(p(w)‾ b(w))P_{k}=\sum_{w\in W^{\circ}_{d,k}}c_{w}\operatorname{Re}\bigl(\overline{p(w)}\,b(w)\bigr) converges, and ⟨p,b⟩d=Re⁡(p(∅)‾ b(∅))+∑k=1∞Pk=∑k=1∞Pk\langle p,b\rangle_{d}=\operatorname{Re}\bigl(\overline{p(\varnothing)}\,b(\varnothing)\bigr)+\sum_{k=1}^{\infty}P_{k}=\sum_{k=1}^{\infty}P_{k}, because b(∅)=0b(\varnothing)=0 by (B1). Let n∈Nn\in\mathbb{N}. Each word has only one length (Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §words), so (k,w)↦w(k,w)\mapsto w is a bijection from the pairs with k∈[n]k\in[n] and w∈Wd,k∘w\in W^{\circ}_{d,k} onto W≤n∘W^{\circ}_{\le n}. Claim 1 of Properties of a Sum over a Finite Index Set, Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs and reindexing therefore give sn:=∑k=1nPk=∑w∈W≤n∘cwRe⁡(p(w)‾ b(w))s_{n}:=\sum_{k=1}^{n}P_{k}=\sum_{w\in W^{\circ}_{\le n}}c_{w}\operatorname{Re}\bigl(\overline{p(w)}\,b(w)\bigr).

The real part commutes with finite sums and with real factors. Indeed, 2Re⁡ζ=ζ+ζ‾2\operatorname{Re}\zeta=\zeta+\overline{\zeta} (claim 2 of Properties of Complex Conjugation and Modulus); combine this with Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §conjugate, additivity, homogeneity and claim 1 of the same lemma. Hence sn=Re⁡∑w∈W≤n∘cwp(w)‾ b(w)s_{n}=\operatorname{Re}\sum_{w\in W^{\circ}_{\le n}}c_{w}\overline{p(w)}\,b(w). By Feedback Drifts of Polynomial Controls on Unitary Laws §drift and homogeneity, the summand equals ∑i∈[d]cwp(w)‾λ(aiDwi)\sum_{i\in[d]}c_{w}\overline{p(w)}\lambda(a^{i}D^{i}_{w}). We exchange the two finite sums: Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs applies twice, with reindexing along (w,i)↦(i,w)(w,i)\mapsto(i,w). By (S) with Q=W≤n∘Q=W^{\circ}_{\le n} and The Truncated Cyclic Gradient of a Gauge Vector §gradient, we then obtain

sn=Re⁡∑i∈[d] ∑w∈W≤n∘cwp(w)‾ λ(aiDwi)=Re⁡∑i∈[d]λ(aiZp,ni)=∑i∈[d]λ(ai Ξp,ni),s_{n}=\operatorname{Re}\sum_{i\in[d]}\ \sum_{w\in W^{\circ}_{\le n}}c_{w}\overline{p(w)}\,\lambda\bigl(a^{i}D^{i}_{w}\bigr)=\operatorname{Re}\sum_{i\in[d]}\lambda\bigl(a^{i}Z^{i}_{p,n}\bigr)=\sum_{i\in[d]}\lambda\bigl(a^{i}\,\Xi^{i}_{p,n}\bigr),

the last equality by (a) and the commutation of Re⁡\operatorname{Re} with finite sums. The partial sums sns_{n} of the convergent series ∑kPk\sum_{k}P_{k} converge to its sum ⟨p,b⟩d\langle p,b\rangle_{d} in (R,dR)(\mathbb{R},d_{\mathbb{R}}) (Series of Real Numbers §convergent), which proves clause 3.

Step 4 (clause 4: operator preliminaries). Fix a∈Pda\in\mathcal{P}_{d} and λ∈Ld\lambda\in\mathcal{L}_{d}, and write b=ba(λ)b=b_{a}(\lambda). Operators are handled in the notation of Complex Hilbert Spaces and Bounded Linear Maps: Standing Notation. For a tracial W*-probability space and a dd-tuple VV of unitaries in it, we use the notation V(l)V^{(l)}, V(w)V^{(w)} and λV\lambda_{V} fixed in the statement of Unitary Laws Are the Laws of Unitary Tuples, and Are Realised Together with a Free Semicircular Family. By Unitary Laws Are the Laws of Unitary Tuples, and Are Realised Together with a Free Semicircular Family §realisation there are a tracial W*-probability space (H,M,Ω)(H,M,\Omega) with trace τ=τM\tau=\tau_{M} and a dd-tuple UU of unitary operators on HH belonging to MM with λU=λ\lambda_{U}=\lambda. The self-adjoint operators SjS_{j} provided there are not used. The triple (H,M,Ω)(H,M,\Omega) is a cyclic tracial operator algebra (Tracial W*-Probability Spaces §space).

(O1) For T∈MT\in M, ∣τ(T)∣=∣⟨Ω,TΩ⟩∣≤∥Ω∥ ∥TΩ∥≤∥T∥op|\tau(T)|=|\langle\Omega,T\Omega\rangle|\le\lVert\Omega\rVert\,\lVert T\Omega\rVert\le\lVert T\rVert_{\mathrm{op}}. The equality is Tracial W*-Probability Spaces §trace with Cyclic Tracial Operator Algebras and Their Traces §trace. For the first inequality, Cauchy-Schwarz Inequality in a Complex Inner Product Space gives ∣⟨Ω,TΩ⟩∣2≤∥Ω∥2∥TΩ∥2|\langle\Omega,T\Omega\rangle|^{2}\le\lVert\Omega\rVert^{2}\lVert T\Omega\rVert^{2}; both sides of the claimed inequality are nonnegative, so claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field applies. For the second, ∥T∥op\lVert T\rVert_{\mathrm{op}} is a bound for TT by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §least-bound, so ∥TΩ∥≤∥T∥op∥Ω∥\lVert T\Omega\rVert\le\lVert T\rVert_{\mathrm{op}}\lVert\Omega\rVert by Bounded Linear Maps between Complex Normed Spaces and the Operator Norm §bounded, and ∥Ω∥=1\lVert\Omega\rVert=1 by Cyclic Tracial Operator Algebras and Their Traces §cyclic.

(O2) τ\tau is linear, τ(I)=1\tau(I)=1, τ(ST)=τ(TS)\tau(ST)=\tau(TS) and τ(S∗)=τ(S)‾\tau(S^{*})=\overline{\tau(S)} for S,T∈MS,T\in M (The Trace, the Conjugation and the Right Action of a Cyclic Tracial Operator Algebra §trace). MM contains II and is closed under sums, complex multiples, products and adjoints (Cyclic Tracial Operator Algebras and Their Traces §star-algebra).

(O3) For every letter ll, ∥U(l)∥op≤1\lVert U^{(l)}\rVert_{\mathrm{op}}\le1. Indeed, by claim 3 of Properties of Unitary Operators each UjU_{j} is a bounded linear operator on HH with operator norm at most 11; by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §least-bound, Uj∈L(H)U_{j}\in\mathcal{L}(H) and that operator norm is ∥Uj∥op\lVert U_{j}\rVert_{\mathrm{op}}, so ∥Uj∥op≤1\lVert U_{j}\rVert_{\mathrm{op}}\le1. Moreover ∥Uj∗∥op=∥Uj∥op\lVert U_{j}^{*}\rVert_{\mathrm{op}}=\lVert U_{j}\rVert_{\mathrm{op}} (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint). Hence every product of such operators, including the empty product II, has operator norm at most 11. This follows by submultiplicativity and ∥I∥op≤1\lVert I\rVert_{\mathrm{op}}\le1 (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations) and induction on the number of factors.

(O4) U(u)U(v)=U(uv)U^{(u)}U^{(v)}=U^{(uv)} for u,v∈W2du,v\in W_{2d}, since U(∅)=IU^{(\varnothing)}=I and the letters of uvuv are those of uu followed by those of vv (Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §concatenation). By induction, a product of consecutive factors U(wm)U^{(w_{m})} of a word ww is U(y)U^{(y)} for the corresponding subword yy.

(O5) A map T∈L(H)T\in\mathcal{L}(H) with T∗T=TT∗=IT^{*}T=TT^{*}=I is unitary. Every vv equals T(T∗v)T(T^{*}v). Also ⟨Tu,Tv⟩=⟨u,T∗Tv⟩=⟨u,v⟩\langle Tu,Tv\rangle=\langle u,T^{*}Tv\rangle=\langle u,v\rangle, because TT is the adjoint of T∗T^{*} (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus, Adjoint of a Linear Map between Complex Inner Product Spaces §adjoint). Conversely, each UjU_{j} is a bounded linear operator on HH by claim 3 of Properties of Unitary Operators, so Uj∈L(H)U_{j}\in\mathcal{L}(H) by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §least-bound. It has an adjoint Uj∗∈L(H)U_{j}^{*}\in\mathcal{L}(H) by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint, and Uj∗Uj=UjUj∗=IU_{j}^{*}U_{j}=U_{j}U_{j}^{*}=I. Indeed, ⟨u,Uj∗Ujv⟩=⟨Uju,Ujv⟩=⟨u,v⟩\langle u,U_{j}^{*}U_{j}v\rangle=\langle U_{j}u,U_{j}v\rangle=\langle u,v\rangle with u=Uj∗Ujv−vu=U_{j}^{*}U_{j}v-v gives ∥u∥2=⟨u,u⟩=⟨u,Uj∗Ujv⟩−⟨u,v⟩=0\lVert u\rVert^{2}=\langle u,u\rangle=\langle u,U_{j}^{*}U_{j}v\rangle-\langle u,v\rangle=0, so u=0u=0 by definiteness (claim 4 of Elementary Properties of a Complex Inner Product), that is, Uj∗Ujv=vU_{j}^{*}U_{j}v=v. Writing v=Ujuv=U_{j}u by surjectivity, UjUj∗v=Uj(Uj∗Uju)=vU_{j}U_{j}^{*}v=U_{j}(U_{j}^{*}U_{j}u)=v.

(O6) (U(w))∗=U(w∗)(U^{(w)})^{*}=U^{(w^{*})} for every w∈W2dw\in W_{2d}. First let ll be a letter. If ε(l)=1\varepsilon(l)=1, then l−1=d+g(l)l^{-1}=d+g(l) has generator g(l)g(l) and sign −1-1 (Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters), so U(l−1)=Ug(l)∗=(U(l))∗U^{(l^{-1})}=U_{g(l)}^{*}=(U^{(l)})^{*}. If ε(l)=−1\varepsilon(l)=-1, then l−1=g(l)l^{-1}=g(l) has sign 11, and (U(l))∗=(Ug(l)∗)∗=Ug(l)=U(l−1)(U^{(l)})^{*}=(U_{g(l)}^{*})^{*}=U_{g(l)}=U^{(l^{-1})}, because TT is the adjoint of T∗T^{*} (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus). For w=∅w=\varnothing we have I∗=II^{*}=I. For ww of length kk, the same clause and induction on kk give (U(w))∗=(U(wk))∗⋯(U(w1))∗=U(wk−1)⋯U(w1−1)(U^{(w)})^{*}=(U^{(w_{k})})^{*}\cdots(U^{(w_{1})})^{*}=U^{(w_{k}^{-1})}\cdots U^{(w_{1}^{-1})}. This is U(w∗)U^{(w^{*})}, since the ii-th letter of w∗w^{*} is (wj)−1(w_{j})^{-1} with i+j=k+1i+j=k+1 (Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §adjoint, Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §reversal).

Step 5 (the control operators). Let i∈[d]i\in[d]. Let FiF_{i} be a support set of aia^{i} with nin_{i} elements, and φi:[ni]→Fi\varphi_{i}:[n_{i}]\to F_{i} a bijection. Put

Ai′=∑q=1niai(φi(q)) U(φi(q)),Ai=12(Ai′+Ai′∗),A'_{i}=\sum_{q=1}^{n_{i}}a^{i}\bigl(\varphi_{i}(q)\bigr)\,U^{(\varphi_{i}(q))},\qquad A^{i}=\tfrac12\bigl(A'_{i}+A'^{*}_{i}\bigr),

finite sums in the complex vector space L(H)\mathcal{L}(H) (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations). Both operators lie in MM by (O2). By Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus, (Ai)∗=12(Ai′∗+Ai′)=Ai(A^{i})^{*}=\frac12(A'^{*}_{i}+A'_{i})=A^{i}, so AiA^{i} is self-adjoint. Put N=1+∑i∈[d]∥Ai∥opN=1+\sum_{i\in[d]}\lVert A^{i}\rVert_{\mathrm{op}}. Then 1≤N1\le N, and ∥Ai∥op≤N\lVert A^{i}\rVert_{\mathrm{op}}\le N for every ii (Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §monotone).

(T) τ(AiU(r))=λ(aier)\tau(A^{i}U^{(r)})=\lambda(a^{i}e_{r}) for every r∈W2dr\in W_{2d}. By (O2), (O4) and distributivity of composition over sums (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations), τ(Ai′U(r))=∑q=1niai(φi(q)) τ(U(φi(q)r))\tau(A'_{i}U^{(r)})=\sum_{q=1}^{n_{i}}a^{i}(\varphi_{i}(q))\,\tau(U^{(\varphi_{i}(q)r)}). Since λU=λ\lambda_{U}=\lambda and by Sum over a Finite Index Set, this equals ∑v∈Fiai(v)λ(vr)\sum_{v\in F_{i}}a^{i}(v)\lambda(vr). On the other hand, ai=∑v∈Fiai(v)eva^{i}=\sum_{v\in F_{i}}a^{i}(v)e_{v} pointwise: at w∈Fiw\in F_{i} only the term v=wv=w is nonzero (Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing), and outside FiF_{i} both sides vanish. So 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 (by induction on the number of summands), together with λ(evr)=λ(vr)\lambda(e_{vr})=\lambda(vr), give λ(aier)=∑v∈Fiai(v)λ(ever)=∑v∈Fiai(v)λ(vr)=τ(Ai′U(r))\lambda(a^{i}e_{r})=\sum_{v\in F_{i}}a^{i}(v)\lambda(e_{v}e_{r})=\sum_{v\in F_{i}}a^{i}(v)\lambda(vr)=\tau(A'_{i}U^{(r)}). Next, by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus, Ai′∗U(r)A'^{*}_{i}U^{(r)} is the adjoint of (U(r))∗Ai′=U(r∗)Ai′(U^{(r)})^{*}A'_{i}=U^{(r^{*})}A'_{i}, using (O6). So (O2) and the case just proved give τ(Ai′∗U(r))=τ(U(r∗)Ai′)‾=τ(Ai′U(r∗))‾=λ(aier∗)‾\tau(A'^{*}_{i}U^{(r)})=\overline{\tau(U^{(r^{*})}A'_{i})}=\overline{\tau(A'_{i}U^{(r^{*})})}=\overline{\lambda(a^{i}e_{r^{*}})}. By Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §adjoint, Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §algebra (with (er∗)∗=er(e_{r^{*}})^{*}=e_{r}, since (r∗)∗=r(r^{*})^{*}=r), self-adjointness of aia^{i}, and Word Polynomials under a Unitary Law: Linearity, Traciality, Positivity, the Cauchy-Schwarz Inequality and the l1 Bound §tracial:

λ(aier∗)‾=λ((aier∗)∗)=λ(er ai)=λ(aier).\overline{\lambda(a^{i}e_{r^{*}})}=\lambda\bigl((a^{i}e_{r^{*}})^{*}\bigr)=\lambda\bigl(e_{r}\,a^{i}\bigr)=\lambda\bigl(a^{i}e_{r}\bigr).

Hence τ(AiU(r))=12λ(aier)+12λ(aier)=λ(aier)\tau(A^{i}U^{(r)})=\frac12\lambda(a^{i}e_{r})+\frac12\lambda(a^{i}e_{r})=\lambda(a^{i}e_{r}).

Step 6 (Cayley transforms). Put s0=14N>0s_{0}=\frac{1}{4N}>0. Let ss be real with ∣s∣≤s0|s|\le s_{0}, so that ∣s∣N≤14|s|N\le\frac14, and let i∈[d]i\in[d]. Put Xi=is2Ai∈MX_{i}=\frac{\mathrm{i}s}{2}A^{i}\in M. Then ∥Xi∥op=∣s∣2∥Ai∥op≤18<1\lVert X_{i}\rVert_{\mathrm{op}}=\frac{|s|}{2}\lVert A^{i}\rVert_{\mathrm{op}}\le\frac18<1 by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations. By The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §neumann, Ti=I−XiT_{i}=I-X_{i} is a bijection of HH onto HH with Ti−1∈L(H)T_{i}^{-1}\in\mathcal{L}(H) and ∥Ti−1∥op≤11−1/8=87\lVert T_{i}^{-1}\rVert_{\mathrm{op}}\le\frac{1}{1-1/8}=\frac87. As Ti∈MT_{i}\in M and M=M′′M=M'' (Tracial W*-Probability Spaces §space), Ti−1∈MT_{i}^{-1}\in M by The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §commutant. Put Ti′=I+XiT'_{i}=I+X_{i} and Ci=Ti′Ti−1∈MC_{i}=T'_{i}T_{i}^{-1}\in M.

(a) Ci∗Ci=CiCi∗=IC_{i}^{*}C_{i}=C_{i}C_{i}^{*}=I. As ss is real and AiA^{i} is self-adjoint, Xi∗=is2‾Ai=−XiX_{i}^{*}=\overline{\tfrac{\mathrm{i}s}{2}}A^{i}=-X_{i} (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus, Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-unique); so Ti∗=Ti′T_{i}^{*}=T'_{i} and Ti′∗=TiT'^{*}_{i}=T_{i}. Also TiTi′=I−Xi2=Ti′TiT_{i}T'_{i}=I-X_{i}^{2}=T'_{i}T_{i}, so Ti−1T_{i}^{-1} commutes with Ti′T'_{i} by The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §commutant. Since Ti−1∈L(H)T_{i}^{-1}\in\mathcal{L}(H) and HH is a complex Hilbert space, Ti−1T_{i}^{-1} has an adjoint (Ti−1)∗∈L(H)(T_{i}^{-1})^{*}\in\mathcal{L}(H) by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint. From TiTi−1=Ti−1Ti=IT_{i}T_{i}^{-1}=T_{i}^{-1}T_{i}=I and Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus we get (Ti−1)∗Ti′=I(T_{i}^{-1})^{*}T'_{i}=I and Ti′(Ti−1)∗=IT'_{i}(T_{i}^{-1})^{*}=I. So Ti′T'_{i} is a bijection with inverse Ti′−1=(Ti−1)∗T'^{-1}_{i}=(T_{i}^{-1})^{*}, and Ci∗=Ti′−1TiC_{i}^{*}=T'^{-1}_{i}T_{i}. Then Ci∗Ci=Ti′−1TiTi′Ti−1=Ti′−1Ti′TiTi−1=IC_{i}^{*}C_{i}=T'^{-1}_{i}T_{i}T'_{i}T_{i}^{-1}=T'^{-1}_{i}T'_{i}T_{i}T_{i}^{-1}=I. Multiplying Ti−1Ti′=Ti′Ti−1T_{i}^{-1}T'_{i}=T'_{i}T_{i}^{-1} by Ti′−1T'^{-1}_{i} on both sides gives Ti′−1Ti−1=Ti−1Ti′−1T'^{-1}_{i}T_{i}^{-1}=T_{i}^{-1}T'^{-1}_{i}, and hence CiCi∗=Ti′Ti−1Ti′−1Ti=Ti′Ti′−1Ti−1Ti=IC_{i}C_{i}^{*}=T'_{i}T_{i}^{-1}T'^{-1}_{i}T_{i}=T'_{i}T'^{-1}_{i}T_{i}^{-1}T_{i}=I.

(b) The path. Put Vi=CiUi∈MV_{i}=C_{i}U_{i}\in M and V(s)=(V1,…,Vd)V(s)=(V_{1},\dots,V_{d}). By (O5) and (a), Vi∗Vi=Ui∗Ci∗CiUi=IV_{i}^{*}V_{i}=U_{i}^{*}C_{i}^{*}C_{i}U_{i}=I and ViVi∗=CiUiUi∗Ci∗=IV_{i}V_{i}^{*}=C_{i}U_{i}U_{i}^{*}C_{i}^{*}=I, so ViV_{i} is unitary by (O5). Define γ(s)=λV(s)\gamma(s)=\lambda_{V(s)}; it belongs to Ld\mathcal{L}_{d} by Unitary Laws Are the Laws of Unitary Tuples, and Are Realised Together with a Free Semicircular Family §law. This defines γ\gamma on {s∈R:−s0≤s≤s0}\{s\in\mathbb{R}:-s_{0}\le s\le s_{0}\}. For s=0s=0 we have Xi=0X_{i}=0, Ci=IC_{i}=I and V(0)=UV(0)=U, so γ(0)=λU=λ\gamma(0)=\lambda_{U}=\lambda.

(c) Expansion. Since Ti′=Ti+2XiT'_{i}=T_{i}+2X_{i}, Ci=I+2XiTi−1C_{i}=I+2X_{i}T_{i}^{-1}. Also Ti−1=I+XiTi−1T_{i}^{-1}=I+X_{i}T_{i}^{-1}, because Ti−1−XiTi−1=TiTi−1=IT_{i}^{-1}-X_{i}T_{i}^{-1}=T_{i}T_{i}^{-1}=I. Hence, with 2Xi=is Ai2X_{i}=\mathrm{i}s\,A^{i},

Di:=Ci−I=is Ai+Ri,Ri=2Xi2Ti−1,∥Ri∥op≤2(∣s∣N2)287≤s2N2.D_{i}:=C_{i}-I=\mathrm{i}s\,A^{i}+R_{i},\qquad R_{i}=2X_{i}^{2}T_{i}^{-1},\qquad\lVert R_{i}\rVert_{\mathrm{op}}\le2\Bigl(\frac{|s|N}{2}\Bigr)^{2}\frac87\le s^{2}N^{2}.

By Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus, Di∗=Ci∗−I=−is Ai+Ri∗D_{i}^{*}=C_{i}^{*}-I=-\mathrm{i}s\,A^{i}+R_{i}^{*}, and ∥Ri∗∥op=∥Ri∥op\lVert R_{i}^{*}\rVert_{\mathrm{op}}=\lVert R_{i}\rVert_{\mathrm{op}} by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint. By Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations and ∣s∣N≤14|s|N\le\frac14, both DiD_{i} and Di∗D_{i}^{*} have operator norm at most ∣s∣N+s2N2≤2∣s∣N≤12|s|N+s^{2}N^{2}\le2|s|N\le\frac12.

Step 7 (expansion along a word). Let 0<∣s∣≤s00<|s|\le s_{0} and let w∈Wd∘w\in W^{\circ}_{d} have length k∈Nk\in\mathbb{N}. For m∈[k]m\in[k] put Lm=U(wm)L_{m}=U^{(w_{m})}, and

Bm=DjmLm  if εm=1,Bm=LmDjm∗  if εm=−1.B_{m}=D_{j_{m}}L_{m}\ \text{ if }\varepsilon_{m}=1,\qquad B_{m}=L_{m}D_{j_{m}}^{*}\ \text{ if }\varepsilon_{m}=-1 .

Then V(wm)=Lm+BmV^{(w_{m})}=L_{m}+B_{m}. If εm=1\varepsilon_{m}=1, V(wm)=CjmUjm=(I+Djm)LmV^{(w_{m})}=C_{j_{m}}U_{j_{m}}=(I+D_{j_{m}})L_{m}. If εm=−1\varepsilon_{m}=-1, V(wm)=(CjmUjm)∗=Ujm∗Cjm∗=Lm(I+Djm∗)V^{(w_{m})}=(C_{j_{m}}U_{j_{m}})^{*}=U_{j_{m}}^{*}C_{j_{m}}^{*}=L_{m}(I+D_{j_{m}}^{*}) (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus). By (O3) and Step 6(c), ∥Bm∥op≤2∣s∣N\lVert B_{m}\rVert_{\mathrm{op}}\le2|s|N. For J⊆[k]J\subseteq[k] let ΠJ=F1J⋯FkJ\Pi^{J}=F^{J}_{1}\cdots F^{J}_{k}, where FmJ=BmF^{J}_{m}=B_{m} for m∈Jm\in J and FmJ=LmF^{J}_{m}=L_{m} for m∉Jm\notin J. Multiplying out V(w)=(L1+B1)⋯(Lk+Bk)V^{(w)}=(L_{1}+B_{1})\cdots(L_{k}+B_{k}) uses distributivity of composition over sums (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations) and induction on kk. It gives V(w)=∑J⊆[k]ΠJV^{(w)}=\sum_{J\subseteq[k]}\Pi^{J}, a sum over the 2k2^{k} subsets of [k][k] (their number, by induction on kk). By (O3) and submultiplicativity, ∥ΠJ∥op≤∏m∈J∥Bm∥op\lVert\Pi^{J}\rVert_{\mathrm{op}}\le\prod_{m\in J}\lVert B_{m}\rVert_{\mathrm{op}}.

(i) J=∅J=\varnothing: Π∅=U(w)\Pi^{\varnothing}=U^{(w)} by (O4), so τ(Π∅)=λU(w)=λ(w)\tau(\Pi^{\varnothing})=\lambda_{U}(w)=\lambda(w).

(ii) J={m}J=\{m\}. If εm=1\varepsilon_{m}=1, let P=L1⋯Lm−1P=L_{1}\cdots L_{m-1} and Q=Lm⋯LkQ=L_{m}\cdots L_{k}. By (O4), P=U(w[1,m−1])P=U^{(w_{[1,m-1]})} (P=IP=I if m=1m=1), Q=U(w[m,k])Q=U^{(w_{[m,k]})}, and Π{m}=PDjmQ\Pi^{\{m\}}=PD_{j_{m}}Q. By (O4) and Rotations and Cyclic Derivatives of Words in Unitary Letters §rotations, QP=U(w[m,k]w[1,m−1])=U(rm)QP=U^{(w_{[m,k]}w_{[1,m-1]})}=U^{(r_{m})} if 1<m1<m, and QP=U(w)=U(r1)QP=U^{(w)}=U^{(r_{1})} if m=1m=1. If εm=−1\varepsilon_{m}=-1, let P=L1⋯Lm=U(w[1,m])P=L_{1}\cdots L_{m}=U^{(w_{[1,m]})} and Q=Lm+1⋯Lk=U(w[m+1,k])Q=L_{m+1}\cdots L_{k}=U^{(w_{[m+1,k]})} (Q=IQ=I if m=km=k). Then Π{m}=PDjm∗Q\Pi^{\{m\}}=PD_{j_{m}}^{*}Q, and again QP=U(rm)QP=U^{(r_{m})} by Rotations and Cyclic Derivatives of Words in Unitary Letters §rotations. In both cases Step 6(c) gives Π{m}=P(iεms Ajm+R)Q\Pi^{\{m\}}=P\bigl(\mathrm{i}\varepsilon_{m}s\,A^{j_{m}}+R\bigr)Q with R=RjmR=R_{j_{m}} or R=Rjm∗R=R_{j_{m}}^{*}. By (O2) and (T), τ(PAjmQ)=τ(AjmQP)=τ(AjmU(rm))=λ(ajmerm)\tau(PA^{j_{m}}Q)=\tau(A^{j_{m}}QP)=\tau(A^{j_{m}}U^{(r_{m})})=\lambda(a^{j_{m}}e_{r_{m}}). By (O1), (O3) and submultiplicativity, ∣τ(PRQ)∣≤∥R∥op≤s2N2|\tau(PRQ)|\le\lVert R\rVert_{\mathrm{op}}\le s^{2}N^{2}. Hence, by the linearity of τ\tau, τ(Π{m})=s iεmλ(ajmerm)+ρ{m}\tau(\Pi^{\{m\}})=s\,\mathrm{i}\varepsilon_{m}\lambda(a^{j_{m}}e_{r_{m}})+\rho_{\{m\}} with ρ{m}:=τ(PRQ)\rho_{\{m\}}:=\tau(PRQ), and ∣ρ{m}∣≤s2N2≤4s2N2|\rho_{\{m\}}|\le s^{2}N^{2}\le4s^{2}N^{2}.

(iii) JJ with at least two elements. As 0≤2∣s∣N≤12≤10\le2|s|N\le\frac12\le1, ∥ΠJ∥op≤(2∣s∣N)2=4s2N2\lVert\Pi^{J}\rVert_{\mathrm{op}}\le(2|s|N)^{2}=4s^{2}N^{2}. So ρJ:=τ(ΠJ)\rho_{J}:=\tau(\Pi^{J}) satisfies ∣ρJ∣≤4s2N2|\rho_{J}|\le4s^{2}N^{2} by (O1).

By the linearity of τ\tau, γ(s)(w)=τ(V(w))=∑J⊆[k]τ(ΠJ)\gamma(s)(w)=\tau(V^{(w)})=\sum_{J\subseteq[k]}\tau(\Pi^{J}). Let J\mathcal{J} be the set of nonempty subsets of [k][k]; it is nonempty since [k]∈J[k]\in\mathcal{J} (k≥1k\ge1), and it has 2k−1<2k2^{k}-1<2^{k} elements. For J∈JJ\in\mathcal{J} put φ(J)=s iεmλ(ajmerm)\varphi(J)=s\,\mathrm{i}\varepsilon_{m}\lambda(a^{j_{m}}e_{r_{m}}) if J={m}J=\{m\} and φ(J)=0\varphi(J)=0 if JJ has at least two elements; by (ii) and (iii), τ(ΠJ)=φ(J)+ρJ\tau(\Pi^{J})=\varphi(J)+\rho_{J} with ∣ρJ∣≤4s2N2|\rho_{J}|\le4s^{2}N^{2} for every J∈JJ\in\mathcal{J}. Splitting off J=∅J=\varnothing by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §disjoint-union (with F={∅}F=\{\varnothing\} and G=JG=\mathcal{J}), and using (i) and additivity,

γ(s)(w)=λ(w)+∑J∈Jφ(J)+∑J∈JρJ.\gamma(s)(w)=\lambda(w)+\sum_{J\in\mathcal{J}}\varphi(J)+\sum_{J\in\mathcal{J}}\rho_{J}.

The set S\mathcal{S} of one-element subsets of [k][k] is a nonempty subset of J\mathcal{J} outside which φ\varphi vanishes, and m↦{m}m\mapsto\{m\} is a bijection from [k][k] onto S\mathcal{S}. So Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing, reindexing, homogeneity and (B1) give ∑J∈Jφ(J)=∑m∈[k]s iεmλ(ajmerm)=s b(w)\sum_{J\in\mathcal{J}}\varphi(J)=\sum_{m\in[k]}s\,\mathrm{i}\varepsilon_{m}\lambda(a^{j_{m}}e_{r_{m}})=s\,b(w). Finally, Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §modulus, Real Sums over a Finite Index Set: Comparison, Nonnegativity, Monotonicity, Term Bounds, Absolute Values, Counting and Limits §comparison and the count of J\mathcal{J} (a sum of a nonnegative constant over a set with nn elements is nn times the constant, as in Step 2) give

∣γ(s)(w)−λ(w)−s b(w)∣=∣∑J∈JρJ∣≤(2k−1)⋅4s2N2≤4⋅2ks2N2(0<∣s∣≤s0, w∈Wd∘ of length k).(∗)\bigl|\gamma(s)(w)-\lambda(w)-s\,b(w)\bigr|=\Bigl|\sum_{J\in\mathcal{J}}\rho_{J}\Bigr|\le(2^{k}-1)\cdot4s^{2}N^{2}\le4\cdot2^{k}s^{2}N^{2}\qquad(0<|s|\le s_{0},\ w\in W^{\circ}_{d}\text{ of length }k).\tag{$\ast$}

Step 8 (the two-sided tangent vector). For 0<∣s∣≤s00<|s|\le s_{0} let vs=1s(ιd(γ(s))−ιd(λ))−bv_{s}=\frac1s\bigl(\iota_{d}(\gamma(s))-\iota_{d}(\lambda)\bigr)-b. It lies in EdE_{d} by The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §space, The Embedding of Unitary Laws into the Word Gauge Space and the Gauge Distance §embedding and Step 2. By The Word Gauge Space Is a Real Hilbert Space, and the Unitary Laws Form a Compact Metric Space in It §sobolev and the pointwise operations, vs(w)=1s(γ(s)(w)−λ(w))−b(w)v_{s}(w)=\frac1s\bigl(\gamma(s)(w)-\lambda(w)\bigr)-b(w) for w∈Wd∘w\in W^{\circ}_{d}. At w=∅w=\varnothing, vs(∅)=0v_{s}(\varnothing)=0: indeed γ(s)(∅)=τ(I)=1=λ(∅)\gamma(s)(\varnothing)=\tau(I)=1=\lambda(\varnothing) by (O2) and Laws of d-Tuples of Unitaries §normalised, and b(∅)=0b(\varnothing)=0 by (B1). For w∈Wd,k∘w\in W^{\circ}_{d,k}, (∗\ast) and claims 4 and 8 of Properties of Complex Conjugation and Modulus give ∣vs(w)∣=∣s∣−1∣γ(s)(w)−λ(w)−s b(w)∣≤4⋅2k∣s∣N2|v_{s}(w)|=|s|^{-1}|\gamma(s)(w)-\lambda(w)-s\,b(w)|\le4\cdot2^{k}|s|N^{2}. Hence cw∣vs(w)∣2≤16 s2N4 (4θd)kc_{w}|v_{s}(w)|^{2}\le16\,s^{2}N^{4}\,(4\theta_{d})^{k}, by claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field and claim 3 of Properties of Natural Number Powers in a Field. By Counting Cyclically Reduced Words: Finiteness and the Bound (2d)^k §count and claim 3 of Properties of Natural Number Powers in a Field, the kk-th block sum satisfies

0≤∑w∈Wd,k∘cw∣vs(w)∣2≤16 s2N4(8d θd)k=16 s2N4(1768)k.0\le\sum_{w\in W^{\circ}_{d,k}}c_{w}|v_{s}(w)|^{2}\le16\,s^{2}N^{4}(8d\,\theta_{d})^{k}=16\,s^{2}N^{4}\bigl(\tfrac1{768}\bigr)^{k}.

By Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric and Elementary Properties of Series of Real Numbers §linearity, ∑k=1∞16 s2N4(1768)k=16767s2N4≤s2N4\sum_{k=1}^{\infty}16\,s^{2}N^{4}(\frac1{768})^{k}=\frac{16}{767}s^{2}N^{4}\le s^{2}N^{4}. So by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §comparison and The Word Gauge: Weights, the Space of Weighted Square-Summable Word Functions, Its Norm and Inner Product, and the Length-Weighted Gauge §norm, ∥vs∥d2=0+∑k=1∞∑w∈Wd,k∘cw∣vs(w)∣2≤(∣s∣N2)2\lVert v_{s}\rVert_{d}^{2}=0+\sum_{k=1}^{\infty}\sum_{w\in W^{\circ}_{d,k}}c_{w}|v_{s}(w)|^{2}\le(|s|N^{2})^{2}. Hence ∥vs∥d≤∣s∣N2\lVert v_{s}\rVert_{d}\le|s|N^{2} by claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field.

Now let ε>0\varepsilon>0 be given and choose δ=εN−2>0\delta=\varepsilon N^{-2}>0. For every real ss with 0<∣s∣≤s00<|s|\le s_{0} and ∣s∣<δ|s|<\delta, ∥vs∥d≤∣s∣N2<ε\lVert v_{s}\rVert_{d}\le|s|N^{2}<\varepsilon. By Step 6(b), γ\gamma maps {s:−s0≤s≤s0}\{s:-s_{0}\le s\le s_{0}\} into Ld\mathcal{L}_{d} with γ(0)=λ\gamma(0)=\lambda. So by Inward and Two-Sided Tangent Vectors to the Space of Unitary Laws §two-sided, ba(λ)∈Tλ±b_{a}(\lambda)\in T^{\pm}_{\lambda}, which is the set of Unitary Laws with Free Unitary Noise: Standing Data §noise. ■\blacksquare

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…