TheoremBase

Glue the joint law of the position with both momenta and their sum to an arbitrary realisation of the summed plan, apply the two jet inequalities there, add them, and transfer back by law invariance.

Proof

Each result cited is universally quantified over the data in its own statement. For L2L^{2} dd-tuples of one tracial W*-probability space, sums and real multiples are those of HdH^{d} and ⟨⋅,⋅⟩2\langle\cdot,\cdot\rangle_{2} is the inner product of HdH^{d}, with norm ∥⋅∥2\lVert\cdot\rVert_{2}, by Sums, Real Multiples and the Pairing of Square-Integrable Tuples in a Tracial W*-Probability Space §operations and Sums, Real Multiples and the Pairing of Square-Integrable Tuples in a Tracial W*-Probability Space §pairing; so the pairing is bilinear over R\mathbb{R}, and a tuple of norm 00 is the zero tuple. Lifts satisfy fM(Y)=f(law(Y))f_{M}(Y)=f(\mathrm{law}(Y)) by Lifts of Functions on Square-Integrable Noncommutative Laws to Square-Integrable Tuples §lift; in particular (φ1+φ2)M(Y)=(φ1)M(Y)+(φ2)M(Y)(\varphi_{1}+\varphi_{2})_{M}(Y)=(\varphi_{1})_{M}(Y)+(\varphi_{2})_{M}(Y) for every L2L^{2} dd-tuple YY.

Step 0 (law invariance). Let AA and A′A' be L2L^{2} kk-tuples of two (possibly different) tracial W*-probability spaces with law(A)=law(A′)\mathrm{law}(A)=\mathrm{law}(A'), and let T,ST,S be affine data from kk to dd variables with zero constant part. By the entrywise formula of Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §operations, the pair (TA,SA)(TA,SA) equals UAUA, where UU is the affine datum from kk to 2d2d variables whose first dd rows are those of TT and whose last dd rows are those of SS; likewise (TA′,SA′)=UA′(TA',SA')=UA'. By Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §push-forward, law(TA)=T#law(A)=T#law(A′)=law(TA′)\mathrm{law}(TA)=T_{\#}\mathrm{law}(A)=T_{\#}\mathrm{law}(A')=\mathrm{law}(TA') and similarly law(UA)=law(UA′)\mathrm{law}(UA)=\mathrm{law}(UA'). By Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §moments, applied to TATA and to UAUA and to their primed counterparts,

∥TA∥22=M^(law(TA))=∥TA′∥22,⟨TA,SA⟩2=∑j=1dmj,d+j(law(UA))=⟨TA′,SA′⟩2.\lVert TA\rVert_{2}^{2}=\widehat{M}(\mathrm{law}(TA))=\lVert TA'\rVert_{2}^{2},\qquad\langle TA,SA\rangle_{2}=\sum_{j=1}^{d}\mathrm{m}_{j,d+j}(\mathrm{law}(UA))=\langle TA',SA'\rangle_{2}.

Also fM(TA)=f(law(TA))=f(law(TA′))f_{M}(TA)=f(\mathrm{law}(TA))=f(\mathrm{law}(TA')) for every f:Σd2→Rf:\Sigma^{2}_{d}\to\mathbb{R}, where the second lift is taken in the space of A′A'. Every sub-tuple, sum and difference of blocks of a tuple used below is of the form TATA with such a datum TT, by the same entrywise formula. So norms, pairings and lift values of such combinations depend only on the law of AA.

Step 1 (the summed law is a plan at λ\lambda). law(X,P1+P2)∈Σ2d2\mathrm{law}(X,P_{1}+P_{2})\in\Sigma^{2}_{2d} by Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §law. By Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §push-forward, pr#1law(X,P1+P2)=law(pr1(X,P1+P2))=law(X)=λ\mathrm{pr}^{1}_{\#}\mathrm{law}(X,P_{1}+P_{2})=\mathrm{law}(\mathrm{pr}^{1}(X,P_{1}+P_{2}))=\mathrm{law}(X)=\lambda, since the entries of pr1(X,P1+P2)\mathrm{pr}^{1}(X,P_{1}+P_{2}) are those of XX (Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §operations, Affine Data and Affine Substitutions of Noncommutative Polynomials §coordinate). Hence law(X,P1+P2)\mathrm{law}(X,P_{1}+P_{2}) is a plan at λ\lambda by Plans at a Square-Integrable Noncommutative Law §plan.

Step 2 (Claim 1, superjets). Fix a real η>0\eta>0. By Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §superjet and Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §super, applied to φi\varphi_{i}, the plan law(X,Pi)\mathrm{law}(X,P_{i}) and slack δi\delta_{i} with the positive number η/2\eta/2 in place of η\eta, there are reals ri>0r_{i}>0 (i=1,2i=1,2), depending only on η\eta, such that for every tracial W*-probability space (K,N,Ψ)(K,N,\Psi) and all L2L^{2} dd-tuples A,B,A′A,B,A' of it with law(A,B)=law(X,Pi)\mathrm{law}(A,B)=\mathrm{law}(X,P_{i}) and ∥A′−A∥2<ri\lVert A'-A\rVert_{2}<r_{i},

(φi)N(A′)≤φi(λ)+⟨B,A′−A⟩2+(δi+η/2)∥A′−A∥2.(\varphi_{i})_{N}(A')\le\varphi_{i}(\lambda)+\langle B,A'-A\rangle_{2}+(\delta_{i}+\eta/2)\lVert A'-A\rVert_{2}.

Put r=min⁡{r1,r2}>0r=\min\{r_{1},r_{2}\}>0. Now let (H~,M~,Ω~)(\tilde H,\tilde M,\tilde\Omega) be any tracial W*-probability space and X~,P~,X~′\tilde X,\tilde P,\tilde X' any L2L^{2} dd-tuples of it with law(X~,P~)=law(X,P1+P2)\mathrm{law}(\tilde X,\tilde P)=\mathrm{law}(X,P_{1}+P_{2}) and ∥X~′−X~∥2<r\lVert\tilde X'-\tilde X\rVert_{2}<r.

Gluing. Put π1=law((X,P1+P2),(P1,P2))∈Σ4d2\pi_{1}=\mathrm{law}((X,P_{1}+P_{2}),(P_{1},P_{2}))\in\Sigma^{2}_{4d} and π2=law(X~,P~,X~′)∈Σ3d2\pi_{2}=\mathrm{law}(\tilde X,\tilde P,\tilde X')\in\Sigma^{2}_{3d}. With k=2dk=2d and the data F2d,FdF^{2d},F^{d} of Gluing Two Square-Integrable Noncommutative Laws along a Common Marginal, which select the first 2d2d entries, Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §push-forward gives F#2dπ1=law(X,P1+P2)=law(X~,P~)=F#dπ2F^{2d}_{\#}\pi_{1}=\mathrm{law}(X,P_{1}+P_{2})=\mathrm{law}(\tilde X,\tilde P)=F^{d}_{\#}\pi_{2}. By Gluing Two Square-Integrable Noncommutative Laws along a Common Marginal §glue (with k=2dk=2d, m=2dm=2d, n=dn=d and these π1,π2\pi_{1},\pi_{2}) there are a tracial W*-probability space (H^,M^,Ω^)(\hat H,\hat M,\hat\Omega), an L2L^{2} 2d2d-tuple (X^,P^)(\hat X,\hat P) (written by its two blocks of dd entries), an L2L^{2} 2d2d-tuple (Y1,Y2)(Y_{1},Y_{2}) and an L2L^{2} dd-tuple X^′\hat X' of it with

law((X^,P^),(Y1,Y2))=law((X,P1+P2),(P1,P2)),law(X^,P^,X^′)=law(X~,P~,X~′).\mathrm{law}((\hat X,\hat P),(Y_{1},Y_{2}))=\mathrm{law}((X,P_{1}+P_{2}),(P_{1},P_{2})),\qquad\mathrm{law}(\hat X,\hat P,\hat X')=\mathrm{law}(\tilde X,\tilde P,\tilde X').

Consequences of law invariance. By Step 0 applied to the first equality of laws: ∥P^−Y1−Y2∥2=∥(P1+P2)−P1−P2∥2=0\lVert\hat P-Y_{1}-Y_{2}\rVert_{2}=\lVert(P_{1}+P_{2})-P_{1}-P_{2}\rVert_{2}=0, so P^=Y1+Y2\hat P=Y_{1}+Y_{2}; and law(X^,Yi)=law(X,Pi)\mathrm{law}(\hat X,Y_{i})=\mathrm{law}(X,P_{i}) for i=1,2i=1,2 (push-forward under the datum selecting the corresponding blocks, Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §push-forward). By Step 0 applied to the second equality of laws: ∥X^′−X^∥2=∥X~′−X~∥2<r\lVert\hat X'-\hat X\rVert_{2}=\lVert\tilde X'-\tilde X\rVert_{2}<r, ⟨P^,X^′−X^⟩2=⟨P~,X~′−X~⟩2\langle\hat P,\hat X'-\hat X\rangle_{2}=\langle\tilde P,\tilde X'-\tilde X\rangle_{2}, and (φ1+φ2)M^(X^′)=(φ1+φ2)M~(X~′)(\varphi_{1}+\varphi_{2})_{\hat M}(\hat X')=(\varphi_{1}+\varphi_{2})_{\tilde M}(\tilde X').

Adding the inequalities. The triples (X^,Yi,X^′)(\hat X,Y_{i},\hat X') in (H^,M^,Ω^)(\hat H,\hat M,\hat\Omega) are admissible in the displayed inequality for i=1,2i=1,2, since law(X^,Yi)=law(X,Pi)\mathrm{law}(\hat X,Y_{i})=\mathrm{law}(X,P_{i}) and ∥X^′−X^∥2<r≤ri\lVert\hat X'-\hat X\rVert_{2}<r\le r_{i}. Adding the two resulting inequalities, and using the additivity of lifts, bilinearity of the pairing and Y1+Y2=P^Y_{1}+Y_{2}=\hat P,

(φ1+φ2)M^(X^′)≤(φ1+φ2)(λ)+⟨P^,X^′−X^⟩2+(δ1+δ2+η)∥X^′−X^∥2.(\varphi_{1}+\varphi_{2})_{\hat M}(\hat X')\le(\varphi_{1}+\varphi_{2})(\lambda)+\langle\hat P,\hat X'-\hat X\rangle_{2}+(\delta_{1}+\delta_{2}+\eta)\lVert\hat X'-\hat X\rVert_{2}.

By the consequences of law invariance, each term equals its counterpart for (X~,P~,X~′)(\tilde X,\tilde P,\tilde X'), so

(φ1+φ2)M~(X~′)≤(φ1+φ2)(λ)+⟨P~,X~′−X~⟩2+(δ1+δ2+η)∥X~′−X~∥2.(\varphi_{1}+\varphi_{2})_{\tilde M}(\tilde X')\le(\varphi_{1}+\varphi_{2})(\lambda)+\langle\tilde P,\tilde X'-\tilde X\rangle_{2}+(\delta_{1}+\delta_{2}+\eta)\lVert\tilde X'-\tilde X\rVert_{2}.

As η>0\eta>0 was arbitrary and rr depends only on η\eta, the plan law(X,P1+P2)\mathrm{law}(X,P_{1}+P_{2}) at λ\lambda (Step 1) is a plan superdifferential of φ1+φ2\varphi_{1}+\varphi_{2} at λ\lambda with slack δ1+δ2\delta_{1}+\delta_{2} in the sense of Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §super, that is, it lies in Jδ1+δ2+(φ1+φ2)(λ)J^{+}_{\delta_{1}+\delta_{2}}(\varphi_{1}+\varphi_{2})(\lambda) by Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §superjet.

Step 3 (Claim 2, subjets). The argument of Step 2 applies verbatim with Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §sub and Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §subjet in place of Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §super and Plan Superdifferentials, Plan Subdifferentials and Plan Jets with Slack on Square-Integrable Noncommutative Laws §superjet: the inequalities are reversed and the slack terms (δi+η/2)∥A′−A∥2(\delta_{i}+\eta/2)\lVert A'-A\rVert_{2} enter with a minus sign, so adding them gives

(φ1+φ2)M~(X~′)≥(φ1+φ2)(λ)+⟨P~,X~′−X~⟩2−(δ1+δ2+η)∥X~′−X~∥2(\varphi_{1}+\varphi_{2})_{\tilde M}(\tilde X')\ge(\varphi_{1}+\varphi_{2})(\lambda)+\langle\tilde P,\tilde X'-\tilde X\rangle_{2}-(\delta_{1}+\delta_{2}+\eta)\lVert\tilde X'-\tilde X\rVert_{2}

for every admissible realisation (X~,P~,X~′)(\tilde X,\tilde P,\tilde X') with ∥X~′−X~∥2<r\lVert\tilde X'-\tilde X\rVert_{2}<r. Hence law(X,P1+P2)∈Jδ1+δ2−(φ1+φ2)(λ)\mathrm{law}(X,P_{1}+P_{2})\in J^{-}_{\delta_{1}+\delta_{2}}(\varphi_{1}+\varphi_{2})(\lambda).

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…