TheoremBase

Proof of Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound

lemmalem:bounded-maps-complex-inner-product-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 24,014 chars · 36 deps · depth 13 Reason: Proof of the complex bounded-maps toolkit (adjoints via real Riesz).

Proves the eight claims in turn: the operator norm is the least bound, the operations and norm inequalities follow pointwise, the real part of the inner product gives a real Hilbert structure, and adjoints are obtained from the real Riesz representation theorem via the identity h(u,v)=Re h(u,v)-i Re h(u,iv). Completeness is proved by pointwise limits with a uniform estimate, and the quadratic-form bound by polarisation and the parallelogram identity.

Proof

Throughout, subscripts on inner products and norms are omitted when the space is clear from the arguments, 00 also denotes the zero vector of each vector space, and x−y=x+(−y)x-y=x+(-y), where −y=(−1)y-y=(-1)y by claim 5 of Elementary Identities in a Vector Space. Elementary real order and arithmetic is used as licensed by The Real Numbers: Standing Notation and Background.

Preliminaries. (P1) Complex arithmetic. Let z,w∈Cz,w\in\mathbb{C} and a∈Ra\in\mathbb{R}, and write z=x+yiz=x+yi and w=p+qiw=p+qi with real x,y,p,qx,y,p,q, as allowed by claim 3 of Canonical Form and Arithmetic of Complex Numbers. The rules of claim 4 of that lemma give z+w=(x+p)+(y+q)iz+w=(x+p)+(y+q)i, az=(ax)+(ay)iaz=(ax)+(ay)i (writing a=a+0ia=a+0i), iz=(−y)+xiiz=(-y)+xi and (−i)z=y+(−x)i(-i)z=y+(-x)i, and z‾=x+(−y)i\overline{z}=x+(-y)i by Complex Conjugate. By the uniqueness in claim 3 of Canonical Form and Arithmetic of Complex Numbers and by Real and Imaginary Parts of a Complex Number therefore

Re⁡(z+w)=Re⁡z+Re⁡w,Re⁡(az)=aRe⁡z,Re⁡a=a,Re⁡(iz)=−Im⁡z,Re⁡((−i)z)=Im⁡z,Re⁡z‾=Re⁡z,z=Re⁡z+(Im⁡z) i;\operatorname{Re}(z+w)=\operatorname{Re}z+\operatorname{Re}w,\quad\operatorname{Re}(az)=a\operatorname{Re}z,\quad\operatorname{Re}a=a,\quad\operatorname{Re}(iz)=-\operatorname{Im}z,\quad\operatorname{Re}\bigl((-i)z\bigr)=\operatorname{Im}z,\quad\operatorname{Re}\overline{z}=\operatorname{Re}z,\quad z=\operatorname{Re}z+(\operatorname{Im}z)\,i;

moreover i‾=−i\overline{i}=-i (as i=0+1ii=0+1i), so i‾ i=−(ii)=1\overline{i}\,i=-(ii)=1 by condition 2 of The Complex Numbers. From Properties of Complex Conjugation and Modulus we use: conjugation respects sums and products, is involutive, and fixes exactly the real numbers, so a‾=a\overline{a}=a and −1‾=−1\overline{-1}=-1 (claim 1); z+z‾=2Re⁡zz+\overline{z}=2\operatorname{Re}z (claim 2); ∣z∣=0|z|=0 only if z=0z=0 (claim 3); ∣zw∣=∣z∣ ∣w∣|zw|=|z|\,|w| (claim 4); Re⁡z≤∣z∣\operatorname{Re}z\le|z| and −Re⁡z≤∣z∣-\operatorname{Re}z\le|z|, so the absolute value of the real number Re⁡z\operatorname{Re}z is at most ∣z∣|z| (claim 6); and ∣a∣=a|a|=a when 0≤a0\le a, while ∣a∣|a| is aa or −a-a for every real aa (claim 8).

(P2) Norms. Let XX be one of V,W,UV,W,U and x,y∈Xx,y\in X. The induced norm is a norm by claim 2 of The Induced Norm is a Norm, and Induces a Metric, so by conditions 1-3 of Norm on a Complex Vector Space we have 0≤∥x∥0\le\lVert x\rVert with equality only if x=0x=0, ∥λx∥=∣λ∣ ∥x∥\lVert\lambda x\rVert=|\lambda|\,\lVert x\rVert for λ∈C\lambda\in\mathbb{C}, and ∥x+y∥≤∥x∥+∥y∥\lVert x+y\rVert\le\lVert x\rVert+\lVert y\rVert; in particular ∥0∥=∥0⋅0∥=0\lVert0\rVert=\lVert0\cdot0\rVert=0. By Norm Induced by a Complex Inner Product, ⟨x,x⟩=∥x∥2\langle x,x\rangle=\lVert x\rVert^{2}, and by claim 1 of The Induced Norm is a Norm, and Induces a Metric, ∣⟨x,y⟩∣≤∥x∥ ∥y∥|\langle x,y\rangle|\le\lVert x\rVert\,\lVert y\rVert. The metric of XX is d(x,y)=∥x−y∥d(x,y)=\lVert x-y\rVert. Every linear map TT satisfies T0=T(0⋅0)=0⋅T0=0T0=T(0\cdot0)=0\cdot T0=0 by claim 3 of Elementary Identities in a Vector Space and the homogeneity of TT (Linear Map).

(P3) Separation. If x,y∈Xx,y\in X satisfy ⟨x,v⟩=⟨y,v⟩\langle x,v\rangle=\langle y,v\rangle for every v∈Xv\in X, then x=yx=y. Indeed, by claims 1 and 2 of Elementary Properties of a Complex Inner Product and −1‾=−1\overline{-1}=-1 we get ⟨x−y,v⟩=⟨x,v⟩−⟨y,v⟩=0\langle x-y,v\rangle=\langle x,v\rangle-\langle y,v\rangle=0 for every vv; taking v=x−yv=x-y, claim 4 of that lemma gives x−y=0x-y=0, hence x=yx=y.

Claim 1 (Least bound). Let T∈L(V,W)T\in\mathcal{L}(V,W) and let BTB_{T} be its set of bounds, so that ∥T∥op=inf⁡BT\lVert T\rVert_{\mathrm{op}}=\inf B_{T} by Bounded Linear Maps between Complex Normed Spaces and the Operator Norm §operator-norm, a greatest lower bound in the sense of Lower Bound and Greatest Lower Bound in a Totally Ordered Set. Every bound is nonnegative by Bounded Linear Maps between Complex Normed Spaces and the Operator Norm §bounded, so 00 is a lower bound of BTB_{T} and hence 0≤∥T∥op0\le\lVert T\rVert_{\mathrm{op}}. Let v∈Vv\in V. If v=0v=0, then ∥Tv∥=0=∥T∥op∥v∥\lVert Tv\rVert=0=\lVert T\rVert_{\mathrm{op}}\lVert v\rVert by (P2). If v≠0v\neq0, then 0<∥v∥0<\lVert v\rVert by (P2), and each C∈BTC\in B_{T} satisfies ∥Tv∥≤C∥v∥\lVert Tv\rVert\le C\lVert v\rVert, that is ∥Tv∥/∥v∥≤C\lVert Tv\rVert/\lVert v\rVert\le C; so ∥Tv∥/∥v∥\lVert Tv\rVert/\lVert v\rVert is a lower bound of BTB_{T}, hence at most the greatest lower bound ∥T∥op\lVert T\rVert_{\mathrm{op}}, and multiplying by ∥v∥>0\lVert v\rVert>0 gives ∥Tv∥≤∥T∥op∥v∥\lVert Tv\rVert\le\lVert T\rVert_{\mathrm{op}}\lVert v\rVert. Thus ∥T∥op\lVert T\rVert_{\mathrm{op}} is a bound for TT. Every bound CC lies in BTB_{T}, so ∥T∥op≤C\lVert T\rVert_{\mathrm{op}}\le C because ∥T∥op\lVert T\rVert_{\mathrm{op}} is a lower bound of BTB_{T}.

Now let T:V→VT:V\to V be linear, that is, a linear operator on VV, and give VV the norm ∥⋅∥V\lVert\cdot\rVert_{V} of (P2). A real C≥0C\ge0 is a bound for TT in the sense of Bounded Linear Maps between Complex Normed Spaces and the Operator Norm §bounded exactly when ∥Tu∥≤C∥u∥\lVert Tu\rVert\le C\lVert u\rVert for every u∈Vu\in V, which is exactly the condition for CC to be a bound for TT in the sense of Bound for a Linear Operator and Bounded Linear Operator. Hence TT has a bound in one sense if and only if it has one in the other, i.e. T∈L(V)T\in\mathcal{L}(V) if and only if TT is a bounded linear operator. In that case the first paragraph shows that ∥T∥op\lVert T\rVert_{\mathrm{op}} is a bound for TT that is at most every bound for TT, so it is an operator norm of TT in the sense of Operator Norm; since TT has exactly one operator norm by Existence and Uniqueness of the Operator Norm, it is ∥T∥op\lVert T\rVert_{\mathrm{op}}.

Claim 2 (Operations). First, for any linear maps S,T:V→WS,T:V\to W, R:W→UR:W\to U and c∈Cc\in\mathbb{C}, the maps S+TS+T, cTcT and RTRT are linear: for u,u′∈Vu,u'\in V and λ∈C\lambda\in\mathbb{C}, linearity of S,T,RS,T,R (Linear Map) and conditions 1, 2, 5 and 7 of Vector Space over a Field in WW and UU give (S+T)(u+u′)=(S+T)u+(S+T)u′(S+T)(u+u')=(S+T)u+(S+T)u', (S+T)(λu)=λSu+λTu=λ(S+T)u(S+T)(\lambda u)=\lambda Su+\lambda Tu=\lambda(S+T)u, (cT)(u+u′)=cTu+cTu′(cT)(u+u')=cTu+cTu', (cT)(λu)=(cλ)Tu=(λc)Tu=λ(cTu)(cT)(\lambda u)=(c\lambda)Tu=(\lambda c)Tu=\lambda(cTu), R(T(u+u′))=R(Tu)+R(Tu′)R(T(u+u'))=R(Tu)+R(Tu') and R(T(λu))=λR(Tu)R(T(\lambda u))=\lambda R(Tu).

Now let S,T∈L(V,W)S,T\in\mathcal{L}(V,W), R∈L(W,U)R\in\mathcal{L}(W,U), c∈Cc\in\mathbb{C} and v∈Vv\in V. By (P2) and claim 1,

∥(S+T)v∥≤∥Sv∥+∥Tv∥≤(∥S∥op+∥T∥op)∥v∥,∥(cT)v∥=∣c∣ ∥Tv∥≤∣c∣ ∥T∥op∥v∥,\lVert(S+T)v\rVert\le\lVert Sv\rVert+\lVert Tv\rVert\le\bigl(\lVert S\rVert_{\mathrm{op}}+\lVert T\rVert_{\mathrm{op}}\bigr)\lVert v\rVert,\qquad\lVert(cT)v\rVert=|c|\,\lVert Tv\rVert\le|c|\,\lVert T\rVert_{\mathrm{op}}\lVert v\rVert, ∥(RT)v∥≤∥R∥op∥Tv∥≤∥R∥op∥T∥op∥v∥,∥IVv∥=1⋅∥v∥,\lVert(RT)v\rVert\le\lVert R\rVert_{\mathrm{op}}\lVert Tv\rVert\le\lVert R\rVert_{\mathrm{op}}\lVert T\rVert_{\mathrm{op}}\lVert v\rVert,\qquad\lVert I_{V}v\rVert=1\cdot\lVert v\rVert,

where the middle step of the third chain multiplies the bound for TvTv by ∥R∥op≥0\lVert R\rVert_{\mathrm{op}}\ge0. The multipliers on the right are nonnegative real numbers, so they are bounds; hence S+T,cT∈L(V,W)S+T,cT\in\mathcal{L}(V,W), RT∈L(V,U)RT\in\mathcal{L}(V,U) and IV∈L(V)I_{V}\in\mathcal{L}(V) (IVI_{V} is linear), and the second part of claim 1 gives ∥S+T∥op≤∥S∥op+∥T∥op\lVert S+T\rVert_{\mathrm{op}}\le\lVert S\rVert_{\mathrm{op}}+\lVert T\rVert_{\mathrm{op}}, ∥cT∥op≤∣c∣ ∥T∥op\lVert cT\rVert_{\mathrm{op}}\le|c|\,\lVert T\rVert_{\mathrm{op}}, ∥RT∥op≤∥R∥op∥T∥op\lVert RT\rVert_{\mathrm{op}}\le\lVert R\rVert_{\mathrm{op}}\lVert T\rVert_{\mathrm{op}} and ∥IV∥op≤1\lVert I_{V}\rVert_{\mathrm{op}}\le1. For the reverse inequality ∣c∣ ∥T∥op≤∥cT∥op|c|\,\lVert T\rVert_{\mathrm{op}}\le\lVert cT\rVert_{\mathrm{op}}: if c=0c=0, then ∣c∣=0|c|=0 by claim 8 of Properties of Complex Conjugation and Modulus and 0≤∥cT∥op0\le\lVert cT\rVert_{\mathrm{op}} by claim 1. If c≠0c\neq0, then 0<∣c∣0<|c| by (P1), c−1(cT)=Tc^{-1}(cT)=T by conditions 5 and 6 of Vector Space over a Field in WW, and ∣c−1∣ ∣c∣=∣1∣=1|c^{-1}|\,|c|=|1|=1 by (P1); the inequality just proved, applied to c−1c^{-1} and cTcT, gives ∥T∥op≤∣c−1∣ ∥cT∥op\lVert T\rVert_{\mathrm{op}}\le|c^{-1}|\,\lVert cT\rVert_{\mathrm{op}}, and multiplying by ∣c∣|c| gives the claim.

The zero map 0:V→W0:V\to W is linear and, since ∥0∥=0\lVert0\rVert=0 by (P2), has the bound 00; so 0∈L(V,W)0\in\mathcal{L}(V,W). Two maps V→WV\to W are equal when they agree at every v∈Vv\in V. Evaluating at each vv, conditions 1, 2, 5, 6, 7 and 8 of Vector Space over a Field for L(V,W)\mathcal{L}(V,W) with the operations above follow from the same conditions in WW; condition 3 holds with the zero map, as Tv+0=TvTv+0=Tv; and condition 4 holds with (−1)T∈L(V,W)(-1)T\in\mathcal{L}(V,W), as Tv+(−1)Tv=Tv+(−Tv)=0Tv+(-1)Tv=Tv+(-Tv)=0 by claim 5 of Elementary Identities in a Vector Space. So L(V,W)\mathcal{L}(V,W) is a complex vector space with zero vector the zero map. Finally, for R,R′∈L(W,U)R,R'\in\mathcal{L}(W,U), S,T∈L(V,W)S,T\in\mathcal{L}(V,W), c∈Cc\in\mathbb{C} and v∈Vv\in V, linearity of RR gives R(Sv+Tv)=R(Sv)+R(Tv)R(Sv+Tv)=R(Sv)+R(Tv) and R(cTv)=c R(Tv)R(cTv)=c\,R(Tv), while (R+R′)(Tv)=R(Tv)+R′(Tv)(R+R')(Tv)=R(Tv)+R'(Tv) and (cR)(Tv)=c R(Tv)(cR)(Tv)=c\,R(Tv) by definition; hence R(S+T)=RS+RTR(S+T)=RS+RT, R(cT)=c(RT)R(cT)=c(RT), (R+R′)T=RT+R′T(R+R')T=RT+R'T and (cR)T=c(RT)(cR)T=c(RT).

Claim 3 (Underlying real structure). Since R⊆C\mathbb{R}\subseteq\mathbb{C} with the same sums and products (condition 1 of The Complex Numbers) and the same identity 11 (claim 1 of Canonical Form and Arithmetic of Complex Numbers), conditions 1-8 of Vector Space over a Field for XX with real scalars are instances of those with complex scalars, so XX is a vector space over R\mathbb{R}. Let u,u′,v,w∈Xu,u',v,w\in X, c∈Cc\in\mathbb{C} and a∈Ra\in\mathbb{R}. By the third hypothesis on hh, the first two, and (P1),

h(u+u′,v)=h(v,u)+h(v,u′)‾=h(u,v)+h(u′,v),h(cu,v)=c h(v,u)‾=c‾ h(u,v).(∗)h(u+u',v)=\overline{h(v,u)+h(v,u')}=h(u,v)+h(u',v),\qquad h(cu,v)=\overline{c\,h(v,u)}=\overline{c}\,h(u,v).\qquad(\ast)

By (P1), β(v,u)=Re⁡h(u,v)‾=Re⁡h(u,v)=β(u,v)\beta(v,u)=\operatorname{Re}\overline{h(u,v)}=\operatorname{Re}h(u,v)=\beta(u,v), so β\beta is symmetric; β(u,v+w)=Re⁡(h(u,v)+h(u,w))=β(u,v)+β(u,w)\beta(u,v+w)=\operatorname{Re}\bigl(h(u,v)+h(u,w)\bigr)=\beta(u,v)+\beta(u,w) and β(u,av)=Re⁡(a h(u,v))=a β(u,v)\beta(u,av)=\operatorname{Re}\bigl(a\,h(u,v)\bigr)=a\,\beta(u,v), so v↦β(u,v)v\mapsto\beta(u,v) is linear over R\mathbb{R} in the sense of Linear Map, and by symmetry so is u↦β(u,v)u\mapsto\beta(u,v); and since h(v,v)h(v,v) is real, β(v,v)=Re⁡h(v,v)=h(v,v)≥0\beta(v,v)=\operatorname{Re}h(v,v)=h(v,v)\ge0. Thus β\beta is symmetric, bilinear and positive semidefinite as in Cauchy Sequences for a Positive Semidefinite Symmetric Bilinear Form: Cauchy-Schwarz, the Space of Cauchy Sequences, Convergence of Pairings, and Null Sequences. By (∗)(\ast), the second hypothesis and (P1),

β(iu,iv)=Re⁡(i‾ i h(u,v))=β(u,v),β(u,iv)=Re⁡(i h(u,v))=−Im⁡h(u,v),β(iu,v)=Re⁡((−i) h(u,v))=Im⁡h(u,v),\beta(iu,iv)=\operatorname{Re}\bigl(\overline{i}\,i\,h(u,v)\bigr)=\beta(u,v),\qquad\beta(u,iv)=\operatorname{Re}\bigl(i\,h(u,v)\bigr)=-\operatorname{Im}h(u,v),\qquad\beta(iu,v)=\operatorname{Re}\bigl((-i)\,h(u,v)\bigr)=\operatorname{Im}h(u,v),

so β(u,iv)=−β(iu,v)\beta(u,iv)=-\beta(iu,v), and h(u,v)=Re⁡h(u,v)+(Im⁡h(u,v)) i=β(u,v)−i β(u,iv)h(u,v)=\operatorname{Re}h(u,v)+(\operatorname{Im}h(u,v))\,i=\beta(u,v)-i\,\beta(u,iv).

Now let X=VX=V and h=⟨⋅,⋅⟩Vh=\langle\cdot,\cdot\rangle_{V}; the hypotheses on hh are conditions 2, 3, 1 and 4 of Complex Inner Product Space. Conditions (a), (b), (c) of Real Inner Product Space §inner-product are the symmetry of β\beta and the linearity of u↦β(u,v)u\mapsto\beta(u,v) proved above, and (d) holds because β(x,x)=⟨x,x⟩V≥0\beta(x,x)=\langle x,x\rangle_{V}\ge0, which vanishes only for x=0x=0 by condition 4 of Complex Inner Product Space. So β=⟨⋅,⋅⟩V,R\beta=\langle\cdot,\cdot\rangle_{V,\mathbb{R}} is an inner product. Its norm ∣x∣|x| is the nonnegative real number with ∣x∣2=⟨x,x⟩V,R=⟨x,x⟩V|x|^{2}=\langle x,x\rangle_{V,\mathbb{R}}=\langle x,x\rangle_{V}, and ∥x∥V\lVert x\rVert_{V} is a nonnegative real number with the same square by Norm Induced by a Complex Inner Product, so ∣x∣=∥x∥V|x|=\lVert x\rVert_{V} by the uniqueness in Existence and Uniqueness of the Nonnegative Square Root. The zero vector and additive inverses are determined by the addition alone (claims 1 and 2 of Elementary Identities in a Vector Space), so x−yx-y is the same vector in both structures, and the distance ∣x−y∣|x-y| equals ∥x−y∥V\lVert x-y\rVert_{V}, the metric of Complex Hilbert Space. Being Cauchy (Cauchy Sequence in a Metric Space) or convergent (Convergent Sequence in a Metric Space) refers only to the metric, so the two metric spaces are complete together (Complete Metric Space); hence if VV is a complex Hilbert space, VV with ⟨⋅,⋅⟩V,R\langle\cdot,\cdot\rangle_{V,\mathbb{R}} is a real Hilbert space.

Claim 4 (Uniqueness of adjoints). Let S,S′:W→VS,S':W\to V be adjoints of TT. For w∈Ww\in W, Adjoint of a Linear Map between Complex Inner Product Spaces §adjoint gives ⟨Sw,v⟩=⟨w,Tv⟩=⟨S′w,v⟩\langle Sw,v\rangle=\langle w,Tv\rangle=\langle S'w,v\rangle for every v∈Vv\in V, so Sw=S′wSw=S'w by (P3). Hence S=S′S=S'.

Claim 5 (Existence of adjoints). Let VV be a complex Hilbert space and T∈L(V,W)T\in\mathcal{L}(V,W). By claim 3, VV with ⟨⋅,⋅⟩V,R=Re⁡⟨⋅,⋅⟩V\langle\cdot,\cdot\rangle_{V,\mathbb{R}}=\operatorname{Re}\langle\cdot,\cdot\rangle_{V} is a real Hilbert space with norm ∥⋅∥V\lVert\cdot\rVert_{V}. Fix w∈Ww\in W and put ℓw(v)=Re⁡⟨w,Tv⟩W\ell_{w}(v)=\operatorname{Re}\langle w,Tv\rangle_{W} for v∈Vv\in V. For v,v′∈Vv,v'\in V and a∈Ra\in\mathbb{R}, linearity of TT, conditions 2 and 3 of Complex Inner Product Space and (P1) give ℓw(v+v′)=ℓw(v)+ℓw(v′)\ell_{w}(v+v')=\ell_{w}(v)+\ell_{w}(v') and ℓw(av)=Re⁡(a⟨w,Tv⟩)=a ℓw(v)\ell_{w}(av)=\operatorname{Re}\bigl(a\langle w,Tv\rangle\bigr)=a\,\ell_{w}(v); and by (P1), (P2) and claim 1,

∣ℓw(v)∣≤∣⟨w,Tv⟩∣≤∥w∥ ∥Tv∥≤∥w∥ ∥T∥op∥v∥.|\ell_{w}(v)|\le|\langle w,Tv\rangle|\le\lVert w\rVert\,\lVert Tv\rVert\le\lVert w\rVert\,\lVert T\rVert_{\mathrm{op}}\lVert v\rVert .

So ℓw\ell_{w} is a bounded linear functional on this real Hilbert space, and by The Riesz Representation Theorem for a Real Hilbert Space §existence we may choose zw∈Vz_{w}\in V with ℓw(v)=⟨v,zw⟩V,R\ell_{w}(v)=\langle v,z_{w}\rangle_{V,\mathbb{R}} for every v∈Vv\in V; by the symmetry in claim 3, ℓw(v)=Re⁡⟨zw,v⟩V\ell_{w}(v)=\operatorname{Re}\langle z_{w},v\rangle_{V}. Applying the identity h(u,v)=β(u,v)−i β(u,iv)h(u,v)=\beta(u,v)-i\,\beta(u,iv) of claim 3 first to ⟨⋅,⋅⟩V\langle\cdot,\cdot\rangle_{V} and then to ⟨⋅,⋅⟩W\langle\cdot,\cdot\rangle_{W}, and using T(iv)=i TvT(iv)=i\,Tv,

⟨zw,v⟩V=ℓw(v)−i ℓw(iv)=Re⁡⟨w,Tv⟩W−iRe⁡⟨w,i Tv⟩W=⟨w,Tv⟩W(v∈V).\langle z_{w},v\rangle_{V}=\ell_{w}(v)-i\,\ell_{w}(iv)=\operatorname{Re}\langle w,Tv\rangle_{W}-i\operatorname{Re}\langle w,i\,Tv\rangle_{W}=\langle w,Tv\rangle_{W}\qquad(v\in V).

Define T∗:W→VT^{*}:W\to V by T∗w=zwT^{*}w=z_{w} (one choice for each ww). For w,w′∈Ww,w'\in W, a∈Ca\in\mathbb{C} and v∈Vv\in V, claims 1 and 2 of Elementary Properties of a Complex Inner Product give

⟨T∗(w+w′),v⟩=⟨w,Tv⟩+⟨w′,Tv⟩=⟨T∗w+T∗w′,v⟩,⟨T∗(aw),v⟩=a‾ ⟨w,Tv⟩=⟨a T∗w,v⟩,\langle T^{*}(w+w'),v\rangle=\langle w,Tv\rangle+\langle w',Tv\rangle=\langle T^{*}w+T^{*}w',v\rangle,\qquad\langle T^{*}(aw),v\rangle=\overline{a}\,\langle w,Tv\rangle=\langle a\,T^{*}w,v\rangle,

so T∗(w+w′)=T∗w+T∗w′T^{*}(w+w')=T^{*}w+T^{*}w' and T∗(aw)=a T∗wT^{*}(aw)=a\,T^{*}w by (P3). Thus T∗T^{*} is linear and, by the displayed identity, an adjoint of TT in the sense of Adjoint of a Linear Map between Complex Inner Product Spaces §adjoint; by claim 4 it is the adjoint.

Bounds. Let w∈Ww\in W. The real number ∥T∗w∥2=⟨T∗w,T∗w⟩V=⟨w,T(T∗w)⟩W\lVert T^{*}w\rVert^{2}=\langle T^{*}w,T^{*}w\rangle_{V}=\langle w,T(T^{*}w)\rangle_{W} is nonnegative, so it equals its modulus by (P1), and by (P2) and claim 1

∥T∗w∥2≤∥w∥ ∥T(T∗w)∥≤∥w∥ ∥T∥op∥T∗w∥.\lVert T^{*}w\rVert^{2}\le\lVert w\rVert\,\lVert T(T^{*}w)\rVert\le\lVert w\rVert\,\lVert T\rVert_{\mathrm{op}}\lVert T^{*}w\rVert .

If T∗w=0T^{*}w=0, then ∥T∗w∥=0≤∥T∥op∥w∥\lVert T^{*}w\rVert=0\le\lVert T\rVert_{\mathrm{op}}\lVert w\rVert; otherwise 0<∥T∗w∥0<\lVert T^{*}w\rVert by (P2), and dividing gives ∥T∗w∥≤∥T∥op∥w∥\lVert T^{*}w\rVert\le\lVert T\rVert_{\mathrm{op}}\lVert w\rVert. Hence ∥T∥op≥0\lVert T\rVert_{\mathrm{op}}\ge0 is a bound for T∗T^{*}, so T∗∈L(W,V)T^{*}\in\mathcal{L}(W,V) and ∥T∗∥op≤∥T∥op\lVert T^{*}\rVert_{\mathrm{op}}\le\lVert T\rVert_{\mathrm{op}} by claim 1. Conversely, for v∈Vv\in V the adjoint identity with w=Tvw=Tv gives ∥Tv∥2=⟨Tv,Tv⟩W=⟨T∗(Tv),v⟩V\lVert Tv\rVert^{2}=\langle Tv,Tv\rangle_{W}=\langle T^{*}(Tv),v\rangle_{V}, and in the same way ∥Tv∥2≤∥T∗(Tv)∥ ∥v∥≤∥T∗∥op∥Tv∥ ∥v∥\lVert Tv\rVert^{2}\le\lVert T^{*}(Tv)\rVert\,\lVert v\rVert\le\lVert T^{*}\rVert_{\mathrm{op}}\lVert Tv\rVert\,\lVert v\rVert, whence ∥Tv∥≤∥T∗∥op∥v∥\lVert Tv\rVert\le\lVert T^{*}\rVert_{\mathrm{op}}\lVert v\rVert (trivially if Tv=0Tv=0). By claim 1, ∥T∥op≤∥T∗∥op\lVert T\rVert_{\mathrm{op}}\le\lVert T^{*}\rVert_{\mathrm{op}}, so the two norms are equal.

Claim 6 (Calculus of adjoints). All maps below are linear by the first paragraph of claim 2, and each identity is checked against Adjoint of a Linear Map between Complex Inner Product Spaces §adjoint; by claim 4 the adjoint found is the adjoint. Let v∈Vv\in V, w∈Ww\in W and x∈Ux\in U. By condition 1 of Complex Inner Product Space, ⟨Tv,w⟩=⟨w,Tv⟩‾=⟨T∗w,v⟩‾=⟨v,T∗w⟩\langle Tv,w\rangle=\overline{\langle w,Tv\rangle}=\overline{\langle T^{*}w,v\rangle}=\langle v,T^{*}w\rangle, so TT is the adjoint of T∗:W→VT^{*}:W\to V. By claims 1 and 2 of Elementary Properties of a Complex Inner Product, conditions 2 and 3 of Complex Inner Product Space and c‾‾=c\overline{\overline{c}}=c,

⟨(S∗+T∗)w,v⟩=⟨w,Sv⟩+⟨w,Tv⟩=⟨w,(S+T)v⟩,⟨c‾ T∗w,v⟩=c ⟨w,Tv⟩=⟨w,(cT)v⟩,\langle(S^{*}+T^{*})w,v\rangle=\langle w,Sv\rangle+\langle w,Tv\rangle=\langle w,(S+T)v\rangle,\qquad\langle\overline{c}\,T^{*}w,v\rangle=c\,\langle w,Tv\rangle=\langle w,(cT)v\rangle, ⟨T∗(R∗x),v⟩V=⟨R∗x,Tv⟩W=⟨x,R(Tv)⟩U,⟨IVv′,v⟩=⟨v′,IVv⟩(v′∈V),\langle T^{*}(R^{*}x),v\rangle_{V}=\langle R^{*}x,Tv\rangle_{W}=\langle x,R(Tv)\rangle_{U},\qquad\langle I_{V}v',v\rangle=\langle v',I_{V}v\rangle\quad(v'\in V),

so S+TS+T, cTcT, RTRT and IVI_{V} have the adjoints S∗+T∗S^{*}+T^{*}, c‾ T∗\overline{c}\,T^{*}, T∗R∗T^{*}R^{*} and IVI_{V}. For a linear map A:V→VA:V\to V, which is a linear operator, the condition of Self-Adjoint Operator (⟨Au,v⟩=⟨u,Av⟩\langle Au,v\rangle=\langle u,Av\rangle for all u,v∈Vu,v\in V) is literally the condition of Adjoint of a Linear Map between Complex Inner Product Spaces §adjoint for AA to be an adjoint of AA (with W=VW=V), so the two are equivalent. Next, by condition 1 of Complex Inner Product Space, the adjoint identity and (P1),

⟨v,T∗Tv⟩V=⟨T∗(Tv),v⟩V‾=⟨Tv,Tv⟩W‾=∥Tv∥W2‾=∥Tv∥W2,\langle v,T^{*}Tv\rangle_{V}=\overline{\langle T^{*}(Tv),v\rangle_{V}}=\overline{\langle Tv,Tv\rangle_{W}}=\overline{\lVert Tv\rVert_{W}^{2}}=\lVert Tv\rVert_{W}^{2},

the last number being real. Finally, the computation for RTRT uses only that the three spaces are complex inner product spaces; applied to T:V→WT:V\to W and to T∗:W→VT^{*}:W\to V (in the roles of TT and RR, with VV in the role of UU; the adjoint of T∗T^{*} is TT), it shows that T∗TT^{*}T has the adjoint T∗TT^{*}T, so T∗TT^{*}T is self-adjoint by the equivalence just proved; and ⟨v,T∗Tv⟩=∥Tv∥2\langle v,T^{*}Tv\rangle=\lVert Tv\rVert^{2} is a nonnegative real number for every vv, so T∗TT^{*}T is positive semi-definite.

Claim 7 (Completeness). Let WW be a complex Hilbert space and (Tk)(T_{k}) as stated. By claim 2 each Tk−TlT_{k}-T_{l} lies in L(V,W)\mathcal{L}(V,W), and (Tk−Tl)v=Tkv−Tlv(T_{k}-T_{l})v=T_{k}v-T_{l}v by claim 5 of Elementary Identities in a Vector Space, so by claim 1

∥Tkv−Tlv∥≤∥Tk−Tl∥op∥v∥(k,l∈N, v∈V).(†)\lVert T_{k}v-T_{l}v\rVert\le\lVert T_{k}-T_{l}\rVert_{\mathrm{op}}\lVert v\rVert\qquad(k,l\in\mathbb{N},\ v\in V).\qquad(\dagger)

Pointwise limit. Fix v∈Vv\in V. Given ε>0\varepsilon>0, choose NN from the hypothesis for ε/(∥v∥+1)\varepsilon/(\lVert v\rVert+1); then for k,l≥Nk,l\ge N, (†)(\dagger) gives ∥Tkv−Tlv∥≤ε∥v∥/(∥v∥+1)<ε\lVert T_{k}v-T_{l}v\rVert\le\varepsilon\lVert v\rVert/(\lVert v\rVert+1)<\varepsilon. So (Tkv)(T_{k}v) is a Cauchy sequence in WW, which converges because WW is complete (Complex Hilbert Space, Complete Metric Space); let TvTv be its limit, unique by Uniqueness of Limits in a Metric Space.

TT is linear. Let u,u′∈Vu,u'\in V and λ∈C\lambda\in\mathbb{C}. Given ε>0\varepsilon>0, choose NN so large that ∥Tku−Tu∥<ε/2\lVert T_{k}u-Tu\rVert<\varepsilon/2, ∥Tku′−Tu′∥<ε/2\lVert T_{k}u'-Tu'\rVert<\varepsilon/2 and ∥Tku−Tu∥<ε/(∣λ∣+1)\lVert T_{k}u-Tu\rVert<\varepsilon/(|\lambda|+1) for k≥Nk\ge N. Then for k≥Nk\ge N, by (P2),

∥Tk(u+u′)−(Tu+Tu′)∥≤∥Tku−Tu∥+∥Tku′−Tu′∥<ε,∥Tk(λu)−λ Tu∥=∣λ∣ ∥Tku−Tu∥<ε.\lVert T_{k}(u+u')-(Tu+Tu')\rVert\le\lVert T_{k}u-Tu\rVert+\lVert T_{k}u'-Tu'\rVert<\varepsilon,\qquad\lVert T_{k}(\lambda u)-\lambda\,Tu\rVert=|\lambda|\,\lVert T_{k}u-Tu\rVert<\varepsilon .

So (Tk(u+u′))(T_{k}(u+u')) converges to Tu+Tu′Tu+Tu' and (Tk(λu))(T_{k}(\lambda u)) to λ Tu\lambda\,Tu; since they converge to T(u+u′)T(u+u') and T(λu)T(\lambda u) by definition, Uniqueness of Limits in a Metric Space gives T(u+u′)=Tu+Tu′T(u+u')=Tu+Tu' and T(λu)=λ TuT(\lambda u)=\lambda\,Tu.

Uniform estimate. Let ε>0\varepsilon>0 and choose NN from the hypothesis for ε\varepsilon. Fix k≥Nk\ge N and v∈Vv\in V, and let δ>0\delta>0. Choose MM with ∥Tlv−Tv∥<δ\lVert T_{l}v-Tv\rVert<\delta for l≥Ml\ge M, and put l=max⁡(N,M)l=\max(N,M). By (P2) and (†)(\dagger),

∥Tkv−Tv∥≤∥Tkv−Tlv∥+∥Tlv−Tv∥≤ε∥v∥+δ.\lVert T_{k}v-Tv\rVert\le\lVert T_{k}v-T_{l}v\rVert+\lVert T_{l}v-Tv\rVert\le\varepsilon\lVert v\rVert+\delta .

As δ>0\delta>0 was arbitrary, Comparison of Real Numbers with Arbitrary Positive Slack §slack-above gives ∥(Tk−T)v∥=∥Tkv−Tv∥≤ε∥v∥\lVert(T_{k}-T)v\rVert=\lVert T_{k}v-Tv\rVert\le\varepsilon\lVert v\rVert. So ε\varepsilon is a bound for the linear map Tk−TT_{k}-T (linear by claim 2), whence Tk−T∈L(V,W)T_{k}-T\in\mathcal{L}(V,W) and ∥Tk−T∥op≤ε\lVert T_{k}-T\rVert_{\mathrm{op}}\le\varepsilon for all k≥Nk\ge N by claim 1. In particular TN−T∈L(V,W)T_{N}-T\in\mathcal{L}(V,W), and T=TN+(−1)(TN−T)T=T_{N}+(-1)(T_{N}-T) (evaluate at each vv), so T∈L(V,W)T\in\mathcal{L}(V,W) by claim 2. Given ε>0\varepsilon>0, the above with ε/2\varepsilon/2 yields NN with ∣∥Tk−T∥op−0∣=∥Tk−T∥op≤ε/2<ε\bigl|\lVert T_{k}-T\rVert_{\mathrm{op}}-0\bigr|=\lVert T_{k}-T\rVert_{\mathrm{op}}\le\varepsilon/2<\varepsilon for k≥Nk\ge N, so (∥Tk−T∥op)\bigl(\lVert T_{k}-T\rVert_{\mathrm{op}}\bigr) converges to 00 in the sense of Limit of a Sequence of Real Numbers.

Pointwise convergence for every such TT. Let T∈L(V,W)T\in\mathcal{L}(V,W) with (∥Tk−T∥op)\bigl(\lVert T_{k}-T\rVert_{\mathrm{op}}\bigr) converging to 00, and let v∈Vv\in V and ε>0\varepsilon>0. Choose NN with ∥Tk−T∥op<ε/(∥v∥+1)\lVert T_{k}-T\rVert_{\mathrm{op}}<\varepsilon/(\lVert v\rVert+1) for k≥Nk\ge N. Then claim 1 gives, for k≥Nk\ge N, d(Tkv,Tv)=∥(Tk−T)v∥≤∥Tk−T∥op∥v∥<εd(T_{k}v,Tv)=\lVert(T_{k}-T)v\rVert\le\lVert T_{k}-T\rVert_{\mathrm{op}}\lVert v\rVert<\varepsilon, so (Tkv)(T_{k}v) converges to TvTv in WW (Convergent Sequence in a Metric Space).

Claim 8 (Quadratic-form bound). Let AA and cc be as stated; AA is a linear operator and self-adjoint as in Self-Adjoint Operator. For x∈Vx\in V, ⟨x,Ax⟩=⟨Ax,x⟩=⟨x,Ax⟩‾\langle x,Ax\rangle=\langle Ax,x\rangle=\overline{\langle x,Ax\rangle} by self-adjointness and condition 1 of Complex Inner Product Space, so ⟨x,Ax⟩\langle x,Ax\rangle is real by (P1), and ⟨x,Ax⟩≤c∥x∥2\langle x,Ax\rangle\le c\lVert x\rVert^{2} and −⟨x,Ax⟩≤c∥x∥2-\langle x,Ax\rangle\le c\lVert x\rVert^{2} because its modulus is ⟨x,Ax⟩\langle x,Ax\rangle or −⟨x,Ax⟩-\langle x,Ax\rangle (P1).

Expansion. For a linear B:V→VB:V\to V and u,w∈Vu,w\in V, condition 2 of Complex Inner Product Space and claim 1 of Elementary Properties of a Complex Inner Product give the first identity below; writing u−w=u+(−1)wu-w=u+(-1)w and B(u−w)=Bu+(−1)BwB(u-w)=Bu+(-1)Bw, condition 3 of Complex Inner Product Space, claim 2 of Elementary Properties of a Complex Inner Product and −1‾=−1\overline{-1}=-1 give the second:

⟨u+w,B(u+w)⟩=⟨u,Bu⟩+⟨u,Bw⟩+⟨w,Bu⟩+⟨w,Bw⟩,⟨u−w,B(u−w)⟩=⟨u,Bu⟩−⟨u,Bw⟩−⟨w,Bu⟩+⟨w,Bw⟩.\langle u+w,B(u+w)\rangle=\langle u,Bu\rangle+\langle u,Bw\rangle+\langle w,Bu\rangle+\langle w,Bw\rangle,\qquad\langle u-w,B(u-w)\rangle=\langle u,Bu\rangle-\langle u,Bw\rangle-\langle w,Bu\rangle+\langle w,Bw\rangle .

With B=AB=A, subtracting, and using ⟨w,Au⟩=⟨Aw,u⟩=⟨u,Aw⟩‾\langle w,Au\rangle=\langle Aw,u\rangle=\overline{\langle u,Aw\rangle} (self-adjointness, condition 1 of Complex Inner Product Space) and claim 2 of Properties of Complex Conjugation and Modulus,

⟨u+w,A(u+w)⟩−⟨u−w,A(u−w)⟩=2(⟨u,Aw⟩+⟨u,Aw⟩‾)=4Re⁡⟨u,Aw⟩.\langle u+w,A(u+w)\rangle-\langle u-w,A(u-w)\rangle=2\bigl(\langle u,Aw\rangle+\overline{\langle u,Aw\rangle}\bigr)=4\operatorname{Re}\langle u,Aw\rangle .

With B=IVB=I_{V}, adding, and (P2): ∥u+w∥2+∥u−w∥2=2∥u∥2+2∥w∥2\lVert u+w\rVert^{2}+\lVert u-w\rVert^{2}=2\lVert u\rVert^{2}+2\lVert w\rVert^{2}. By the first paragraph applied to x=u+wx=u+w and to x=u−wx=u-w,

4Re⁡⟨u,Aw⟩≤c∥u+w∥2+c∥u−w∥2=2c(∥u∥2+∥w∥2)(u,w∈V).4\operatorname{Re}\langle u,Aw\rangle\le c\lVert u+w\rVert^{2}+c\lVert u-w\rVert^{2}=2c\bigl(\lVert u\rVert^{2}+\lVert w\rVert^{2}\bigr)\qquad(u,w\in V).

Now fix w∈Vw\in V. If Aw=0Aw=0, then ∥Aw∥=0≤c∥w∥\lVert Aw\rVert=0\le c\lVert w\rVert. Otherwise w≠0w\neq0 (as A0=0A0=0 by (P2)), so 0<∥Aw∥0<\lVert Aw\rVert and 0<∥w∥0<\lVert w\rVert; put s=∥w∥/∥Aw∥>0s=\lVert w\rVert/\lVert Aw\rVert>0 and u=s Awu=s\,Aw. By claim 2 of Elementary Properties of a Complex Inner Product, (P1) and (P2), ⟨u,Aw⟩=s∥Aw∥2=∥w∥ ∥Aw∥\langle u,Aw\rangle=s\lVert Aw\rVert^{2}=\lVert w\rVert\,\lVert Aw\rVert, a real number, so Re⁡⟨u,Aw⟩=∥w∥ ∥Aw∥\operatorname{Re}\langle u,Aw\rangle=\lVert w\rVert\,\lVert Aw\rVert, and ∥u∥=∣s∣ ∥Aw∥=∥w∥\lVert u\rVert=|s|\,\lVert Aw\rVert=\lVert w\rVert. The last inequality becomes 4∥w∥ ∥Aw∥≤4c∥w∥24\lVert w\rVert\,\lVert Aw\rVert\le4c\lVert w\rVert^{2}, and dividing by 4∥w∥>04\lVert w\rVert>0 gives ∥Aw∥≤c∥w∥\lVert Aw\rVert\le c\lVert w\rVert. Hence the real number c≥0c\ge0 is a bound for AA, so A∈L(V)A\in\mathcal{L}(V), and ∥A∥op≤c\lVert A\rVert_{\mathrm{op}}\le c by claim 1.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…