Alphabeta Math
Pipeline-generated
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Banach Algebras Spectrum and Holomorphic Functional Calculus

1 · Prerequisites

2 · Summary

This page develops the spectral theory of a single element of a unital complex Banach algebra, from the Neumann series to the holomorphic functional calculus. The opening definitions fix the conventions — a nonzero associative algebra with a complete submultiplicative norm and a unit of norm one, invertibility by two-sided inverses, and the spectrum as the set of scalars at which the shifted element fails to be invertible in the stated algebra. The Neumann series makes the open unit ball around each invertible element consist of invertible elements and gives the quantitative bound b1a1/(1a1ba), from which openness of the general linear group and continuity of inversion follow.

The spectrum itself is then shown to be a nonempty compact subset of the disc of radius a: boundedness and closedness come from the Neumann expansion and from openness of the invertibles, while nonemptiness is proved by applying bounded functionals to the resolvent, invoking Liouville's theorem for the scalar case and separating points in the dual. The spectral radius is defined as the maximal modulus on this compact nonempty set, and the spectral radius formula r(a)=limnan1/n=infnan1/n is proved by resolving the resolvent into a power series on each disc properly inside the disc of radius 1/r(a) and applying Cauchy's coefficient estimate together with the dual unit-ball formula for the norm. Polynomial spectral mapping σ(p(a))=p(σ(a)) is proved separately, and the real-operator spectrum is defined through the canonical rotation-supremum complexification, with the bounded comparison between compatible models recorded so that the definition is model-independent.

The second half builds the holomorphic functional calculus. Banach-algebra-valued contour integrals are constructed from tagged Riemann sums and identified with the Bochner integral; the vector-valued version of the homology form of Cauchy's theorem is proved by scalarisation and dual separation. A finite polygonal cycle with index one on a compact set and zero outside a prescribed open neighbourhood is constructed from a grid, together with a nested pair with separated traces. These are the cycles along which the Dunford integral f(a)=12πiΓf(z)R(z,a)dz is defined; contour and germ independence are proved before the notation is used, and the calculus is shown to be a unital algebra homomorphism that reproduces polynomials and reciprocals of nonvanishing functions, to satisfy the holomorphic spectral mapping theorem σ(f(a))=f(σ(a)) and the composition law g(f(a))=(gf)(a), and to produce Riesz projections for clopen spectral subsets, with the invariant splitting X=ran(PE)ker(PE) and the restriction spectra E and σ(T)E on the corresponding nonzero summands; an empty spectral part gives the zero summand, to which the page's nonzero-algebra convention assigns no spectrum.

The page closes with the Calkin algebra B(X)/K(X) and Atkinson's theorem in quotient language, and with the five spectral parts: the disjoint point, continuous and residual spectra, the approximate point and compression spectra, the identity σr=σcpσp, the covering relation σ=σapσcp under Dependent Choice, and the theorem that the boundary of the spectrum lies in the approximate point spectrum. Full choice strength is stated wherever it is used: the nonemptiness of the spectrum, the spectral radius formula and the calculus spend the Axiom of Choice through Hahn–Banach separation and compact spectrum nonemptiness, the quotient completeness of the Calkin algebra uses Countable Choice, and the bounded-inverse and approximate-pointer arguments use Dependent Choice and Countable Choice respectively.

3 · Logical flowchart

4 · Definitions, theorems and proofs

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Unital Banach algebra

Definition

A unital complex Banach algebra is a nonzero complex vector space A equipped with

  • an associative bilinear multiplication A×AA, (a,b)ab, and
  • a norm under which A is a Banach space (Banach space),

such that

  • the norm is submultiplicative: abab for all a,bA, and
  • there is a unit 1A with 1a=a1=a for every aA, normalized by 1=1.

The phrase nonzero is part of the definition: the identity is required to exist, and the zero algebra {0} has no element satisfying 10. In every unital Banach algebra the unit is unique, because if 1 is a second element acting as an identity then 1=11=1; from now on 1 denotes that element. The norm condition 1=1 is a normalization rather than a consequence of submultiplicativity, which would give only 11 for a nonzero unit; the two conventions 1=1 and 11 agree on every nonzero unital algebra, since submultiplicativity turns the latter into an equality.

Remarks

  • The scalar field is complex and fixed. Every spectrum, resolvent and holomorphic-calculus statement on this page is about complex unital Banach algebras. Real Banach algebras are not silently complexified: the one place where a real structure enters, the spectrum of a real operator, is defined through a specified complexification in Complexification and spectrum of a real operator.

  • Multiplication is bilinear and associative, and nothing more. No commutativity, involution, or approximate unit is assumed. The algebra B(X) of bounded operators on a nonzero complex Banach space (ex-bounded-operators-form-a-noncommutative-banach-algebra) is the motivating noncommutative example, and C(K,C) for compact Hausdorff K (ex-continuous-functions-form-a-commutative-banach-algebra) the motivating commutative one.

  • Completeness is with respect to the submultiplicative norm. A complete normed algebra whose norm is merely equivalent to a submultiplicative one is not thereby a unital Banach algebra in this sense; rescaling a norm to λ with λ>1 preserves completeness and submultiplicativity but destroys the normalization 1=1.

  • The unit is not a separate structure. It is determined by the multiplication, so an algebra homomorphism between unital Banach algebras that preserves multiplication and the unit is exactly a multiplicative linear map sending 1 to 1; this is the convention used for characters on the following page of this track.

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Invertible element and general linear group of a Banach algebra

Definition

Let A be a unital complex Banach algebra (Unital Banach algebra). An element aA is invertible when there is bA with

ab=ba=1.

Such an element b is then unique: if b and b both satisfy the two equations, then

b=b1=b(ab)=(ba)b=1b=b,

using associativity and the unit law. The unique b is called the inverse of a and is written a1. The set of all invertible elements of A is denoted

A×:={aA:a is invertible}

and is called the general linear group of A. It is a group under multiplication:

  • 1A× with 11=1;
  • if a,bA× then abA× with (ab)1=b1a1, since (ab)(b1a1)=a(bb1)a1=aa1=1 and symmetrically;
  • (a1)1=a by symmetry of the defining equations.

The map aa1 is the inversion map of A×.

Remarks

  • Two-sided inverses are required, and one-sided inverses do not suffice. If ab=1 and ba1, then a is not invertible by definition, and this situation really occurs in a unital complex Banach algebra: on 2(N0) the right shift Ten=en1 for n1, Te0=0, and the left shift Sen=en+1 satisfy TS=1 while ST=1P, where P is the orthogonal projection onto Ce0, so T has a right inverse and is not invertible (ex-spectrum-of-the-unilateral-shift). Thus ab=1 alone does not force ba=1 in a general unital Banach algebra; both inverses are always verified explicitly below.

  • The group need not be dense or connected, but it is open. In a Banach algebra A× is an open subset of A and inversion is continuous there; this is Invertible group is open and inversion is continuous. Openness is what makes the resolvent set of an element open and hence makes the spectrum closed Spectrum and resolvent set in a Banach algebra.

  • Nonunital algebras are not covered here. In a Banach algebra without a unit there is no element 1 to compare with, so invertibility is not defined by this definition. The companion examples introduce the unitization AC1 and declare that spectra of elements of a nonunital algebra are always computed in that named unitization (ex-unitization-of-a-nonunital-banach-algebra).

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Neumann series

Statement

Let A be a unital complex Banach algebra, let aA with a<1, and for NN write SN:=n=0Nan, a finite sum with a0:=1 (Unital Banach algebra). Then

  1. the series n0an converges in A, and its sum S:=limNSN satisfies (1a)S=S(1a)=1; in particular 1a is invertible with (1a)1=S (Invertible element and general linear group of a Banach algebra);
  2. for every NN the tail estimate (1a)1SN    aN+11a holds. The hypothesis a<1 is not symmetric: the estimate is in terms of a, and 1a is invertible whenever a lies in the open unit ball.

Facts & Assumptions

Given: A unital complex Banach algebra A, an element aA with a<1, the partial sums SN=n=0Nan, and the number q:=a[0,1).

[L1]

A is complete under its norm, 1=1, and xyxy for all x,yA; multiplication is associative and bilinear (Unital Banach algebra).

[L2]

An element cA is invertible exactly when there is bA with cb=bc=1, and that b is then unique, written c1 (Invertible element and general linear group of a Banach algebra).

[L3]

A normed space V is a Banach space if and only if every absolutely convergent series in V converges (Series criterion for Banach spaces).

Proof

technique · direct
1.1

For every nN one has anan=qn: this holds at n=0 because a0=1=1=q0, and inductively an+1=anaanaqnq=qn+1.

L1algebra
1.2

For every NN the telescoping identities (1a)SN=1aN+1 and SN(1a)=1aN+1 hold, by distributivity and anan+1=an(1a) summed over 0nN.

L1algebra
1.3

Multiplication is jointly continuous in the norm: for x,x,y,yA one has xyxyxyy+xxy by [L1], so xx and yy force xyxy.

L1algebra
2.1

Since q[0,1), the geometric series satisfies n0qn=1/(1q) and its tails satisfy n>Nqn=qN+1/(1q); with [step 1.1] this gives n0an1/(1q)<.

step 1.1algebra
3.1

The series n0an is absolutely convergent, so it converges to an element S=limNSNA by [L3] and completeness of A.

L3step 2.1L1
4.1

Since aN+10 in A by [step 1.1] and qN+10, letting N in the identities of [step 1.2] is legitimate: (1a)SN(1a)S and SN(1a)S(1a) by [step 1.3] and [step 3.1], while the right hand sides 1aN+1 tend to 1; hence (1a)S=1 and S(1a)=1.

step 1.2step 1.3step 3.1step 1.1
5.1

By [L2] the element 1a is invertible with (1a)1=S=n0an, which proves claim 1; moreover for every N the difference of the sum and the partial sum is the tail (1a)1SN=n>Nan, whose norm is at most n>Nqn=qN+1/(1q) by [step 1.1] and [step 2.1], which is claim 2.

step 4.1step 2.1L2algebra
TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Invertible group is open and inversion is continuous

Statement

Let A be a unital complex Banach algebra (Unital Banach algebra) and let aA× be invertible with inverse a1 (Invertible element and general linear group of a Banach algebra). Then:

  1. every bA with a1ba<1 is invertible, with b1=(1a1(ab))1a1andb1a11a1ba;
  2. A× is an open subset of A;
  3. inversion A×A×, aa1, is continuous at every point of A× (with the relative topology on A×).

Facts & Assumptions

Given: A unital complex Banach algebra A, an invertible aA×, and an element bA with a1ba<1. Put x:=a1(ab), so that b=a(1x).

[L1]

The norm on A is submultiplicative, 1=1, multiplication is associative and bilinear, and the norm is continuous with respect to itself: uvuv and uuuu+uu (Unital Banach algebra).

[L2]

An element cA is invertible exactly when it has a two-sided inverse, which is then unique; inverses satisfy (uv)1=v1u1 for invertible u,v, and (u1)1=u (Invertible element and general linear group of a Banach algebra).

[L3]

If y<1 then 1y is invertible with (1y)1=n0yn and every tail bound (1y)1nNynyN+1/(1y); in particular (1y)11/(1y) (Neumann series).

Proof

technique · direct
1.1

The element x:=a1(ab) satisfies xa1ab=a1ba<1 by [L1], and b=a(ab)=aaa1(ab)=a(1x), where the middle step uses aa1=1 from [L2].

L1L2algebra
1.2

For the difference of inverses one has the algebraic identity b1a1=b1(ab)a1 whenever both inverses exist, because b1(ab)a1=b1aa1b1ba1=b1a1, using [L2].

L2L1algebra
2.1

By [L3] applied to x with x<1, the element 1x is invertible with (1x)1=n0xn and (1x)11/(1x)1/(1a1ba).

step 1.1L3
3.1

Since b=a(1x) with both factors invertible, [L2] gives that b is invertible with b1=(1x)1a1=(1a1(ab))1a1, and taking norms with [L1] and [step 2.1] gives b1(1x)1a1a1/(1a1ba); this is claim 1.

step 2.1L2L1algebra
4.1

Claim 2 follows: given aA×, every b with ba<1/a1 satisfies the hypothesis verified in [step 3.1] and hence lies in A×, so A× contains the open ball of that radius about a.

step 3.1L1
4.2

In particular, whenever ba<1/(2a1) the bound of [step 3.1] gives b1a1/(112)=2a1.

step 3.1algebra
5.1

Combining [step 1.2] with [step 4.2] and [L1], for ba<1/(2a1) one has b1a1b1baa12a12ba, which tends to 0 as ba; this is claim 3.

step 1.2step 4.2L1algebra
6.1

Claims 1, 2 and 3 are exactly the three assertions of the statement, so the theorem is proved.

step 3.1step 4.1step 5.1
DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Spectrum and resolvent set in a Banach algebra

Definition

Let A be a unital complex Banach algebra (Unital Banach algebra) and let aA. The spectrum of a in A is the set

σA(a)  :=  {zC  :  z1aA×},

where A× is the group of invertible elements (Invertible element and general linear group of a Banach algebra). The complement

ρA(a)  :=  CσA(a)  =  {zC:z1aA×}

is the resolvent set of a, and for zρA(a) the element

R(z,a)  :=  (z1a)1    A

is the resolvent of a at z. Thus R(z,a) is characterized by the two equations

(z1a)R(z,a)=R(z,a)(z1a)=1,

and it is the unique element with these properties. The subscript in σA(a) and ρA(a) records the ambient algebra: if BA is a closed unital subalgebra containing a and having the same unit, then B is itself a unital complex Banach algebra in the inherited norm, so both spectra are defined. They can differ, and the spectrum in the smaller algebra contains the spectrum in the larger one: σB(a)σA(a).

The zero algebra is excluded by the unital-algebra convention. For a=0, σA(0)={0}: z1 has inverse z11 for z0, whereas 0 cannot have a two-sided inverse since 10.

Remarks

  • The ambient algebra is always part of the data. Every use of a spectrum below states the algebra in which it is computed. For aA and a closed unital subalgebra BA with aB and the same unit, containment of the invertible groups B×A× gives σB(a)σA(a), and the companion page exhibits a case where the inclusion is strict (cex-spectrum-can-shrink-in-a-larger-banach-algebra). The complex spectrum of an element of a nonunital Banach algebra is not defined by this formula; it is taken in the unitization, as recorded in ex-unitization-of-a-nonunital-banach-algebra.

  • Operator spectra are the special case A=B(X). For a nonzero complex Banach space X and TB(X) one has z1TB(X) and σB(X)(T) agrees with the spectrum of T computed in any unital Banach subalgebra of B(X) that contains T and the identity and is closed under inverses of its elements. The closed unital algebra generated by T need not be inverse-closed, so no agreement with that algebra is asserted (ex-bounded-operators-form-a-noncommutative-banach-algebra).

  • The spectrum is closed and bounded; under AC it is nonempty. The map zz1a is continuous and A× is open (Invertible group is open and inversion is continuous), so ρA(a) is open and σA(a) is closed. The Neumann series gives {z:z>a}ρA(a) and hence bounds the spectrum by a; under the Axiom of Choice (The Axiom of Choice), the nonempty-spectrum conclusion is the substance of Spectrum is nonempty compact and norm bounded. Nothing in the present definition assumes either conclusion.

  • The resolvent convention fixed here is R(z,a)=(z1a)1. With this order of the factors the resolvent identity reads R(z,a)R(w,a)=(wz)R(z,a)R(w,a) and the derivative of the resolvent map is R(z,a)2 (Resolvent identity, Resolvent is Banach-valued holomorphic). For the opposite convention R~(z,a)=(az1)1=R(z,a), the identity has factor zw and the derivative is +R~(z,a)2.

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Resolvent identity

Statement

Let A be a unital complex Banach algebra and let a,bA. With the resolvent R(z,a)=(z1a)1 of Spectrum and resolvent set in a Banach algebra:

  1. for all z,wρA(a), R(z,a)R(w,a)=(wz)R(z,a)R(w,a); in particular R(z,a) and R(w,a) commute;
  2. for all zρA(a)ρA(b), R(z,a)R(z,b)=R(z,a)(ab)R(z,b).

Both identities are equalities of two-sided products; no commutativity of A is assumed, and the factor order shown is the one that is used later.

Facts & Assumptions

Given: A unital complex Banach algebra A, elements a,bA, complex numbers z,w with z,wρA(a) and zρA(b).

[L1]

The norm is submultiplicative, 1=1, and multiplication is associative, bilinear, and satisfies 1u=u1=u for all uA (Unital Banach algebra).

[L2]

For zρA(a) the resolvent R(z,a) is the unique element of A with (z1a)R(z,a)=R(z,a)(z1a)=1, and similarly for b; if u,v are invertible then (uv)1=v1u1 (Spectrum and resolvent set in a Banach algebra, Invertible element and general linear group of a Banach algebra).

Proof

technique · direct
1.1

Each resolvent is a two-sided inverse of its own argument: R(z,a)(z1a)=(z1a)R(z,a)=1, R(w,a)(w1a)=(w1a)R(w,a)=1 and R(z,b)(z1b)=(z1b)R(z,b)=1 by [L2]; and the scalar identities (z1a)(w1a)=(zw)1 and (z1a)(z1b)=ba are immediate. No commutativity between a and b, and no commutativity between R(z,a) and R(z,b), is claimed or needed: the two computations below multiply each resolvent against its own argument only.

L1L2algebra
2.1

Multiplying the identity (z1a)(w1a)=(zw)1 on the left by R(z,a) and on the right by R(w,a) yields R(z,a)(z1a)R(w,a)R(z,a)(w1a)R(w,a)=(zw)R(z,a)R(w,a); the two terms on the left equal R(w,a) and R(z,a) respectively, so R(w,a)R(z,a)=(zw)R(z,a)R(w,a), which is claim 1.

step 1.1L2algebra
2.2

Multiplying the identity (z1a)(z1b)=ba on the left by R(z,a) and on the right by R(z,b) yields R(z,a)(z1a)R(z,b)R(z,a)(z1b)R(z,b)=R(z,a)(ba)R(z,b); the two terms on the left equal R(z,b) and R(z,a) respectively, so R(z,b)R(z,a)=R(z,a)(ba)R(z,b), which is claim 2 in the stated form after moving the term and reversing the sign: R(z,a)R(z,b)=R(z,a)(ab)R(z,b).

step 1.1L2algebra
3.1

Claim 1 and claim 2 are exactly the two displayed identities of the statement, so the lemma is proved.

step 2.1step 2.2
TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Resolvent is Banach-valued holomorphic

Statement

Let A be a unital complex Banach algebra and let aA with resolvent set ρA(a) and resolvent R(z,a)=(z1a)1 (Spectrum and resolvent set in a Banach algebra). Then:

  1. ρA(a) is an open subset of C, so σA(a) is closed;
  2. for every z0ρA(a) and every hC with hR(z0,a)<1 the Neumann expansion R(z0+h,a)=n0(h)nR(z0,a)n+1 converges in A and exhibits z0+hρA(a);
  3. the map zR(z,a) is holomorphic on ρA(a) in the norm sense, with derivative R(z,a)=R(z,a)2(zρA(a)), and in particular it is norm continuous there, with the local estimate R(z0+h,a)R(z0,a)hR(z0,a)2/(1hR(z0,a)) for hR(z0,a)<1.

Facts & Assumptions

Given: A unital complex Banach algebra A, an element aA, a point z0ρA(a) with R0:=R(z0,a), and hC with hR0<1.

[L1]

A is complete, the norm is submultiplicative with 1=1, and multiplication is associative and bilinear (Unital Banach algebra).

[L2]

R(z,a) is the unique two-sided inverse of z1a, so (z1a)R(z,a)=R(z,a)(z1a)=1, and R(z0+h,a) exists exactly when (z0+h)1a is invertible (Spectrum and resolvent set in a Banach algebra).

[L3]

If y<1 then 1y is invertible with (1y)1=n0yn and (1y)11/(1y) (Neumann series).

[L4]

Inversion is continuous on the invertible group, and ρA(a) is therefore open: it is the preimage of the open set A× under the continuous map zz1a (Invertible group is open and inversion is continuous).

Proof

technique · direct
1.1

Put x:=hR0, so that xhR0<1 by [L1], and (z0+h)1a=(z0a)(1x): indeed (z0a)(1+hR0)=(z0a)+h(z0a)R0=(z0a)+h, using (z0a)R0=1 from [L2].

L1L2algebra
2.1

By [L3] applied to x, the element 1x is invertible with (1x)1=n0(h)nR0n and (1x)11/(1hR0).

step 1.1L3
3.1

By [step 1.1] and [step 2.1], (z0+h)1a is a product of two invertible elements, hence invertible, with R(z0+h,a)=(1x)1R0=n0(h)nR0n+1; combined with [L4] this shows that ρA(a) is open, which is claim 1.

step 1.1step 2.1L2L4
4.1

The map zR(z,a) is norm continuous at z0: from [step 3.1], R(z0+h,a)R0=n1(h)nR0n+1, whose norm is at most n1hnR0n+1=hR02/(1hR0), and this tends to 0 with h; independently, continuity of inversion [L4] applied to the continuous map zz1a gives the same conclusion.

step 3.1L4L1
4.2

For nonzero h with hR0<1, divide the expansion of [step 3.1] by h after subtracting R0: R(z0+h,a)R0h+R02=n2(1)nhn1R0n+1. The right-hand side converges in A and has norm at most n2hn1R0n+1=hR031hR0, which tends to 0 as h0.

step 3.1L1algebra
5.1

Thus the norm difference quotient of hR(z0+h,a) at 0 converges to R02. Since z0 was arbitrary in the open set ρA(a), the resolvent is Banach-valued holomorphic there and R(z0,a)=R(z0,a)2.

step 4.2L4
6.1

The three claims are established: claim 1 by [step 3.1], claim 3 together with its continuity and estimate by [step 4.1] and [step 5.1], and claim 2 is exactly the expansion of [step 3.1].

step 3.1step 4.1step 5.1
TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Spectrum is nonempty compact and norm bounded

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a nonzero unital complex Banach algebra and let aA, with spectrum σA(a) and resolvent R(z,a)=(z1a)1 as in Spectrum and resolvent set in a Banach algebra. Then

  1. σA(a) is a compact subset of the closed disc {zC:za};
  2. σA(a).

The Axiom of Choice is used exactly once, in the form of the Hahn–Banach separation supplied by The dual space separates points of a normed space; the closedness, boundedness and nonemptiness arguments are otherwise choice-free.

Facts & Assumptions

Given: An assumed Axiom of Choice, a nonzero unital complex Banach algebra A, an element aA, and the spectrum, resolvent set and resolvent of a (Spectrum and resolvent set in a Banach algebra).

[L1]

A is complete, 1=1, uvuv and 1u=u, and 01 because A is nonzero; in particular 0 is not invertible, since 0b=01 for every b (Unital Banach algebra, Invertible element and general linear group of a Banach algebra).

[L2]

zρA(a) exactly when z1a is invertible, and then R(z,a)=(z1a)1 satisfies (z1a)R(z,a)=R(z,a)(z1a)=1 (Spectrum and resolvent set in a Banach algebra).

[L3]

If y<1 then 1y is invertible with (1y)1=n0yn (Neumann series).

[L4]

The resolvent set ρA(a) is open, and zR(z,a) is norm holomorphic there and hence norm continuous (Resolvent is Banach-valued holomorphic, Invertible group is open and inversion is continuous).

[L5]

Every bounded entire function CC is constant (Liouville's theorem: every bounded entire function is constant).

[L6]

If xy in a complex normed space then there is a bounded linear functional φ with φ(x)φ(y) (The dual space separates points of a normed space).

[A1]

The standing hypothesis is the Axiom of Choice, used here through [L6] and nowhere else (The Axiom of Choice).

Proof

technique · direct
1.1

If z>a then a/z=a/z<1, so by [L3] the element 1a/z is invertible and hence z1a=z(1a/z) is invertible with R(z,a)=z1(1a/z)1=n0zn1an. The Neumann-series norm estimate gives R(z,a)z1/(1a/z)=1/(za). Therefore zρA(a) and σA(a){za}.

L1L2L3algebra
2.1

The set ρA(a) is open by [L4], so its complement σA(a) is closed; combined with the boundedness of [step 1.1] this makes σA(a) a closed bounded subset of C, hence compact, which is claim 1.

step 1.1L4
3.1

Suppose, for contradiction, that σA(a)=, so that ρA(a)=C and R(z,a) is defined for every zC.

step 2.1L2
4.1

The element R(0,a)=(0a)1=a1 is nonzero: if a1=0 then 1=aa1=0, contradicting [L1]; here a is invertible because 0ρA(a).

step 3.1L1L2algebra
5.1

By [L6], applied to the distinct points R(0,a) and 0 in A, there is a bounded linear functional φ:AC with φ(R(0,a))0.

step 4.1L6A1
6.1

Define g:CC by g(z):=φ(R(z,a)). Then g is holomorphic on C: at each z0 the resolvent is complex differentiable with R(z0,a)=R(z0,a)2 by [L4], and a bounded linear functional is complex differentiable with φ(x)=φ for xA, so the chain rule gives g(z0)=φ(R(z0,a)2); thus g is entire.

step 3.1step 5.1L4algebra
7.1

The function g is bounded: on the compact set {za+1} the norm R(z,a) is bounded by some C1< because zR(z,a) is norm continuous by [L4], and for z>a+1 the estimate in [step 1.1] gives R(z,a)1/(za)1; hence g(z)φmax(C1,1) for every zC.

step 6.1step 1.1L4algebra
8.1

By [L5] the bounded entire function g is constant; since the estimate in [step 1.1] gives R(z,a)1/(za)0 as z and φ is continuous, g(z)0 along z, so the constant value is 0 and g0.

step 1.1step 7.1L5algebra
9.1

But g(0)=φ(R(0,a))0 by the choice of φ in [step 5.1], contradicting g0; hence σA(a), which is claim 2.

step 8.1step 5.1
10.1

Claim 1 was proved in [step 2.1] and claim 2 in [step 9.1], so the spectrum of a is a nonempty compact subset of the disc of radius a.

step 2.1step 9.1
DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Spectral radius

Definition

Assume the Axiom of Choice (The Axiom of Choice). Let A be a nonzero unital complex Banach algebra and let aA. By Spectrum is nonempty compact and norm bounded the spectrum σA(a) (Spectrum and resolvent set in a Banach algebra) is a nonempty compact subset of C contained in the closed disc of radius a. The real-valued function zz is continuous, so its image σA(a) is a nonempty compact subset of [0,) and has a maximum. The spectral radius of a is the real number

r(a)  :=  max{z:zσA(a)}.

It satisfies 0r(a)a. When the ambient algebra must be recorded, the notation is rA(a); for a bounded operator T on a nonzero complex Banach space the convention is that r(T)=max{z:zσ(T)} is computed in the algebra B(X) of bounded operators, whose spectrum convention is fixed by the spectrum definition above.

Remarks

  • The maximum is a maximum because the spectrum is compact and nonempty. This is the only place where the Axiom of Choice enters the definition: it is inherited from Spectrum is nonempty compact and norm bounded, whose nonemptiness proof uses Hahn–Banach separation. Selecting the maximum of the compact set of moduli uses no further choice, since a nonempty compact subset of R contains its supremum.

  • Monotonicity under containment. If B is a closed unital subalgebra of A containing a and the same unit, then B is a unital Banach algebra in the inherited norm and σA(a)σB(a): invertibility in B implies invertibility in A. Consequently rA(a)rB(a).

  • Constancy on scalar multiples. For λC one has σ(λ1)={λ} and hence r(λ1)=λ: the element λ1 is the unit rescaled, and z1λ1=(zλ)1 is invertible exactly when zλ. This computation is used in the counterexample cex-norm-need-not-equal-spectral-radius.

  • The radius is not the norm in general. The inequality r(a)a is strict for many elements; the definitive relation r(a)=limnan1/n is the theorem Spectral radius formula. In particular r(a)=0 is possible for nonzero a, and then σA(a)={0}.

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Polynomial spectral mapping

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra, let aA, and let pC[z] be a complex polynomial with constant term c0 and degree at most n. Form p(a):=k=0nckakA, with a0:=1. Then

σA(p(a))  =  p(σA(a))  =  {p(λ):λσA(a)},

where spectra are taken in the ambient algebra A (Spectrum and resolvent set in a Banach algebra). The identity holds for constant polynomials as well: if pc then p(a)=c1 and both sides equal {c}.

Facts & Assumptions

Given: The Axiom of Choice, a unital complex Banach algebra A, an element aA, and a complex polynomial p; write p(a)=kckak, a finite sum of scalar multiples of powers of a.

[L1]

The algebra A is associative, the multiplication is bilinear and 1u=u1=u; every polynomial in a commutes with a, and powers of a satisfy the usual index laws (Unital Banach algebra).

[L2]

For uA the element z1u is invertible exactly when zρA(u), and invertibility is a two-sided condition; commuting invertible elements have commuting inverses (Spectrum and resolvent set in a Banach algebra, Invertible element and general linear group of a Banach algebra).

[L3]

If u,vA commute and uv is invertible, then u and v are invertible: with w:=(uv)1 one has u(vw)=uvw=1 and (vw)u=vwu=vuw=uvw=1, so u1=vw; symmetrically v1=wu. [L1, L2, algebra]

[L4]

Every nonconstant complex polynomial of degree d has a factorisation q(z)=cj=1d(zλj) with c0 and λjC the roots of q (Fundamental theorem of algebra by Liouville's theorem).

[L5]

Under the Axiom of Choice, the spectrum of every element of a nonzero unital complex Banach algebra is nonempty (Spectrum is nonempty compact and norm bounded).

Proof

technique · direct
1.1

Constant case: if pc then p(a)=c1 and, for zC, the element z1c1=(zc)1 is invertible exactly when zc — its inverse is then (zc)11 — while at z=c it is 0, which is not invertible in a nonzero algebra. Hence σA(c1)={c}; and σA(a) is nonempty by [L5], so p(σA(a))={c} as well.

L1L2L5algebra
1.2

Nonconstant case, factor step: for λC the polynomial q(z):=p(z)p(λ) vanishes at λ, so q(z)=(zλ)r(z) for a polynomial r of degree degp1; evaluating at a gives p(a)p(λ)1=(aλ1)r(a).

L1algebra
1.3

Root factorisation of the translated polynomial: for μC the polynomial zp(z)μ has degree degp1 and a leading coefficient cdegp0, so by [L4] there are λ1,,λdC with p(z)μ=cdegpj=1d(zλj); evaluating at a gives p(a)μ1=cdegpj=1d(aλj1), a product of commuting elements.

L4L1algebra
2.1

Forward inclusion: if μσA(p(a)) then μp(σA(a)). Indeed, suppose p(z)μ has no zero in σA(a); by [step 1.3] the roots λj of p(z)μ satisfy p(λj)=μ, so λjσA(a) and each aλj1 is invertible; the product p(a)μ1=cdegpj(aλj1) of commuting invertible elements is invertible, so μσA(p(a)).

step 1.3L2algebra
2.2

Reverse inclusion: if λσA(a) then p(λ)σA(p(a)). For if p(a)p(λ)1 were invertible, then by [step 1.2] the commuting product (aλ1)r(a) would be invertible, so [L3] would make aλ1 invertible, contradicting λσA(a).

step 1.2L3L2
3.1

Combining [step 2.1] and [step 2.2] with [step 1.1] gives σA(p(a))=p(σA(a)) in the nonconstant case and σA(c1)={c}=p(σA(a)) in the constant case, which is the assertion.

step 1.1step 2.1step 2.2

Remarks

  • Where the fundamental theorem of algebra is used. The forward inclusion [step 2.1] needs the existence of all roots of p(z)μ, which is Fundamental theorem of algebra by Liouville's theorem. The reverse inclusion needs only polynomial division by the known linear factor zλ.

  • The statement is about the ambient algebra. Both spectra in the theorem are computed in the same unital Banach algebra A; the identity can fail for spectra taken in different algebras, since spectra may shrink in a larger algebra (cex-spectrum-can-shrink-in-a-larger-banach-algebra).

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

LemmaStatement: AI-adaptedProof: AI-adaptedaudited 2026-09-22Open item page →

Submultiplicative root limit

Statement

Let (un)n1 be a positive-indexed family of nonnegative real numbers satisfying the submultiplicative inequality

um+n    umunfor all m,n1.

Then the N-indexed sequence vj:=uj+11/(j+1) (j0) converges in R. Writing the limit of the positive-indexed root family for this sequence limit,

limnun1/n  =  infn1un1/n.

The case of a vanishing term is included: if uk=0 for some k1, then un=0 for all nk and both sides equal 0.

Facts & Assumptions

Given: A positive-indexed family (un)n1 of reals with un0 and um+numun for all m,n1; put u0:=1 and extend the inequality by this convention, so that um+0=umumu0. Define vj=uj+11/(j+1) for jN; no zeroth root is defined.

[L1]

For every a0 and n1 there is a unique a1/n0 with (a1/n)n=a; moreover 01/n=0 and a1/n>0 when a>0 (Existence and uniqueness of n-th roots: a unique a1/n0 with (a1/n)n=a).

[L2]

For x,y0 and n1: xy if and only if xnyn, and x<y if and only if xn<yn. Consequently (xy)1/n=x1/ny1/n and xy implies x1/ny1/n, since both sides are nonnegative and have the same n-th power (Monotonicity of xxn and of nan, Existence and uniqueness of n-th roots: a unique a1/n0 with (a1/n)n=a).

[L3]

For a sequence (xk) of reals the limit superior and limit inferior are elements of R with lim infkxklim supkxk, and (xk) converges to LR exactly when lim infkxk=lim supkxk=L (Limit superior and limit inferior of a real sequence as infnsupknxk and supninfknxk in R, A real sequence converges to LR iff lim infxk=lim supxk=L, and diverges to ± iff both equal ±).

[L4]

If xkyk eventually, then lim supkxklim supkyk and lim infkxklim infkyk (If xkyk eventually then lim supxklim supyk and lim infxklim infyk).

[L5]

For every real c>0, the N-indexed sequence dj=c1/(j+1) converges to 1 (For every a>0, a1/n1).

Proof

technique · direct
1.1

Put L:=inf{un1/n:n1}, a real number in [0,) because u11/1=u1 is finite and every term is 0; by definition of the infimum, un1/nL for every n1.

L1algebra
1.2

If uk=0 for some k1 then un=0 for every nk: writing n=k+(nk) with nk0, the hypothesis and the convention u0=1 give unukunk=0, while un0 by assumption.

givenalgebra
1.3

Now suppose un>0 for every n1. Fix k1, put t:=uk1/k>0 and Ck:=max{ur:0r<k}>0; then for every nk, writing n=qk+r with q1 and 0r<k, iterated submultiplicativity gives unukqurukqCk.

givenL1algebra
2.1

Suppose uk=0 for some k. Then un1/n=0 for all nk by [step 1.2] and [L1], and L=0 because the term uk1/k=0 occurs in the set whose infimum is L; hence vj=0 for jk1 and vj0=L.

step 1.2step 1.1L1L3
2.2

With t=uk1/k>0 as in [step 1.3] the bound ukqtnBk holds for n=qk+rk and the constant Bk:=max(1,tk): indeed ukq=(tk)q=tkq=tnr by the integer index laws, and the correction factor satisfies trtkBk when t1 (then r<k makes trtk) and tr1Bk when t1, so in both cases tnr=tntrtnBk.

step 1.3L1L2algebra
2.3

On the other hand every term satisfies vj=uj+11/(j+1)L by [step 1.1], so the liminf clause of [L4] applied to the constant sequence L gives L=lim infjLlim infjvj, that is lim infjvjL.

step 1.1L4
3.1

Define the positive constant Dk:=max(1,tk)Ck. Combining [step 1.3] and [step 2.2] gives untnDk, hence un1/ntDk1/n for every nk, taking n-th roots by [L2].

step 1.3step 2.2L1L2
4.1

Apply [L5] to dj=Dk1/(j+1). Substituting n=j+1 in step 3.1 gives vjtdj whenever j+1k. Since dj1, for every real ε>0 the inequality dj1+ε holds eventually; hence vjt(1+ε) eventually, and [L4] gives lim supjvjt(1+ε).

step 3.1L4L5algebra
5.1

Since ε>0 was arbitrary in [step 4.1] and t=uk1/k, one has lim supjvjuk1/k for every k1, hence lim supjvjL.

step 4.1step 1.1algebra
6.1

In the positive case, [step 5.1] and [step 2.3] yield lim supLlim inflim sup, so all three (for the sequence v) are equal to L and vjL by [L3]; in the vanishing case [step 2.1] gives the same conclusion. Hence in all cases limnun1/n=infn1un1/n.

step 2.1step 5.1step 2.3L3
TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Spectral radius formula

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra and let aA, with spectral radius r(a) (Spectral radius). Then

r(a)  =  limnan1/n  =  infn1an1/n,

Here the root sequence means vj=aj+11/(j+1) for jN; a0=1, and no zeroth root is used. The Axiom of Choice is used only through the spectrum nonemptiness and Hahn–Banach content of Spectral radius, The norm of a vector is the supremum of |f(x)| over the dual unit ball and the declared polynomial spectral-mapping supplier; the analytic estimate itself is choice-free.

Facts & Assumptions

Given: A unital complex Banach algebra A, an element aA, and the spectrum, resolvent set, resolvent and spectral radius of a (Spectrum and resolvent set in a Banach algebra, Spectral radius).

[L1]

A is complete with 1=1, xyxy and am+n=aman for all m,n0 (Unital Banach algebra).

[L2]

If y<1 then 1y is invertible with (1y)1=n0yn and (1y)1nNynyN+1/(1y) (Neumann series).

[L3]

R(z,a)=(z1a)1 on ρA(a), and R(z,a)R(w,a)=(wz)R(z,a)R(w,a) for z,wρA(a) (Spectrum and resolvent set in a Banach algebra, Resolvent identity).

[L4]

ρA(a) is open and zR(z,a) is holomorphic there with continuous norm, with R(z,a)=R(z,a)2 (Resolvent is Banach-valued holomorphic).

[L5]

If f is holomorphic on a disc containing the closed disc of radius ρ>0 around 0, then for every n0 f(n)(0)=n!2πiζ=ρf(ζ)ζn+1dζ (All higher complex derivatives exist and satisfy Cauchy's integral formula on an interior circle).

[L6]

For every integer m and every ρ>0, ζ=ρζmdζ equals 2πi when m=1 and 0 otherwise (On a positively oriented circle about a, the integral of (z-a)^m is zero for every integer m except -1, and is 2 pi i for m=-1).

[L7]

For every x in a complex normed space, x=sup{φ(x):φX, φ1} (The norm of a vector is the supremum of |f(x)| over the dual unit ball).

[L8]

With roots interpreted as the zero-based sequence uj+11/(j+1), if un0 and um+numun for all m,n1, then limnun1/n=infnun1/n (Submultiplicative root limit).

[L9]

For every polynomial p, σA(p(a))=p(σA(a)), so in particular σA(an)={λn:λσA(a)} for n1 (Polynomial spectral mapping).

[L10]

r(a)=max{z:zσA(a)} and r(b)b for every bA (Spectral radius).

[L11]

For every c>0 the sequence dj=c1/(j+1), jN, converges to 1 (For every a>0, a1/n1).

[L12]

For a rectifiable contour, the modulus of the integral is at most its length times an upper bound for the integrand modulus (ML estimate: a contour integral is bounded by a supremum bound times path length).

[A1]

The standing hypothesis is the Axiom of Choice, used through [L10], [L7] and the declared [L9] interface (The Axiom of Choice).

Proof

technique · direct
1.1

First, if a=0, then r(a)=0 by 0r(a)a in [L10], and every positive power and root norm is zero, proving the formula. In the remainder assume a0, so a>0 and divisions by this number are legitimate. For z0 one has 1za=z(az11); hence 1za is invertible exactly when az11 is, that is exactly when z1ρA(a).

L1L3L10A1algebra
1.2

For every n1 one has am+naman for all m,n1, so the positive-indexed family un:=an is submultiplicative and nonnegative, and its root sequence is vj=uj+11/(j+1) for j0.

L1algebra
1.3

For every n1, [L9] gives σA(an)={λn:λσA(a)}, hence by [L10] applied to an and the multiplicativity of the modulus, r(an)=max{λn:λσA(a)}=r(a)n; and r(an)an by the last clause of [L10], so r(a)nan.

L9L10algebra
2.1

For z<1/a one has za=za<1, so [L2] applies to y:=za and gives (1za)1=n0anzn; comparing with the identity of [step 1.1] this is a power series in z whose value at z=0 is 1 and whose linear coefficient is a.

step 1.1L2L1algebra
2.2

Fix a real R>r(a) and put DR:={zC:z<1/R}. For zDR with z0 one has z1>R>r(a), so z1σA(a) because every spectral point has modulus at most r(a); by [step 1.1] the element 1za is invertible. Hence the function h(z):=(1za)1 for z0, h(0):=1, is a well-defined map DRA.

step 1.1L10L3
3.1

h is complex differentiable at every z0DR, z00: by [step 1.1] one has h(z)=z1R(z1,a) for z0, so with u:=z1, v:=z01 the difference is h(z)h(z0)=uR(u)vR(v)=(vu)aR(u)R(v), the identity following from the resolvent identity [L3] in the form R(u)R(v)=(vu)R(u)R(v) together with the rearrangement 1vR(v)=aR(v) of vR(v)=1+aR(v); since vu=(zz0)/(zz0), the difference quotient is h(z)h(z0)zz0=aR(z1)R(z01)zz0, which converges to aR(z01)2/z02 as zz0 by continuity of the resolvent [L4] and of 1/z.

step 2.2step 1.1L3L4
4.1

At z0=0 the map h is complex differentiable with h(0)=a: by [step 2.1], h(z)=1+za+n2anzn for z<1/a, and the remainder is bounded by n2anzn=z2a2/(1za)=o(z). Combined with [step 3.1] this shows that h is holomorphic on DR.

step 2.1step 3.1L2L1algebra
5.1

Let φA be a bounded linear functional and let g:=φh:DRC. Since h is holomorphic by [step 4.1] and φ is continuous linear, g is holomorphic on DR with g(z)=φ(h(z)).

step 4.1L4algebra
6.1

Fix R with r(a)<R<R and put DR:={z:z<1/R}. The argument of steps 2.2-4.1 with R in place of R shows that h is holomorphic on DR; since 1/R<1/R, the disc DR contains the closed disc of radius 1/R around 0, the same inverse formula extends h consistently, and g=φh extends by that formula as well. Thus [L5] applies to this extended g with ρ=1/R and gives g(n)(0)=n!2πiζ=1/Rg(ζ)ζn1dζ for every n0.

step 2.2step 4.1step 5.1L5
7.1

For 0<ρ<min(1/R,1/a) the series of [step 2.1] converges uniformly on the circle ζ=ρ, so g(ζ)=φ((1ζa)1)=k0φ(ak)ζk uniformly there, and [L5] also applies on this smaller circle. For fixed n, the uniform remainder after multiplying by ζn1 is bounded by φρn1(ρa)N+1/(1ρa), which tends to zero. By [L12] its integral tends to zero. Integrating the finite sums and using [L6] therefore gives g(n)(0)=n!φ(an) for every n0.

step 6.1step 2.1L2L5L6L12algebra
8.1

Norm estimate for the coefficients, using [L12] on the circle of length 2π/R: for n0, by [step 7.1] and the integral formula of [step 6.1], φ(an)=12πζ=1/Rg(ζ)ζn1dζ(supζ=1/Rh(ζ))φRn; the supremum is finite because [step 6.1] places the circle ζ=1/R as a compact subset of the larger disc DR, on which the argument of [step 4.1] makes h holomorphic and hence continuous.

step 6.1step 7.1step 4.1L4L12algebra
9.1

Put CR:=supζ=1/Rh(ζ)<. This constant is positive because h(1/R) is invertible and therefore nonzero. Taking the supremum in [step 8.1] over all φ with φ1 and using [L7] gives anCRRn for every n0; hence vjRCR1/(j+1) for every j0. By [L8] and step 1.2, vj has a real limit Q equal to the stated infimum. By [L11], CR1/(j+1)1, and passing to these real limits gives QR. (If Q>R, convergence of both sequences would contradict their termwise inequality.) Since this holds for every R>r(a), Qr(a): otherwise choose R=(Q+r(a))/2.

step 8.1step 1.2L7L8L11algebra
10.1

By [L8] applied to the submultiplicative family un=an of [step 1.2], the limit Q:=limjvj=infnan1/n exists; [step 9.1] gives Qr(a), while [step 1.3] gives an1/nr(a) for every n, hence Qr(a). Therefore Q=r(a) and the formula holds for a0; step 1.1 already proved the zero case.

step 1.1step 1.2step 1.3step 9.1L8
LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Canonical Banach complexification of a real Banach space

Statement

Let X be a real Banach space (Banach space) and let XC:=X×X carry

  • the complex scalar multiplication (a+bi)(x,y):=(axby,  bx+ay), making it a complex vector space, and
  • the rotation-supremum norm ρ(x,y)  :=  supθRcosθxsinθy, the supremum of a bounded set of reals, so that ρ(x,y)[0,).

Then:

  1. ρ is a norm on the complex vector space XC (the complex norm axioms of Real and complex scalar conventions for normed spaces), and XC is complete for it, hence a complex Banach space;
  2. j:XXC, j(x):=(x,0), is a real-linear isometry, and for every bounded real-linear T:XX the map TC(x,y):=(Tx,Ty) is complex-linear and bounded with TC=T;
  3. (canonical comparison) Let Z be a complex Banach space and let jZ:XZ be a real-linear isometry such that every zZ has a unique representation z=jZ(x)+ijZ(y) with x,yX, and let σ:ZZ be a conjugation: real-linear with σ(iz)=iσ(z), σ2=id, σ(jZ(x))=jZ(x) and σ(z)=z for all zZ. Then Φ:XCZ,Φ(x,y):=jZ(x)+ijZ(y) is a complex-linear bijection with Φ2 and Φ12, and ΦTC=TZΦ, where TZ:=ΦTCΦ1 is the extension of T defined by TZ(jZ(x)+ijZ(y)):=jZ(Tx)+ijZ(Ty). If in addition Z carries the rotation-supremum norm relative to jZ, that is jZ(x)+ijZ(y)Z=supθcosθjZ(x)sinθjZ(y)Z, then Φ is an isometry.

The comparison is bounded, not isometric, in general; isometry holds precisely when the comparison model has the same rotation-supremum norm under its unique coordinates.

Facts & Assumptions

Given: A real Banach space X with norm ; the set XC=X×X with complex scalar multiplication (a+bi)(x,y)=(axby,bx+ay); the function ρ(x,y)=supθcosθxsinθy.

[L1]

X is a real normed space that is complete: x0 with equality only for x=0, λx=λx for real λ, and x+yx+y; the closed unit ball and all bounded sets are as in A norm on a real vector space, the induced metric, and the dictionary with the metric axioms, and completeness is Banach space. For complex spaces we use the modulus-homogeneity convention of Real and complex scalar conventions for normed spaces.

[L3]

Every μ0 in C has a representation μ=r(cosθ+isinθ) with r=μ>0 and θR (Every nonzero complex number has a unique polar form r(cosθ+isinθ) with r>0 and π<θπ).

[L4]

For all real u,v, cos(u+v)=cosucosvsinusinv and sin(u+v)=sinucosv+cosusinv (The addition formulas for sine and cosine).

[L5]

A real-linear T:XX is bounded when it has a finite bound (A bounded linear operator between normed spaces). Its operator norm T is the least such bound, and TxTx for all x (The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[L6]

The real field is a complete ordered field: every nonempty subset of R that is bounded above has a least upper bound, and suprema are monotone, satisfy sup(f+g)supf+supg for bounded real functions on a common nonempty index set, and commute with multiplication by a positive scalar (The Cauchy-sequence reals have the least-upper-bound property, Complete ordered field (least-upper-bound property)). Its order properties used below follow directly by comparing upper bounds.

[L7]

For every real angle, sinθ,cosθ1 (Parity and the Pythagorean identity for sine and cosine); (cos0,sin0)=(1,0) and (cos(π/2),sin(π/2))=(0,1) (Quarter-turn values and shifts by pi/2 and pi).

Proof

technique · direct
1.1

For every (x,y) the set {cosθxsinθy:θR} is nonempty and bounded above by x+y, because cosθxsinθycosθx+sinθyx+y; hence ρ(x,y) is a well-defined real number with 0ρ(x,y)x+y, and choosing θ=0 and θ=π2 gives xρ(x,y) and yρ(x,y).

L1L6L7algebra
1.2

j(x)=(x,0) is real-linear, and ρ(x,0)=supθcosθx=x because the value at θ=0 is x and cosθ1 bounds every other value by x; hence j is an isometric real-linear embedding.

L1L6L7algebra
1.3

For a real-linear T:XX the map TC(x,y):=(Tx,Ty) is complex-linear: TC((a+bi)(x,y))=TC(axby,bx+ay)=(aTxbTy,bTx+aTy)=(a+bi)TC(x,y) by real-linearity of T.

L1algebra
1.4

In the situation of claim 3, every zZ is jZ(x)+ijZ(y) for unique x,yX by hypothesis; hence Φ(x,y):=jZ(x)+ijZ(y) is a well-defined bijection XCZ, and it is complex-linear because jZ is real-linear and i2=1.

L1L2algebra
2.1

The conjugation inverts the two components: σ(jZ(x)+ijZ(y))=jZ(x)ijZ(y), because σ is real-linear, fixes jZ(X) pointwise and satisfies σ(iw)=iσ(w); consequently the formulas jZ(x)=12(z+σ(z)) and jZ(y)=12i(zσ(z)) hold for z=jZ(x)+ijZ(y).

step 1.4L2algebra
2.2

ρ(x,y)=0 forces x=y=0 by the bound max(x,y)ρ(x,y) of step 1.1 and definiteness of the norm, so ρ is definite.

step 1.1L1algebra
2.3

ρ satisfies the triangle inequality: for all (x,y),(x,y), and every θ, cosθ(x+x)sinθ(y+y)cosθxsinθy+cosθxsinθyρ(x,y)+ρ(x,y), so the supremum over θ gives ρ(x+x,y+y)ρ(x,y)+ρ(x,y).

step 1.1L1L6algebra
2.4

ρ is absolutely homogeneous for complex scalars: writing μ=a+bi, for θR one has cosθ(axby)sinθ(bx+ay)=(acosθbsinθ)x(bcosθ+asinθ)y; if μ0 and μ=r(cosφ+isinφ) by [L3], then a=rcosφ and b=rsinφ, so [L4] turns the two coefficients into rcos(φ+θ) and rsin(φ+θ); the norm of the resulting vector is rcosψxsinψy with ψ:=φ+θ, and taking suprema over θ (equivalently over ψ) gives ρ(μ(x,y))=rρ(x,y)=μρ(x,y), while μ=0 gives the zero vector.

step 1.1L2L3L4L6algebra
2.5

If in addition T is bounded, then ρ(TC(x,y))=supθT(cosθxsinθy)Tsupθcosθxsinθy=Tρ(x,y) by [L5], and the reverse inequality follows by evaluating at y=0, where the supremum is Tx: the operator norm of TC is exactly T. If X={0}, both unit-ball suprema are zero by [L5], so this conclusion still holds.

step 1.2step 1.3L5L6algebra
2.6

The intertwining is a definitional identity: TZ:=ΦTCΦ1 satisfies TZ(jZ(x)+ijZ(y))=Φ(TC(x,y))=jZ(Tx)+ijZ(Ty), so ΦTC=TZΦ holds by construction.

step 1.3step 1.4algebraL1
2.7

If Z carries the rotation-supremum norm relative to jZ, then Φ(x,y)Z=supθcosθjZ(x)sinθjZ(y)Z=supθjZ(cosθxsinθy)Z=supθcosθxsinθy=ρ(x,y), using real-linearity and isometry of jZ; so Φ is an isometry in that case. Conversely, if Φ is isometric, its defining formula gives exactly this norm equality for every (x,y), which is the stated rotation-supremum condition.

step 1.2step 1.4L1L6algebra
3.1

For z=jZ(x)+ijZ(y) the component estimates x=jZ(x)12(z+σ(z))=z and yz hold, because σ(z)=z by hypothesis and jZ is isometric.

step 2.1L1algebra
3.2

By steps 1.1, 2.2, 2.3 and 2.4, the function ρ satisfies definiteness, the triangle inequality and absolute homogeneity, so it is a norm on the complex vector space XC with scalar multiplication (a+bi)(x,y)=(axby,bx+ay), which is associative and distributive because C is a field.

step 1.1step 2.2step 2.3step 2.4L1L2algebra
4.1

For z=Φ(x,y) one has ρ(x,y)x+y2z by [step 3.1] and Φ(x,y)Z=jZ(x)+ijZ(y)x+y2ρ(x,y) by [step 1.1]; hence Φ1 and Φ are bounded with norms at most 2. Thus TZz4Tz, using its defining composition and step 2.5.

step 1.4step 3.1step 1.1step 2.5step 2.6L1L5algebra
4.2

The norm ρ is equivalent to the product maximum norm (x,y):=max(x,y): indeed (x,y)ρ(x,y)2(x,y) by [step 1.1]. Consequently a ρ-Cauchy sequence in XC is Cauchy for , hence its two coordinate sequences are Cauchy in X and converge by completeness of X, and the coordinatewise limit is the ρ-limit by the same two-sided estimate; so XC is complete for ρ and is a complex Banach space.

step 3.2step 1.1L1L6algebra
5.1

Claims 1, 2 and 3 are established: [step 3.2] and [step 4.2] give the complex Banach space, [step 1.2] and [step 2.5] give the isometric embedding and the same-norm extension, and [step 4.1], [step 2.6] and [step 2.7] give the bounded canonical comparison, its intertwining property and its isometry in the equal-norm case.

step 3.2step 4.2step 1.2step 2.5step 4.1step 2.6step 2.7L1

Remarks

  • The comparison is not claimed to be isometric in general. Bühler–Salamon Exercise 5.4 and the surrounding discussion show that a real Banach space can carry different complexification norms agreeing on its real copy; the rotation-supremum model is one convenient choice, and the canonical map between two compatible models is bounded in both directions but isometric only when norms are the same rotation-supremum construction.

  • Why a real operator's spectrum is defined through the complexification. TC is complex-linear on a complex Banach space, and, when X{0}, the nonzero unital algebra B(XC) applies to it and the whole spectrum theory of this page becomes available; the definition Complexification and spectrum of a real operator records that convention and uses the bounded comparison of claim 3 to show that the resulting spectrum does not depend on the model.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Complexification and spectrum of a real operator

Definition

Assume the Axiom of Choice (The Axiom of Choice), inherited by the spectral-radius definition. Let X be a nonzero real Banach space and let T:XX be a bounded real-linear operator (A bounded linear operator between normed spaces). Let XC=X×X be the canonical complexification with the rotation-supremum norm ρ and let TC(x,y):=(Tx,Ty) be the complex-linear extension, both from Canonical Banach complexification of a real Banach space; thus TCB(XC) with TC=T. The spectrum, resolvent set, resolvent and spectral radius of the real operator T are those of TC computed in the unital Banach algebra B(XC):

σ(T):=σB(XC)(TC),ρ(T):=Cσ(T),R(z,T):=(z1TC)1(zρ(T)),r(T):=r(TC),

with the conventions of Spectrum and resolvent set in a Banach algebra and Spectral radius.

The operator algebra is complete by If (Y) is Banach then (\mathcal B(X,Y)) is Banach. Composition is bilinear and associative, and STuSTu gives submultiplicativity by The operator norm as the least bound and as the unit-sphere or unit-ball supremum. Its identity has norm one: it is bounded by one, and a nonzero vector, normalized to norm one, gives equality. Thus it is nonzero and satisfies Unital Banach algebra. The same argument applies to each comparison model below, since its embedded real copy is nonzero.

Well-definedness (independence of the complexification model). Let Z be another compatible complexification of X in the sense of claim 3 of Canonical Banach complexification of a real Banach space: a complex Banach space with a real-linear isometric embedding jZ:XZ such that Z=jZ(X)ijZ(X), an isometric conjugation σZ, and let TZ(jZ(x)+ijZ(y)):=jZ(Tx)+ijZ(Ty) be the corresponding extension of T. By that lemma the map Φ:XCZ, Φ(x,y)=jZ(x)+ijZ(y), is a bounded complex-linear bijection with bounded inverse Φ1, and ΦTCΦ1=TZ. Hence for every zC

z1TZ  =  Φ(z1TC)Φ1,

so z1TZ is invertible in B(Z) exactly when z1TC is invertible in B(XC), with (z1TZ)1=Φ(z1TC)1Φ1. Taking spectra,

σB(Z)(TZ)=σB(XC)(TC),r(TZ)=r(TC),

because the bijection zz matches the two spectral sets and preserves moduli. So the spectrum and the spectral radius of a real operator do not depend on which compatible complexification computes them, and all of them are computed below in the canonical model.

Remarks

  • Why not "real λ with λIT not invertible". Restricting the discussion to real scalars would discard the genuinely complex part of the spectrum. For example the quarter-turn J(u,v)=(v,u) on Euclidean R2 has complexified spectrum exactly {i,i}: J2=I, so for z2+10 the inverse of zIJ is (zI+J)/(z2+1); at z=i,i the respective nonzero complex vectors (1,i) and (1,i) are in the kernel. Its real-scalar resolvent is all of R, whereas its complex spectrum is nonempty. More precisely the real-scalar noninvertibility set equals σ(T)R. Indeed, a bounded real inverse extends componentwise to a bounded complex inverse. Conversely, for real z the operator zITC commutes with the canonical conjugation C(x,y)=(x,y), which is isometric by replacing θ with θ in the norm formula. Its bounded inverse therefore also commutes with C and restricts to a bounded inverse on its fixed real copy X×{0}. This gives both directions without conflating real and complex spectra.

  • The operator is bounded by hypothesis. The same-norm extension statement of the complexification lemma is used only for bounded real-linear T; it is what makes TC an element of the Banach algebra B(XC), to which the spectral theory of this page applies. For unbounded real operators no spectrum in this sense is defined here.

  • zR(z,T) is a holomorphic B(Z)-valued map on ρ(T) by Resolvent is Banach-valued holomorphic, applied in whichever model is used; the comparison isomorphism above conjugates one resolvent map into the other. In particular the resolvent of a real operator is well defined at a point zC exactly when zσ(T).

TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Gelfand-Mazur

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra (Unital Banach algebra) which is a division algebra: every nonzero element of A is invertible (Invertible element and general linear group of a Banach algebra). Then the map

CA,λλ1,

is an isomorphism of complex algebras and an isometry, and consequently A=C1 and dimCA=1.

Facts & Assumptions

Given: An assumed Axiom of Choice, a unital complex Banach algebra A in which every nonzero element is invertible, and the map φ:CA, φ(λ)=λ1.

[L1]

A is a complex vector space with associative bilinear multiplication, 1a=a1=a for all a, 1=1 and xyxy; in particular 10 (Unital Banach algebra).

[L2]

aA is invertible exactly when some b satisfies ab=ba=1; the only non-invertible element of a division algebra is 0 (Invertible element and general linear group of a Banach algebra).

[L3]

λσA(a) exactly when λ1a is not invertible (Spectrum and resolvent set in a Banach algebra).

[L4]

Under the Axiom of Choice every element of a nonzero unital complex Banach algebra has nonempty spectrum (Spectrum is nonempty compact and norm bounded).

Proof

technique · direct
1.1

The map φ is complex-linear and multiplicative: φ(0)=0 and φ(1)=1; φ(λ+μ)=(λ+μ)1=λ1+μ1; φ(λμ)=(λμ)1=(λ1)(μ1)=φ(λ)φ(μ), using bilinearity and 11=1; and φ is injective, because λ1=0 with λ0 would give 1=λ1λ1=0, contradicting [L1].

L1algebra
1.2

For every aA the spectrum σA(a) is nonempty by [L4].

L4
2.1

Fix aA and pick λσA(a) by [step 1.2]; then λ1a is not invertible by [L3], so λ1a=0 by the division-algebra hypothesis [L2]; hence a=λ1=φ(λ) lies in the image of φ.

step 1.2L2L3
2.2

The isomorphism is isometric: φ(λ)=λ1=λ1=λ by [L1].

step 1.1L1algebra
3.1

Since a was arbitrary, φ is surjective, and by [step 1.1] it is an injective complex-algebra homomorphism; hence it is a complex-algebra isomorphism CA and A=C1.

step 1.1step 2.1algebra
4.1

The statements of the theorem are proved: φ is an algebra isomorphism by [step 3.1] and an isometry by [step 2.2], so a complex unital Banach division algebra is one-dimensional over C.

step 2.2step 3.1

Remarks

  • The Axiom of Choice enters only through the nonemptiness of the spectrum. If one is willing to assume that the spectrum of every element is nonempty, the argument above is choice-free; conversely the theorem is the standard quantitative form of the fact that one-point spectra force division algebras to be scalars.

  • "Division algebra" cannot be weakened to "no zero divisors". The disc algebra A(D) is a unital commutative complex Banach algebra without zero divisors: a product of two functions whose product vanishes on the connected disc D vanishes identically by analytic continuation, so one factor is zero. It is nevertheless not a division algebra, because the coordinate function z is nonzero while σA(D)(z)=D (cex-spectrum-can-shrink-in-a-larger-banach-algebra, ex-maximal-ideal-space-of-the-disc-algebra). The boundary argument for the spectrum produces only a topological zero divisor, that is, an element a admitting unit vectors bn with abn0 or bna0; topological zero divisors need not be algebraic ones, as 1z in the disc algebra shows, so the two notions must not be conflated. The correct replacement of "no zero divisors" is "no nonzero topological zero divisors": every element of the boundary of the invertible group is a topological zero divisor, so a unital complex Banach algebra in which no nonzero element is a topological zero divisor is a division algebra.

  • Use in the Gelfand theory. This is the step that identifies the quotient A/m of a commutative unital Banach algebra by a maximal ideal with C, making characters and maximal ideals correspond; the following page of this track uses the theorem in exactly that form.

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Banach algebra valued contour integral

Definition

Let E be a complex Banach space (Banach space), let γ:[a,b]C be a piecewise C1 contour (Rectifiable complex contours, reversal, concatenation, closedness, and orientation), and let f:γE be continuous. The construction applies in particular to E=A for a unital complex Banach algebra (Unital Banach algebra); no algebra multiplication or unit is used.

For a<b fix a finite subdivision a=t0<<tm=b such that the restriction of γ on each piece has a continuous derivative extension vk to its closed interval. Given a tagged partition P=(sj,ξj) refining these nodes, define S(f,γ,P,ξ)=jf(γ(ξj))vk(j)(ξj)(sjsj1)E, where [sj1,sj] lies in the k(j)-th piece. At a node, use the derivative extension from this piece; the two adjacent intervals may therefore use different values. The contour integral is the norm limit γf(z)dz=limmesh(P)0S(f,γ,P,ξ). Existence and independence of all these choices are verified below. On a singleton parameter interval the integral is defined to be zero.

Equivalently it is the Bochner integral on the finite Lebesgue measure interval abf(γ(t))γ(t)dt, where the finitely many corner values may be assigned arbitrarily. It satisfies γf(z)dzL(γ)supzγf(z). Concatenation adds the integrals and reversal negates them. An increasing piecewise-C1 bijection of compact parameter intervals whose inverse is also piecewise C1 leaves the integral unchanged.

For a finite complex chain Γ=k<rmkγk whose nonzero terms are piecewise C1 contours, and continuous f:ΓE, define Γf(z)dz=k<rmk0mkγkf(z)dz. Zero-coefficient terms are omitted; the empty chain integrates to zero.

Remarks

Existence and Bochner agreement. On the k-th closed piece put Fk(t)=f(γ(t))vk(t). This is uniformly continuous and bounded. Subdivide each piece into 2n equal intervals and use its left endpoint values to obtain finite-valued measurable step functions hn. Assign fixed values at the finitely many nodes. Uniform continuity on the finitely many pieces shows hnF uniformly away from these nodes; here F denotes f(γ)γ with the chosen node values. In particular F is strongly measurable, not merely scalar measurable. Each hn is integrable, and Fhn0 on this finite interval. Thus the definition of Bochner integration (Bochner-integrable function) supplies its integral, and Bochner integrability criterion gives independence of the approximants.

For arbitrary tagged refinements the corresponding step function differs from F in norm by at most a common modulus ω(meshP) off the nodes. Hence its L1 difference from F is at most (ba)ω(meshP). Comparing its simple integral with those of hn, the triangle inequality for finite sums bounds the difference of integrals by the L1 difference. Passing to the limit proves convergence of all tagged sums to the same Bochner value. Different finite subdivisions have a common refinement and the same a.e. function F; changing finitely many endpoint values changes neither integral. This proves all independence claims without a choice of an infinite family of tags.

Norm estimate. The triangle inequality gives S(f,γ,P,ξ)supγfjvk(j)(ξj)(sjsj1). The scalar sums tend to the sum of the speed integrals on the pieces, which is L(γ) by A continuous piecewise-C1 path is rectifiable and its length is the sum of the speed integrals over its pieces. Taking the limit proves the bound. A constant contour and a zero integrand therefore have zero integral, as does a contour with singleton parameter interval.

Increment sums and parameter changes. The same value is the limit of T(f,γ,P,ξ)=jf(γ(ξj))(γ(sj)γ(sj1)). To see this, apply the real mean-value theorem (The mean value theorem, as the case g(x)=x of Cauchy's: for f continuous on [a,b] with a<b and differentiable on (a,b) there is c(a,b) with f(b)f(a)=f(c)(ba)) separately to the two coordinates of γ on an interval contained in one smooth piece. If η is a common modulus of the derivative extensions, the difference between its complex increment and vk(j)(ξj)(sjsj1) is at most 2η(meshP)(sjsj1). Therefore TS2supf(ba)η(meshP)0. This also holds for partitions not containing the original nodes: inserting the finitely many nodes changes only intervals of total length at most 2mmeshP. The bounded derivative extensions bound the variation of γ there by a constant times this length, so both their old and subdivided contributions tend to zero.

Under an increasing reparametrization as specified above, tagged partitions and tags map to tagged partitions and tags with exactly the same increment sums. Uniform continuity of the parameter map makes the image mesh tend to zero. Both contours remain piecewise C1, so their integrals agree. Reversal reverses the order and the signs of the increments. For concatenation, split a partition at the joining parameter and use its two affine pieces; the increment sums split into the two sums. These facts prove the asserted reversal and concatenation identities. Finite linearity in chains follows from their definition. No claim is made here for a reparametrization taking a contour outside the piecewise-C1 domain.

Scalar consistency. When E=C, expansion into real and imaginary parts turns the increment sums into the four Riemann–Stieltjes sums in The complex line integral over a rectifiable path as a componentwise Riemann–Stieltjes integral. Thus the limits agree on the common piecewise-C1 domain. Taking finite sums gives agreement with Integration over a complex chain and the index of a chain. The zero Banach space is allowed and all its integrals are zero. The construction uses completeness, uniform continuity and explicitly prescribed finite subdivisions; it makes no new choice assumption.

LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Contour integral commutes with bounded linear maps

Statement

Let A be a unital complex Banach algebra, let E be a complex Banach space, let B:AE be a bounded complex-linear map (A bounded linear operator between normed spaces), let γ be a piecewise C1 complex contour, and let f:γA be continuous. Then

  1. Bf is continuous on γ and B ⁣(γf(z)dz)=γ(Bf)(z)dz, the first integral being that of Banach algebra valued contour integral and the second computed in the Banach space E;
  2. γf(z)dz    L(γ)supzγf(z), where L(γ) is the length of γ.

Facts & Assumptions

Given: A unital complex Banach algebra A, a complex Banach space E, a bounded linear B:AE, a piecewise C1 contour γ:[a,b]C with trace γ and length L(γ), a C1 subdivision a=t0<<tm=b with derivative extensions vk on the closed pieces, and a continuous f:γA.

[L1]

γfdz is the limit, over tagged partitions refining the subdivision, of jf(γ(ξj))vk(j)(ξj)Δj, where the derivative extension belonging to the subinterval is used even when a tag is a corner; the chain version is the corresponding finite sum (Banach algebra valued contour integral).

[L2]

B is complex-linear and bounded, and its operator norm satisfies B(λu+μv)=λB(u)+μB(v) and B(u)Bu for all u,vA and scalars λ,μ (A bounded linear operator between normed spaces, The operator norm as the least bound and as the unit-sphere or unit-ball supremum).

[L3]

For a piecewise C1 path γ the length is the sum of the speed integrals over a C1 subdivision: L(γ)=ktk1tkγ(t)dt; on each such interval the speed is continuous, and the corresponding refined Riemann sums converge to this sum (A continuous piecewise-C1 path is rectifiable and its length is the sum of the speed integrals over its pieces).

Proof

technique · direct
1.1

Bf is continuous as a composition of continuous maps, and for every tagged partition refining the fixed subdivision, jB(f(γ(ξj)))vk(j)(ξj)Δj=B ⁣(jf(γ(ξj))vk(j)(ξj)Δj), because B is linear and the scalars vk(j)(ξj)Δj pull out of B.

L1L2algebra
1.2

For every such tagged partition, jf(γ(ξj))vk(j)(ξj)Δj(supγf)jvk(j)(ξj)Δj.

L1L2algebra
2.1

Passing to the limit in [step 1.1] using continuity of B and the convergence of the Riemann sums in [L1] gives B(γfdz)=γ(Bf)dz, which is claim 1.

step 1.1L1L2
2.2

Passing to the limit in [step 1.2] and using that the speed sums converge to the length, as in [L3], gives the estimate γfdzL(γ)supγf, which is claim 2.

step 1.2L1L3
3.1

The two claims of the statement are exactly [step 2.1] and [step 2.2].

step 2.1step 2.2
LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Banach-valued Cauchy integral vanishes

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra, let UC be open, and let F:UA be continuous and weakly holomorphic: for every bounded linear functional φ:AC (The dual space X^* of a normed space and its dual norm) the scalar function φF:UC is holomorphic. Let Γ be a complex chain which is a cycle, with trace in U (Complex chains, their traces, and cycles) and null-homologous in U (Null-homologous cycles and homologous cycles in an open set). Then

ΓF(z)dz=0,

the integral being that of Banach algebra valued contour integral over the chain Γ. The Axiom of Choice is used exactly once, in the separation step supplied by The dual space separates points of a normed space.

Facts & Assumptions

Given: An assumed Axiom of Choice, an open UC, a continuous weakly holomorphic F:UA, and a chain Γ=k<rmkγk which is a cycle with trace ΓU and is null-homologous in U.

[F1]

For pΓ one has Γdz/(zp)=2πin(Γ,p), and Γgdz=k<rmk0mkγkgdz for every g continuous on Γ; integrals over chains are additive (Banach algebra valued contour integral, Integration over a complex chain and the index of a chain).

[F2]

For a bounded linear φ:AC and a single contour γ, bounded linearity commutes with the contour integral (Contour integral commutes with bounded linear maps). Hence for the finite chain Γ=k<rmkγk and every f continuous on its trace, φ ⁣(Γfdz)=k<rmk0mkφ ⁣(γkfdz)=Γ(φf)dz, by the chain-integral definition in [F1]. Each retained contour has trace contained in Γ, so its integral is defined; zero-coefficient contours are omitted even if their traces lie outside the domain of f. For an empty retained list, both sides are zero by linearity.

[F3]

If ΩC is open, g:ΩC is holomorphic, and Γ is a complex chain which is a cycle with trace in Ω and null-homologous in Ω, then Γgdz=0 (Cauchy's theorem for a null-homologous cycle).

[F4]

If xy in a complex normed space V then there is a bounded linear functional φ on V with φ(x)φ(y) (The dual space separates points of a normed space).

[A1]

The standing hypothesis is the Axiom of Choice, used here through [F4] and nowhere else (The Axiom of Choice).

Proof

technique · direct
1.1

For every bounded linear functional φ:AC the composition φF is holomorphic on U by weak holomorphy, and it is continuous; moreover Γ is a cycle with trace in U that is null-homologous in U by hypothesis, so [F3] applies to g:=φF and gives Γφ(F(z))dz=0.

F1F3
2.1

For every bounded linear φ, φ(ΓFdz)=Γφ(F(z))dz=0: the first equality is [F2], and the second is [step 1.1].

step 1.1F2
3.1

Suppose ΓFdz0. Then [F4] applied to the distinct points x:=ΓFdz and 0 produces a bounded linear functional φ with φ(ΓFdz)0, contradicting [step 2.1]; hence ΓFdz=0.

step 2.1F4A1
LemmaStatement: AI-adaptedProof: AI-adaptedaudited 2026-09-22Open item page →

Admissible cycle around a compact plane set

Statement

Let KUC with K compact and U open. Then:

  1. there is a finite polygonal complex cycle Γ in UK — a finite chain of directed line segments (Complex chains, their traces, and cycles, Filled complex triangles, their oriented three-edge boundary contours, diameter, and perimeter) — such that n(Γ,z)=1for every zK,n(Γ,z)=0for every zU;
  2. there are two such cycles β,γ, both with index 1 on K and 0 outside U, whose traces are disjoint, and which are nested: n(γ,w)=1for every wβ,n(β,w)=0for every wγ.

All indices are those of Integration over a complex chain and the index of a chain. The construction is choice-free: the only selections are from the finitely many cells of a grid, and no supremum over an infinite family is used.

Facts & Assumptions

Given: A compact K and an open U with KUC.

[L1]

There is a compact Jordan set J, a finite union of closed rectangles of one axis-parallel grid with pairwise disjoint interiors, such that KintJJU; we write Q for its finitely many cells (A compact subset of an open Euclidean set has a compact Jordan neighborhood inside that open set).

[L2]

A finite sum of closed complex contours with integer coefficients is a complex chain; the trace of a sum is the union of the traces, a closed contour is a cycle, and the concatenation of the four sides ab,bc,cd,da of an axis-parallel rectangle is a closed contour whose trace is its boundary (Complex chains, their traces, and cycles, Filled complex triangles, their oriented three-edge boundary contours, diameter, and perimeter, Goursat's theorem for rectangles: a holomorphic function integrates to zero around every rectangle contained in its domain).

[L3]

For a chain Γ and pΓ, the index is n(Γ,p)=12πiΓdzzp; it is additive over sums of chains, and reversing a contour negates its index (Integration over a complex chain and the index of a chain, Chain integration and the index are additive in the chain, and reverse with it).

[L4]

For a closed complex contour γ and pγ, a continuous argument θ of γp exists along γ, and n(γ,p)=(θ(b)θ(a))/2π (Every contour missing a point admits a continuous logarithm, unique up to a constant in 2πiZ, The winding number is the increment of a continuous argument divided by 2π).

[L5]

The index of a cycle is continuous, hence locally constant, on the complement of its trace, and is constant on every connected component of that complement (The index of a cycle is locally constant off its trace and vanishes far from it).

[L6]

If f is holomorphic on an open set containing a closed axis-parallel rectangle R, then Rfdz=0 for the positively oriented boundary R=abbccdda (Goursat's theorem for rectangles: a holomorphic function integrates to zero around every rectangle contained in its domain).

Proof

technique · direct
1.1

Cell index inside. Let Q be one of the closed grid rectangles, with Q its positively oriented boundary, and let pQ. Along each of the four sides of Q the point ζp has a continuous argument: by [L4] a continuous argument θ of Qp exists, and along a side the perpendicular foot from p to the side's supporting line lies in the relative interior of the side, because both coordinates of p lie strictly between the corresponding coordinates of Q; hence θ varies monotonically along each side by exactly the angle subtended at p by that side, and the four such angles sum to 2π because the four triangles from p to the sides tile Q and their angles at p cover one full turn. By [L4], n(Q,p)=(θendθstart)/2π=1.

L2L4algebra
1.2

Cell index outside. Let Q be one of the closed grid rectangles and let pQ. Then p is strictly to the left of the left side, to the right of the right side, below the bottom side, or above the top side of Q; in each case the whole trace Q lies in an open half-plane bounded by a line through p, so a continuous argument of Qp takes values in an interval of length π; its increment is a multiple of 2π by [L4], hence 0, and n(Q,p)=0.

L2L4algebra
1.3

Cell index outside by Goursat. For Q and pQ the function z1/(zp) is holomorphic on an open set containing Q, so [L6] gives Qdz/(zp)=0 and hence n(Q,p)=0 again; the two computations agree and either may be used below.

L3L6
1.4

Choose J and its cells Q as in [L1]. For each cell Q, write its four positively oriented directed side contours separately. Form Γ directly as the finite list of those directed side occurrences that are not shared with another cell; a shared grid side has exactly two occurrences, with opposite directions, and neither is put in Γ. This is a chain under [L2], without identifying it with or deleting terms from the list QQ. Its trace is exactly the exposed grid sides, hence J. It is a cycle: the sum of the endpoint-boundary functions of all four sides of every cell is zero, while each omitted opposite pair also has zero endpoint-boundary function, so the remaining endpoint counts cancel at every grid vertex.

L1L2algebra
2.1

For p lying on no grid line, n(Γ,p)=QQn(Q,p)=1 if pJ and =0 if pJ: the first equality holds because the omitted opposite pairs contribute zero to the index by [L3] and the index is additive, and the second because exactly one cell contains p in its interior when pJ, while no cell contains p when pJ.

step 1.1step 1.3step 1.4L3algebra
3.1

The index n(Γ,) is continuous on CJ by [L5]; since the points on no grid line are dense in CJ, [step 2.1] extends by continuity to n(Γ,z)=1 for every zJ and n(Γ,z)=0 for every zJ.

step 2.1L5
4.1

Claim 1 follows with this Γ: KintJ=J and JU, so n(Γ,z)=1 for zK; and zU implies zJ, so n(Γ,z)=0; the trace of Γ is JJKUK.

step 3.1L1
4.2

For claim 2, choose a compact Jordan set J2, again a finite union of grid rectangles, with KintJ2J2intJ, and let β be the cycle obtained from J2 by the construction of [step 1.4]; then β=J2intJ and γ=J are disjoint, and n(γ,w)=1 for every wintJ, in particular for every wJ2=β.

step 1.4step 3.1L1
5.1

With β and γ as in [step 4.2], also n(β,w)=0 for every wγ=J: indeed wJ2 because JJ2=, and [step 3.1] applied to the Jordan set J2 gives n(β,w)=0 for wJ2; moreover n(β,z)=1 for zK because KintJ2.

step 3.1step 4.2
6.1

Claims 1 and 2 are established by [step 4.1] and [step 5.1] together with [step 4.2]: the cycle Γ, and the nested pair β,γ, have the stated index properties and traces.

step 4.1step 4.2step 5.1

Remarks

  • Why the index-one clause is the only one used to define f(a). Both Holomorphic functional calculus and its homomorphism theorem need a cycle whose index is exactly one on the spectrum and zero outside the holomorphy domain; the nested pair of claim 2 is what makes the product rule for the calculus a single separated double integral rather than a limiting argument.

  • Two different cycles, two different constructions of the same index. The argument-increment computation [step 1.1] and the Goursat computation [step 1.3] are independent, and both are used: the first identifies the index of a cell as one, the second as zero outside.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Holomorphic functional calculus

Definition

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra, let aA, and let f be a function holomorphic on an open set UC containing the spectrum σA(a) (Spectrum and resolvent set in a Banach algebra); the spectrum is a nonempty compact subset of the plane (Spectrum is nonempty compact and norm bounded). By Admissible cycle around a compact plane set applied to the compact set σA(a)U there is a finite polygonal complex cycle Γ with trace in UσA(a) such that

n(Γ,z)=1for zσA(a),n(Γ,z)=0for zU;

such a cycle is called admissible for (f,U) (or simply admissible). Define

f(a)  :=  12πiΓf(z)R(z,a)dz    A,

where R(z,a)=(z1a)1 is the resolvent and the integral is that of Banach algebra valued contour integral over the chain Γ (Complex chains, their traces, and cycles). The integrand zf(z)R(z,a) is continuous on the trace of Γ: f is holomorphic on U, and zR(z,a) is norm continuous on the resolvent set Resolvent is Banach-valued holomorphic, which contains Γ. So the integral exists, and f(a)A.

The construction describes the value attached to the germ of f near σA(a): two holomorphic functions f1 on U1 and f2 on U2 with the same germ at σA(a) — that is, agreeing on some neighbourhood of σA(a) — give the same f(a). The value is also independent of which admissible cycle is used; that is Holomorphic functional calculus is contour independent, and until it is proved the notation f(a) refers to the value computed from any one chosen admissible cycle.

Remarks

  • The hypothesis is nonempty and the cycle exists without choice. The spectrum of a is nonempty and compact, so the admissible cycle of Admissible cycle around a compact plane set always exists; the construction inside that lemma uses only finitely many grid cells.

  • Notation for operators. For a nonzero complex Banach space X and TB(X) the definition applies with A=B(X) and gives f(T)=12πiΓf(z)(z1T)1dzB(X), the Dunford integral of the resolvent. The spectrum is taken in B(X) (ex-bounded-operators-form-a-noncommutative-banach-algebra).

  • What is not part of the definition. The definition does not assert that ff(a) is multiplicative, that it preserves polynomials, or that σ(f(a))=f(σ(a)); those properties are proved from this definition in Holomorphic functional calculus homomorphism and Holomorphic spectral mapping and composition. In particular the contour independence of the value is a theorem, and the notation is provisional until then.

  • Wider or smaller domains of holomorphy. Only the germ at σA(a) matters: enlarging U beyond a neighbourhood of the spectrum does not change the value, and shrinking it is allowed as long as it still contains the spectrum and the cycle lies inside it. Both statements follow from Holomorphic functional calculus is contour independent.

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Holomorphic functional calculus is contour independent

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra, aA, and let f be holomorphic on an open set U containing σA(a). Then the value f(a) of Holomorphic functional calculus is independent of the admissible cycle used to compute it. Moreover, if VU is another open set containing σA(a) and g is holomorphic on V with g=f on a neighbourhood of σA(a), then f(a)=g(a) with g computed from V.

Facts & Assumptions

Given: An assumed Axiom of Choice, a unital complex Banach algebra A, an element aA with nonempty compact spectrum σA(a), an open UσA(a), a holomorphic f:UC, and two admissible cycles Γ1,Γ2 in UσA(a).

[L1]

The definition chooses an admissible cycle Γ0 and sets f(a)=12πiΓ0f(z)R(z,a)dz; for every admissible cycle the displayed integral is defined because the integrand is continuous on its trace (Holomorphic functional calculus).

[L2]

A cycle Γ is null-homologous in an open set Ω exactly when n(Γ,p)=0 for every pΩ; equivalently n(Γ,p) vanishes at every point outside Ω (Null-homologous cycles and homologous cycles in an open set).

[L3]

If F is continuous and weakly holomorphic on an open Ω and Γ is a cycle with trace in Ω that is null-homologous in Ω, then ΓFdz=0 (Banach-valued Cauchy integral vanishes).

[L4]

For fixed a the map zR(z,a) is holomorphic on ρA(a) and the product of a scalar holomorphic function with it is weakly holomorphic: for every bounded linear functional φ on A the map zφ(f(z)R(z,a)) is holomorphic on UρA(a) (Resolvent is Banach-valued holomorphic, Spectrum and resolvent set in a Banach algebra).

[L6]

For every compact K contained in an open set U there is a finite polygonal cycle with index 1 on K and index 0 outside U (Admissible cycle around a compact plane set).

Proof

technique · direct
1.1

Put Ω:=UσA(a), an open set containing both traces Γ1,Γ2; the difference Γ1Γ2 is a cycle with trace in Ω whose index at p is n(Γ1,p)n(Γ2,p).

L2L4algebra
1.2

The map F(z):=f(z)R(z,a) is continuous on Ω and weakly holomorphic there: for a bounded linear functional φ the composition φ(F(z))=f(z)φ(R(z,a)) is a product of the holomorphic scalar function f and the holomorphic scalar function φ(R(,a)), hence holomorphic on the open subset Ω of ρA(a).

L4algebra
1.3

Germ independence: let VσA(a) be open and let g be holomorphic on V with f=g on a neighbourhood W of σA(a). Apply [L6] to the compact set σA(a) and the open set UVW: this gives an admissible cycle for both (f,U) and (g,V) with trace in (UVW)σA(a), hence lying in W, where f=g.

L6algebra
2.1

For pσA(a) both indices equal 1, and for pU both equal 0, by admissibility; hence n(Γ1Γ2,p)=0 for every pUσA(a), that is, Γ1Γ2 is null-homologous in Ω:=UσA(a).

step 1.1L2algebra
3.1

By [L3] applied to F and the cycle Γ1Γ2, null-homologous in Ω by [step 2.1], one has Γ1Γ2Fdz=0; by additivity of the chain integral this gives Γ1f(z)R(z,a)dz=Γ2f(z)R(z,a)dz, hence f(a) does not depend on the admissible cycle.

step 1.2step 2.1L1
4.1

On the trace of the cycle of [step 1.3] the two integrands coincide, f(z)R(z,a)=g(z)R(z,a), so the two integrals agree; by [step 3.1] applied to each function separately, f(a)=g(a).

step 1.3step 3.1L1
5.1

Both assertions of the statement are proved: cycle independence by [step 3.1] and germ independence by [step 4.1].

step 3.1step 4.1
TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Holomorphic functional calculus homomorphism

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra and let aA. Write f(a) for the holomorphic functional calculus value of Holomorphic functional calculus, with the contour independence of Holomorphic functional calculus is contour independent in force. Let f,g be holomorphic on open sets Uf,UgσA(a) and let α,βC. Then:

  1. (αf+βg)(a)=αf(a)+βg(a), computed on UfUg;
  2. (fg)(a)=f(a)g(a), computed on UfUg;
  3. 1(a)=1 for the constant function 1, and id(a)=a for the coordinate function id(z)=z, computed on any open set containing σA(a);
  4. for every polynomial p(z)=k=0nckzk one has p(a)=k=0nckak, the sum in the Banach algebra A;
  5. if h is holomorphic and nowhere zero on UhσA(a), then h(a) is invertible with h(a)1=(1/h)(a), where 1/h is holomorphic on Uh.

Thus ff(a) is a unital algebra homomorphism from the algebra of germs of functions holomorphic near σA(a) to A, and it reproduces polynomials and reciprocals of nonvanishing functions.

Facts & Assumptions

Given: An assumed Axiom of Choice, a unital complex Banach algebra A, an element aA, an open set UσA(a), a holomorphic f:UC, and an admissible cycle Γ in UσA(a) with index 1 on σA(a) and index 0 outside U.

[L1]

f(a)=12πiΓf(z)R(z,a)dz, independent of the admissible cycle, and the chain integral is additive over sums of contours with the norm bound ΓhdzL(Γ)supΓh for continuous h (Holomorphic functional calculus, Holomorphic functional calculus is contour independent, Banach algebra valued contour integral, Contour integral commutes with bounded linear maps).

[L9]

The spectrum is contained in the closed disc of radius a (Spectrum is nonempty compact and norm bounded).

[L2]

Resolvent identity: R(w,a)R(z,a)=(R(w,a)R(z,a))/(zw) for distinct w,zρA(a); all resolvents and the element a commute with one another (Resolvent identity, Spectrum and resolvent set in a Banach algebra).

[L3]

Cauchy formula on a cycle: if g is holomorphic on an open Ω and Γ is a cycle with trace in Ω null-homologous in Ω, then n(Γ,p)g(p)=12πiΓg(ζ)/(ζp)dζ for every pΩΓ (Cauchy's integral formula for a null-homologous cycle).

[L4]

Vanishing Cauchy theorem: if h is holomorphic on an open Ω and Γ is a cycle with trace in Ω null-homologous in Ω, then Γhdz=0 (Cauchy's theorem for a null-homologous cycle).

[L5]

Nested cycles: for compact KU with U open there are cycles β,γ with traces in UK, disjoint, with n(β,)=n(γ,)=1 on K, n(γ,w)=1 for wβ and n(β,z)=0 for zγ; each is a finite chain of directed line segments whose boundary function vanishes, although its constituent contours need not be closed (Admissible cycle around a compact plane set).

[L6]

For a cycle Γ and 0Γ, Γz1dz=2πin(Γ,0) by the definition of index (Integration over a complex chain and the index of a chain). For n1, the function zn1 has the primitive zn/n on C{0}, so its integral over Γ vanishes (The integral of a continuous derivative over a cycle is zero). Also Γcdz=0 for every constant c, by summing endpoint increments over the cycle (The contour integral of a constant c is c times the endpoint displacement, Complex chains, their traces, and cycles).

[L7]

y<1 implies (1y)1=n0yn with the series converging in norm, and every convergent series on a compact C contour may be integrated termwise: if hNh uniformly on the trace then ΓhNdzΓhdz by the norm bound of [L1] (Neumann series).

Proof

technique · direct
1.1

Linearity: for a common admissible cycle Γ in UfUg one has (αf+βg)(a)=12πiΓ(αf(z)+βg(z))R(z,a)dz=αf(a)+βg(a), because the chain integral is C-linear in the integrand.

L1algebra
1.2

Unit law, cycle choice: choose r>a and apply [L5] to the compact closed disc K={z:zr} inside C. It gives a finite polygonal cycle Γr whose trace lies outside K and whose index is 1 on K. Since σA(a)K and the constant function 1 is entire, Γr is admissible for its calculus value; by contour independence [L1], 1(a) may be computed on Γr.

L1L5L9algebra
1.3

Coordinate identity: for every admissible cycle Γ and the function id(z)=z one has the pointwise identity zR(z,a)=1+aR(z,a) on Γ, hence id(a)=12πiΓdz+a12πiΓR(z,a)dz=0+a1(a) by [L6] and [L1].

L1L6algebra
1.4

Nested cycles: apply [L5] to the compact set K:=σA(a) and the open set UfUg; this produces cycles β,γ with disjoint traces in (UfUg)σA(a), both admissible for f and for g, with n(γ,w)=1 for every wβ and n(β,z)=0 for every zγ.

L5algebra
2.1

First Cauchy integral: for each fixed wβ the scalar function g is holomorphic on Ug, and γ is a cycle with trace in Ug that is null-homologous in Ug because its index vanishes outside Ug by admissibility; [L3] gives 12πiγg(z)zwdz=n(γ,w)g(w)=g(w).

step 1.4L3
2.2

Second Cauchy integral: for each fixed zγ the function wf(w)zw is holomorphic on Uf{z}, a neighbourhood of β; and β is null-homologous in Uf{z}, because n(β,p)=0 for every pUf by admissibility and n(β,z)=0 by the nesting; hence [L4] gives 12πiβf(w)zwdw=0.

step 1.4L4algebra
2.3

Unit law, value: the compact trace of Γr lies in the open set {z>r}, so q:=maxzΓra/z<1. Hence R(z,a)=1z(1a/z)1=n0anzn1 uniformly on the trace. Integrating termwise by [L7], [L6] gives Γrz1dz=2πin(Γr,0)=2πi and Γrzn1dz=0 for n1. Thus 12πiΓrR(z,a)dz=1 and 1(a)=1.

step 1.2L6L7algebra
3.1

The double integral: the function H(w,z):=f(w)g(z)R(w,a)R(z,a) is continuous on the compact product β×γ; the two-dimensional tagged Riemann sums of H over refined partitions of β and γ converge in A, by the uniform-continuity mesh estimate underlying the Banach-valued contour integral in [L1] applied in both variables, so the two iterated integrals 12πiβ(12πiγHdz)dw and 12πiγ(12πiβHdw)dz exist and agree.

step 1.4step 2.1L1algebra
3.2

Coordinate law, value: combining [step 1.3] with [step 2.3] gives id(a)=a1=a.

step 1.3step 2.3algebra
4.1

Splitting the double integral: by the resolvent identity [L2], R(w,a)R(z,a)=(R(w,a)R(z,a))/(zw) for wβ, zγ, so the double integral of [step 3.1] splits into the sum of the iterated integrals of f(w)g(z)R(w,a)/(zw) and of f(w)g(z)R(z,a)/(zw); the first inner integral over γ equals g(w)R(w,a) by [step 2.1], and the second inner integral over β equals 0 by [step 2.2].

step 2.1step 2.2step 3.1L2algebra
5.1

Multiplicativity: using [step 4.1], f(a)g(a)=12πiβf(w)R(w,a)(12πiγg(z)zwdz)dw12πiγg(z)R(z,a)(12πiβf(w)zwdw)dz=12πiβf(w)g(w)R(w,a)dw=(fg)(a); here each resolvent factor commutes with the scalar coefficient in front of it. This is claim 2.

step 2.1step 2.2step 4.1L1
6.1

Inverse compatibility: for h holomorphic and nowhere zero on UhσA(a) the reciprocal 1/h is holomorphic on Uh and h(1/h)=1 there; by [step 5.1] and [step 2.3], h(a)(1/h)(a)=1(a)=1 and symmetrically (1/h)(a)h(a)=1, so h(a) is invertible with h(a)1=(1/h)(a); this is claim 5.

step 2.3step 5.1algebra
7.1

Polynomials and conclusion: a constant function zc is c1, so its calculus value is c1(a)=c1 by [step 1.1] and [step 2.3]; the coordinate function has value a by [step 3.2]; multiplicativity [step 5.1], linearity [step 1.1] and induction on the degree therefore assemble p(a)=k=0nckak for every polynomial, which is claim 4. Claims 1, 2, 3 and 5 were proved in [step 1.1], [step 5.1], [step 2.3], [step 3.2] and [step 6.1].

step 1.1step 2.3step 3.2step 5.1step 6.1
TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Holomorphic spectral mapping and composition

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra and let aA. Let f be holomorphic on an open set UσA(a), so that f(a) is defined (Holomorphic functional calculus). Then:

  1. spectral mapping: σA(f(a))=f(σA(a))={f(λ):λσA(a)};
  2. composition: if V is an open set with f[U]V and g:VC is holomorphic, then g(f(a))=(gf)(a), where (gf)(a) is computed from the holomorphic function gf:UC by the calculus on A.

Clause 1 includes locally constant functions: if f is constant on a component of σA(a) its image there is a single point, and no connectedness of U or of σA(a) is assumed.

Facts & Assumptions

Given: An assumed Axiom of Choice, a unital complex Banach algebra A, an element aA, an open UσA(a), a holomorphic f:UC, and for clause 2 an open Vf[U] with a holomorphic g:VC.

[L1]

The calculus hh(a) is linear, multiplicative, unital, sends the coordinate function to a and satisfies h(a)1=(1/h)(a) for nowhere vanishing holomorphic h (Holomorphic functional calculus homomorphism, Holomorphic functional calculus).

[L2]

For holomorphic f and fixed z the filled difference quotient ζg(ζ,z):=(f(ζ)f(z))/(ζz) for ζz, extended by f(z) at ζ=z, is holomorphic in each variable on U; in particular hλ(ζ):=g(ζ,λ) is holomorphic on U with f(ζ)f(λ)=(ζλ)hλ(ζ) (The filled difference quotient is holomorphic in each variable separately).

[L3]

If u,v commute in A and uv is invertible then so are u and v: with w:=(uv)1 one has u(vw)=1=(vw)u, and symmetrically for v (Spectrum and resolvent set in a Banach algebra).

[L4]

Nested and encircling cycles exist as in Admissible cycle around a compact plane set: for compact K inside open W there is a cycle with index 1 on K and 0 outside W, and two such with disjoint traces and nesting. For a cycle Γ of this kind the set KΓ:=Γ{wC:n(Γ,w)0} is compact: it is closed and bounded because the index vanishes far from the trace (The index of a cycle is locally constant off its trace and vanishes far from it).

[L5]

Cauchy formula on a cycle: for holomorphic h on open Ω and a cycle Γ with trace in Ω null-homologous in Ω, n(Γ,p)h(p)=12πiΓh(ζ)/(ζp)dζ for pΩΓ, and the double integral of a continuous integrand over two such cycles may be iterated in either order (Cauchy's integral formula for a null-homologous cycle, Resolvent identity, the mesh estimate of Banach algebra valued contour integral).

Proof

technique · direct
1.1

Factorization at a spectral point: for λσA(a) the function hλ of [L2] is holomorphic on U and f(z)f(λ)=(zλ)hλ(z) on U; applying the calculus and its multiplicative and affine laws [L1] gives f(a)f(λ)1=(aλ1)hλ(a), a product of two commuting elements.

L2L1algebra
1.2

Reverse inclusion: if μf(σA(a)) then f(z)μ0 for every zσA(a), so 1/(fμ) is holomorphic on some neighbourhood of σA(a); by [L1] applied to the two functions fμ and 1/(fμ), whose product is the constant function 1, one has (f(a)μ1)(1/(fμ))(a)=1=(1/(fμ))(a)(f(a)μ1), so μσA(f(a)).

L1algebra
2.1

Forward inclusion: let λσA(a). If f(a)f(λ)1 were invertible, then by [step 1.1] the commuting product (aλ1)hλ(a) would be invertible, so [L3] would make aλ1 invertible, contradicting λσA(a); hence f(λ)σA(f(a)).

step 1.1L3
3.1

Clause 1 follows from [step 1.2] and [step 2.1]: f(σA(a))σA(f(a))f(σA(a)).

step 1.2step 2.1algebra
4.1

Setup for clause 2: choose a cycle β with trace in UσA(a) and index 1 on σA(a), 0 outside U, by [L4]; then K:=Kβ is a compact subset of U containing σA(a), and f[K]V is compact. Since σA(f(a))=f(σA(a))f[K] by [step 3.1], the calculus applies to g at f(a).

step 3.1L4L1algebra
5.1

The resolvent identity in integral form: for every zf[K] the function w1zf(w) is holomorphic on a neighbourhood of σA(a) (namely on Uf1({z}), which contains K and hence σA(a)), and zf(w)0 there; by [L1] applied to w1zf(w) and the affine function zf, one has (z1f(a))1=12πiβR(w,a)zf(w)dw: both sides are the calculus of reciprocal functions whose product with zf is 1.

step 4.1L1L2algebra
5.2

Choice of the outer cycle and Cauchy evaluation: apply [L4] to the compact set f[K] inside V, obtaining a cycle γ with index 1 on f[K] and 0 outside V; then for every wβ (so f(w)f[K]) the Cauchy formula [L5] applied to g on V along γ gives 12πiγg(z)zf(w)dz=n(γ,f(w))g(f(w))=g(f(w)).

step 4.1L5algebra
6.1

Composition: using the definition of the calculus, [step 5.1] inside the outer integral, and the iterated-integral identity of [L5], g(f(a))=12πiγg(z)(z1f(a))1dz=(12πi)2γβg(z)R(w,a)zf(w)dwdz=12πiβ(12πiγg(z)zf(w)dz)R(w,a)dw=12πiβg(f(w))R(w,a)dw=(gf)(a), where the second-to-last equality is [step 5.2] and the last is the calculus of gf along β.

step 5.1step 5.2L5L1
7.1

Both clauses are proved: clause 1 by [step 3.1] and clause 2 by [step 6.1].

step 3.1step 6.1

Remarks

  • The composition clause is the coverage's inline obligation. The composition law g(f(a))=(gf)(a) is the second half of Bühler–Salamon Theorem 5.25(v) and of Shirbisheh Theorem 2.5.5; it is proved here, after the spectral mapping statement it needs, and not merely cited. The proof requires cycles around the compact image f[K] of a bounded spectral neighbourhood, not the whole preimage of an outer contour.

  • Local constancy of f on spectral components is allowed. Nothing in the argument uses that f separates points of σA(a): the factorization of [step 1.1] is carried out at the single spectral point λ, and the reverse inclusion tests values of f on the spectrum pointwise.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Riesz spectral projection

Definition

Assume the Axiom of Choice (The Axiom of Choice). Let A be a unital complex Banach algebra, let aA, and let EσA(a) be clopen in the spectrum, that is, both E and σA(a)E are relatively open in σA(a) (Spectrum and resolvent set in a Banach algebra). Equivalently EσA(a) is closed in C — hence compact — and σA(a)E is compact as well, and the two are disjoint.

Choose disjoint open sets U1E and U0σA(a)E; such sets exist because E and σA(a)E are disjoint compact subsets of the plane. Let

χE(z):={1,zU1,0,zU0,

a locally constant function on the open neighbourhood U1U0 of σA(a), hence holomorphic there. The Riesz spectral projection of a associated with E is the calculus value

PE  :=  χE(a)  =  12πiΓχE(z)R(z,a)dz    A,

where Γ is any cycle with trace in (U1U0)σA(a) whose index is 1 at every point of E, whose index is 0 at every point of σA(a)E, and whose index is 0 outside U1U0 — for instance the difference callcE, where call is admissible for (χE,U1U0) and cE is a cycle with index 1 on the compact set σA(a)E and index 0 outside U0 (the zero cycle when σA(a)E=): the difference has index 10=1 on E, index 11=0 on σA(a)E, and index 0 outside U1U0, because call has index 0 there and cE has index 0 outside U0U1U0. Such cycles exist by Admissible cycle around a compact plane set applied to the two compact sets σA(a) and σA(a)E.

Here the equality with the displayed integral, and its independence of the separating cycle, do not use contour independence outside its admissible-cycle hypothesis. Indeed, put F(z)=χE(z)R(z,a) on Ω:=(U1E)U0. This is Banach-valued holomorphic: it is R(z,a) on U1E and identically zero on U0, so in particular it extends holomorphically across σA(a)E. If call is admissible and Γ has the separating indices just specified, then callΓ has index zero on E and outside U1U0, hence is null-homologous in Ω. Therefore Banach-valued Cauchy integral vanishes gives callF(z)dz=ΓF(z)dz. The left side is the defining calculus integral for χE(a). Thus every such Γ gives PE, while germ independence of the calculus makes the value independent of the chosen U0,U1.

Remarks

  • The function χE is a germ, and that is all the definition needs. Its definition depends on the chosen neighbourhoods, but every two such locally constant functions agree on a neighbourhood of σA(a), and the calculus depends only on the germ (Holomorphic functional calculus is contour independent).

  • When E=σA(a) or E=. If E=σA(a) then χE=1 on a neighbourhood of the spectrum and PE=1; if E= then χE=0 on a neighbourhood of the spectrum and PE=0. Both are consistent with the definition and with the multiplicativity of the calculus (Holomorphic functional calculus homomorphism).

  • No idempotence is assumed here. That PE2=PE and that PE commutes with a are consequences of multiplicativity of the calculus, not part of the definition; they are proved for operators in Riesz spectral projection properties.

  • Why the contour has index one on E and zero on the rest of the spectrum. This makes the integral a function of the spectral subset E alone: replacing the cycle by another with the same indices does not change the value, as in the calculus at large. For a single isolated eigenvalue λ the projection is the classical residue (ex-riesz-projection-for-a-matrix-with-separated-spectrum).

  • Reading order. The example items named by ID above are homed on later pages of the plan, so they are named rather than hyperlinked: a body link to later material must be declared as a forward reference, and Step-5b closure removes every such declaration. Rehoming those items to an earlier page (an owner-only reading-order change) would make the citations backward and restore the links.

TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Riesz spectral projection properties

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let X be a nonzero complex Banach space, let TB(X), and let Eσ(T) be clopen in the spectrum, with Riesz projection P:=PEB(X) (Riesz spectral projection). Then:

  1. P2=P and PT=TP; consequently the range and the kernel of P are closed T-invariant subspaces and X=ran(P)ker(P);
  2. if ran(P){0}, then the restriction Tran(P) has spectrum E;
  3. if ker(P){0}, then the restriction Tker(P) has spectrum σ(T)E;
  4. if one of these spectral parts is empty, the corresponding summand is the zero subspace and no spectrum is assigned to the zero operator on it under the normalized nonzero-algebra convention of Unital Banach algebra.

Facts & Assumptions

Given: An assumed Axiom of Choice, a nonzero complex Banach space X, a bounded operator TB(X), a clopen subset E of the spectrum σ(T)=σB(X)(T), the locally constant germ χE, and P=χE(T).

[L1]

The calculus is linear, multiplicative and unital: (fg)(T)=f(T)g(T), 1(T)=1, id(T)=T, and for nowhere vanishing h, h(T)1=(1/h)(T) (Holomorphic functional calculus homomorphism, Holomorphic functional calculus).

[L2]

χE2=χE and χEid=idχE as germs near σ(T) (Riesz spectral projection).

[L3]

For a bounded idempotent P on a normed space the range and kernel are closed and X=ran(P)ker(P) (A closed subspace is complemented exactly when it is the range of a bounded projection).

[L4]

The spectrum σB(X)(T) consists exactly of those μC for which Tμ1 is not invertible in B(X) (Spectrum and resolvent set in a Banach algebra, A bounded linear operator between normed spaces).

Proof

technique · direct
1.1

Idempotence and commutation: P2=χE(T)χE(T)=(χEχE)(T)=χE(T)=P by [L1] and [L2]; PT=χE(T)id(T)=(χEid)(T)=(idχE)(T)=id(T)χE(T)=TP by [L1] and [L2].

L1L2
2.1

The splitting: by [step 1.1] the operator P is a bounded projection, so [L3] gives that ran(P) and ker(P) are closed with X=ran(P)ker(P); since T commutes with P, both summands are T-invariant.

step 1.1L3
2.2

Range spectrum, exclusion: assume ran(P){0}, as in claim 2, so its operator spectrum is defined. Let μE. Choose the neighbourhoods U1E, U0σ(T)E of the definition so that μU1 (possible since μE and E is compact). Then the germ h:=χE/(idμ) is holomorphic near σ(T): on U1 it is 1/(zμ) with μU1, and on U0 it is 0. By [L1], P=(Tμ)h(T)=h(T)(Tμ). The operator h(T) commutes with P, so it preserves ran(P); on that nonzero summand its restriction is a two-sided inverse of Tμ. Hence μσ(Tran(P)).

step 1.1L1
2.3

Kernel spectrum, exclusion: assume ker(P){0}, as in claim 3, so its operator spectrum is defined. Let μσ(T)E, that is, μE or μσ(T). Choose U0,U1 with μU0 when μE, and define q:=(1χE)/(idμ) on the complement part. Then 1P=(1χE)(T), and the same computation with 1χE in place of χE shows that Tμ has the restriction of q(T) as a two-sided inverse on ker(P). Hence μσ(Tker(P)).

step 1.1L1
3.1

Range spectrum, inclusion: continue under ran(P){0}. Let μE and suppose that Tran(P)μ were invertible on ran(P), with inverse S. Put H:=1χEidμ(T), a bounded operator because the germ is holomorphic near σ(T) (its numerator vanishes on U1E and μσ(T)E), and put V:=H+SP on X=ran(P)ker(P). Then (Tμ)V=(1P)+P=1 and V(Tμ)=1 by the same multiplicativity computation, so Tμ would be invertible on X, contradicting μEσ(T). Hence μσ(Tran(P)).

step 2.1step 2.2L1L4
3.2

Kernel spectrum, inclusion: continue under ker(P){0}. Symmetrically, if μσ(T)E and Tker(P)μ were invertible with inverse S, then the germ χEidμ, read as 1zμ on a neighbourhood of E avoiding μ and as 0 on a neighbourhood of σ(T)E, is holomorphic near σ(T). The operator V:=(χEidμ)(T)+S(1P) would be a two-sided inverse of Tμ on X, contradicting μσ(T). Hence μσ(Tker(P)).

step 2.1step 2.3L1L4
4.1

Claims 2 and 3 follow from [step 2.2], [step 3.1] and [step 2.3], [step 3.2] respectively; claim 1 was proved in [step 1.1] and [step 2.1]; claim 4 is the convention recorded in the statement, applied to an empty spectral part, and no spectrum is claimed for the zero operator.

step 1.1step 2.1step 2.2step 3.1step 3.2
DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Calkin algebra

Definition

Assume the Axiom of Countable Choice (The Axiom of Countable Choice (ACω)). Let X be an infinite-dimensional complex Banach space and let B(X) be the Banach algebra of bounded operators (If (Y) is Banach then (\mathcal B(X,Y)) is Banach, The operator norm as the least bound and as the unit-sphere or unit-ball supremum). Let K(X)B(X) be the set of compact operators (Compact linear operator). Then:

The Calkin algebra of X is the quotient algebra

C(X)  :=  B(X)/K(X),

with the quotient vector-space structure, the quotient norm A+K(X):=inf{AK:KK(X)}, and the multiplication (A+K)(B+K):=AB+K. Multiplication is well-defined because K(X) is a two-sided ideal, and it is submultiplicative: for K1,K2K and representatives A,B, (A+K1)(B+K2)=AB+(AK2+K1B+K1K2) with the bracket in K, so taking infima gives (A+K)(B+K)A+KB+K. The quotient is complete for the quotient norm by A quotient of a Banach space by a closed subspace is Banach. Its unit is 1:=IX+K(X).

Remarks

  • The unit has norm one and is nonzero. The quotient norm satisfies IX+KIX=1. The quotient norm is definite because K is closed, and IXK, so c=IX+K>0. The unit is idempotent and the quotient norm is submultiplicative, giving cc2. Division by c yields c1, hence c=1.

  • Finite-dimensional X is excluded, not normalized away. If dimX< then every bounded operator has finite-dimensional range and is compact (Bounded finite rank operators are compact), K(X)=B(X) and the quotient is the zero algebra, which carries no unit in the sense of Unital Banach algebra. The definition therefore restricts to infinite-dimensional X; the finite-dimensional case is the zero quotient and is not called a Calkin algebra here.

  • Where Countable Choice is spent. It is used in the norm-limit compactness theorem, in selecting the approximating sequence above to turn sequential closure into norm closure, and in quotient completeness. The ideal formulas, identity noncompactness and the unit-norm argument introduce no further choice beyond those supplied closed-quotient facts.

  • The Calkin algebra forgets compact perturbations. Two operators have the same coset exactly when they differ by a compact operator, so C(X) records the "Fredholm part" of B(X); this is what makes the Atkinson theorem a statement about invertibility in C(X) (Atkinson in Calkin algebra language).

CorollaryStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Atkinson in Calkin algebra language

Statement

Assume the Axiom of Choice (The Axiom of Choice). Let X be an infinite-dimensional complex Banach space and let TB(X). Then the following are equivalent:

  1. T is Fredholm;
  2. the coset T+K(X) is invertible in the Calkin algebra C(X)=B(X)/K(X) (Calkin algebra);
  3. there is a bounded SB(X) with STIX and TSIX both compact (Compact linear operator) — a bounded two-sided parametrix modulo compact operators.

Facts & Assumptions

Given: An assumed Axiom of Choice, an infinite-dimensional complex Banach space X, a bounded operator TB(X), and the Calkin algebra C(X) with unit 1=IX+K(X).

[L1]

T is Fredholm if and only if there is a bounded linear S with STIX and TSIX compact (Atkinson).

[L2]

In the quotient algebra C(X) one has T+K invertible if and only if there is S+K with (T+K)(S+K)=1 and (S+K)(T+K)=1; these equations are exactly TSIXK(X) and STIXK(X) (Calkin algebra).

[L3]

The Calkin algebra is built under Countable Choice, which is available here because the standing hypothesis is the stronger Axiom of Choice (The Axiom of Choice), whose standard consequences include Countable Choice (The Axiom of Countable Choice (ACω)).

Proof

technique · direct
1.1

Equivalence of 2 and 3: the equation (T+K)(S+K)=IX+K of [L2] holds exactly when TSIXK(X), and the other product equation holds exactly when STIXK(X); so T+K is invertible precisely when T admits a bounded two-sided parametrix modulo compact operators.

L2L3
2.1

Equivalence of 1 and 3 is [L1]; combining it with [step 1.1] gives that 1, 2 and 3 are equivalent.

step 1.1L1

Remarks

  • No new Fredholm theory is hidden here. The corollary is a restatement of the Atkinson theorem in the quotient algebra: the only content beyond Atkinson is that quotient invertibility and the existence of a two-sided parametrix modulo compact operators are the same condition, which is the definition of the quotient multiplication.

  • Both products are required. One-sided quotient invertibility would only give one of the two compactness conditions; the theorem and the definition both ask for two-sided invertibility, and the remark here records that the order of the products is preserved: ST and TS appear in the two conditions separately.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Point continuous and residual spectrum

Definition

Let X be a nonzero complex Banach space, let TB(X) be a bounded operator (A bounded linear operator between normed spaces) and let λC. Write Tλ for TλIX, where IX is the identity, and recall that λσ(T) exactly when Tλ is not invertible (Spectrum and resolvent set in a Banach algebra).

The spectral value λ is

  • in the point spectrum σp(T) when Tλ is not injective, that is, when λ is an eigenvalue;
  • in the continuous spectrum σc(T) when Tλ is injective, has dense range, and is not surjective;
  • in the residual spectrum σr(T) when Tλ is injective and its range is not dense in X.

The three sets are pairwise disjoint by their injectivity and density conditions, and each is contained in σ(T): every listed condition precludes a two-sided inverse in B(X).

Assume additionally Dependent Choice (The axiom of dependent choice: a relation in which every element is related to something admits an N-indexed chain) for the partition assertion. If Tλ is injective and surjective, the bounded inverse theorem Bounded inverse theorem makes its inverse bounded. Consequently, for a spectral value with injective dense range, surjectivity is impossible. Splitting first by injectivity and then by density therefore gives σp(T)σc(T)σr(T)=σ(T)under DC. The definitions themselves do not require DC. For T=0 on the nonzero space, σp(T)={0} and the other two parts are empty, since λI has inverse λ1I for λ0.

Remarks

  • The residual spectrum is the injective case with non-dense range. No closedness is presupposed: λσr(T) simply means that Tλ is injective and its range is not dense in X. The reader should not add the hypothesis that the range be closed; the point of the definition is to separate dense range from non-dense range, and a non-dense range may still fail to be closed.

  • Eigenvalues belong only to the point spectrum. If Tλ is not injective, then λσp(T) and, whatever the range is, λ is not in σc(T) or σr(T) by the disjoint classification above. Consequently σr=σcpσp: after compression values that are eigenvalues are removed, the remaining operators are exactly the injective ones with non-dense range. This identity is recorded and proved in Relations among the five spectral parts rather than identifying σr with the whole compression spectrum.

  • Individual parts need not be closed. The spectrum is closed, but the point spectrum need not be, and the closures of the three disjoint parts may meet at accumulation points of the whole spectrum; under DC their union is the closed spectrum by the partition argument above.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-22Open item page →

Approximate point and compression spectrum

Definition

Assume the Axiom of Countable Choice (The Axiom of Countable Choice (ACω)). Let X be a nonzero complex Banach space and let TB(X) (A bounded linear operator between normed spaces). Write Tλ for TλIX. The approximate point spectrum of T is

σap(T)  :=  {λC:Tλ is not bounded below},

bounded below meaning (Tλ)xcx for all x and some real c>0 (A bounded operator that is bounded below), and the compression spectrum of T is

σcp(T)  :=  {λC:ran(Tλ)X},

the set of λ for which Tλ does not have dense range.

Sequential description of the approximate point spectrum. Under Countable Choice, λσap(T) if and only if there is a sequence (xn) of unit vectors in X with (Tλ)xn0. Indeed, if Tλ is not bounded below then for each n the set {x:x=1, (Tλ)x<1/(n+1)} is nonempty, and Countable Choice selects one unit vector xn for each n; the resulting sequence witnesses the failure of the bound. Conversely a sequence of unit vectors with (Tλ)xn0 rules out every constant c>0 in the estimate.

Remarks

  • The two sets are not spectral parts in the disjoint sense. They may overlap each other and the classical parts: an eigenvalue is in σap but may also be a compression value. A value whose range is dense but not closed is not in σcp and cannot be bounded below; indeed, under Countable Choice a convergent sequence of range points has Cauchy preimages under a lower bound, and completeness then puts its limit back in the range. Thus the value lies in σap. The exact relations are proved in Relations among the five spectral parts; no disjointness is claimed here.

  • The normalization of approximate eigenvectors matters. The definition above demands unit vectors, so a sequence xn0 with (Tλ)xn0 is not evidence: without the normalization every bounded operator would qualify. The unit vectors may be chosen adaptively; the single use of Countable Choice is recorded in the display above.

  • Bounded below is exactly injectivity with closed range in the Banach setting. That equivalence, proved under Dependent Choice in Under Dependent Choice, a bounded operator between Banach spaces is bounded below exactly when it is injective with closed range, is what makes σap and σcp cover the spectrum (Relations among the five spectral parts); the definition here does not assume it.

LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Relations among the five spectral parts

Statement

Assume the Axiom of Dependent Choice (The axiom of dependent choice: a relation in which every element is related to something admits an N-indexed chain). Let X be a nonzero complex Banach space and let TB(X), with the point, continuous and residual spectra of Point continuous and residual spectrum and the approximate point and compression spectra of Approximate point and compression spectrum. Then

  1. σp(T), σc(T) and σr(T) are pairwise disjoint and σp(T)σc(T)σr(T)=σ(T);
  2. σr(T)=σcp(T)σp(T) and σp(T)σap(T);
  3. σ(T)=σap(T)σcp(T).

The three classical parts are a partition of the spectrum; the approximate point and compression spectra cover it as well, but may overlap it and each other.

Facts & Assumptions

Given: An assumed Axiom of Dependent Choice, a nonzero complex Banach space X, a bounded TB(X) and λC; abbreviate Tλ:=TλIX.

[L1]

The three classical cases are exhaustive and exclusive for λσ(T): Tλ fails to be injective, or is injective with dense non-surjective range, or is injective with non-dense range (Point continuous and residual spectrum).

[L2]

λσap(T) when Tλ is not bounded below, and λσcp(T) when Tλ has non-dense range (Approximate point and compression spectrum).

[L3]

Under Dependent Choice, Tλ is bounded below if and only if it is injective with closed range (Under Dependent Choice, a bounded operator between Banach spaces is bounded below exactly when it is injective with closed range).

[L4]

λσ(T) exactly when Tλ is not invertible in B(X) (Spectrum and resolvent set in a Banach algebra).

[L5]

The Axiom of Dependent Choice is the standing hypothesis of the statement (The axiom of dependent choice: a relation in which every element is related to something admits an N-indexed chain).

Proof

technique · direct
1.1

If Tλ is not injective then it is not bounded below: a bounded-below Tλ satisfies cxTλx for c>0, so Tλx=0 forces x=0. Hence σp(T)σap(T), which is the second inclusion of claim 2.

L2L3L5algebra
1.2

If λσr(T) then, by definition, Tλ is injective with non-dense range. Hence λσcp(T) and λσp(T), so σr(T)σcp(T)σp(T). Conversely, if λσcp(T)σp(T), then Tλ has non-dense range and is injective, which is exactly λσr(T). Thus σr(T)=σcp(T)σp(T).

L1L2
1.3

If λσ(T) then Tλ is invertible, hence bounded below and of dense range; so λσap(T)σcp(T).

L2L4algebra
1.4

Claim 3, other inclusion: let λσ(T). If Tλ is not bounded below then λσap(T). If it is bounded below, then by [L3] it is injective with closed range; were the range also dense, closedness would give range =X, so Tλ would be bijective with bounded inverse, hence invertible, contradicting [L4]; therefore the range is not dense and λσcp(T).

L2L3L4
2.1

Claim 3, one inclusion: σap(T)σcp(T)σ(T) is the contrapositive of [step 1.3].

step 1.3
3.1

Claim 1: for λσ(T) the operator Tλ is not invertible by [L4]; by [L1] it falls into exactly one of the three classical cases, so the three sets are disjoint and their union is σ(T). The covering half of claim 3 is [step 1.4] and the reverse half is [step 2.1]; claim 2 consists of [step 1.1] and [step 1.2].

step 1.1step 1.2step 1.4step 2.1L1L4
TheoremStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-22Open item page →

Boundary of spectrum lies in approximate point spectrum

Statement

Assume the Axiom of Countable Choice (The Axiom of Countable Choice (ACω)). Let X be a nonzero complex Banach space and let TB(X). Then every boundary point of the spectrum lies in the approximate point spectrum:

σ(T)    σap(T)

(spectrum as in Spectrum and resolvent set in a Banach algebra, approximate point spectrum as in Approximate point and compression spectrum). The boundary is taken in C.

Facts & Assumptions

Given: An assumed Axiom of Countable Choice, a nonzero complex Banach space X, a bounded TB(X) and a point λσ(T).

[L1]

σ(T) is closed, so λσ(T); and every neighbourhood of λ meets the resolvent set ρ(T)=Cσ(T) (Spectrum and resolvent set in a Banach algebra, Spectrum is nonempty compact and norm bounded).

[L2]

For zρ(T) the resolvent R(z):=(zT)1 is bounded, and 1=(zT)R(z)=R(z)(zT) (Spectrum and resolvent set in a Banach algebra).

[L3]

The invertible group of B(X) is open: if A is invertible and AB<1/A1, then B is invertible (Invertible group is open and inversion is continuous). Here one may take A=Tλn, whose inverse is R(λn) by [L2], and B=Tλ, for which AB=λnλ.

[L4]

λσap(T) exactly when Tλ is not bounded below; a bounded-below operator satisfies (Tλ)xcx for all x and some c>0 (A bounded operator that is bounded below, Approximate point and compression spectrum).

[L5]

The Axiom of Countable Choice is the standing hypothesis, used to select the unit vectors below (The Axiom of Countable Choice (ACω)).

Proof

technique · direct
1.1

Since λ is a boundary point, for each n1 the disc {zλ<1/n} meets ρ(T); Countable Choice selects λnρ(T) with λnλ<1/n for every n; in particular λnλ.

L1L5algebra
1.2

First suppose that the resolvent norms are bounded along this sequence, say R(λn)M for all n, and suppose M1. For n with λnλ<1/(2M) the operator Tλ=(Tλn)+(λnλ)=(Tλn)(1(λnλ)R(λn)) is a product of invertible factors: the displayed identity holds because (Tλn)R(λn)=1 by [L2], and the second factor is invertible by the Neumann series since (λnλ)R(λn)1/2<1. Hence Tλ would be invertible and λρ(T), contradicting λσ(T).

L2L3L1algebra
2.1

Consequently the norms R(λn) are unbounded; passing to a subsequence, which we relabel, we may assume R(λn).

step 1.1step 1.2algebra
3.1

For each n the set of unit vectors x with R(λn)x12R(λn) is nonempty, because the operator norm is the supremum of R(λn)x over the unit sphere; Countable Choice selects such a unit vector xn for every n, and we set yn:=R(λn)xn/R(λn)xn, a unit vector.

step 2.1L5algebra
4.1

Then (Tλn)yn=xn/R(λn)xn2/R(λn)0 and λnλ0, so (Tλ)yn(Tλn)yn+λnλ0 along unit vectors.

step 3.1L2algebra
5.1

By [L4] such a sequence rules out Tλ being bounded below with any constant c>0; hence λσap(T). Since λσ(T) was arbitrary, σ(T)σap(T).

step 4.1L4

5 · Examples, counterexamples and false statements

None yet.

Sources