Kasparov's stabilization theorem

Written by GPT-6.1 Sol (OpenAI) in Codex, Ultra setting, October 2026. Self-checked by the writing AI. Public domain (CC0).

A countably generated Hilbert module need not have an orthonormal basis, and its closed submodules need not be complemented. Stabilization provides a different way to use coordinates: after adding the standard module, every countably generated module becomes standard. The resulting embedding has an adjoint and an orthogonal complement. It is obtained from a carefully chosen dense-range operator, rather than from a general assertion that Hilbert-module operators have polar decompositions.

The prerequisites are Hilbert C*-modules, Adjointable operators, Compact operators, multipliers and the strict topology, and Tensor products and C*-correspondences. Inner products are linear in the second variable. All algebras may be nonunital. Write $H_A$ for the completion of the finite columns over $A$, and $E^\infty=\bigoplus_{j\geq1}E$ for the analogous countable orthogonal sum.

“Countably generated” means that the $A$-linear span of some sequence of vectors is dense. It does not mean that this span already equals the module. It also does not mean separability as a complex Banach space: a nonseparable unital algebra is generated by its unit as a module over itself.

1. Strict positivity and dense range

For a C*-algebra $C$, a positive element $h$ is strictly positive when $\overline{hC}=C$. This is equivalent to saying that the positive contractions $f_\varepsilon(h)=h(h+\varepsilon1)^{-1},\qquad \varepsilon>0, \tag{1.1}$ form an approximate identity as $\varepsilon\downarrow0$. Inverses are taken in a unitization.

Indeed, if $hC$ is dense, then $\|(1-f_\varepsilon(h))h\|\leq\varepsilon$ gives convergence on $hC$, hence on $C$ by contractivity; taking adjoints gives convergence on the other side. Conversely $f_\varepsilon(h)c=h((h+\varepsilon1)^{-1}c)$ belongs to $hC$, since $C$ is an ideal in its unitization. Approximation of every $c$ proves density. For the zero algebra we allow zero as a strictly positive element.

Proposition 1.1. For $h\in\mathcal K(E)^+$, strict positivity in $\mathcal K(E)$ is equivalent to density of $hE$ in $E$.

Proof. Put $C=\mathcal K(E)$. Its action on $E$ is nondegenerate by the compact-action lemma. If $hC$ is dense in $C$, approximation of the coefficients in finite sums $c_ix_i$ shows $\overline{hE}\supseteq\overline{hCE} =\overline{CE}=E.$ Conversely, for $x,y\in E$, choose $z_n$ with $hz_n\to x$. Then $h\theta_{z_n,y}=\theta_{hz_n,y}\longrightarrow\theta_{x,y}.$ Thus $hC$ is dense in the finite-rank span, hence in $C$. ∎

The spectral approximation (1.1) also converges pointwise on $E$ whenever $hE$ is dense: convergence holds on $hE$ by the same scalar estimate and extends by contractivity.

Theorem 1.2. A Hilbert $A$-module $E$ is countably generated if and only if $\mathcal K(E)$ contains a strictly positive element.

Proof. Normalize a sequence of generators $x_j$ to have norm at most one, omitting zero vectors, and set $h=\sum_{j\geq1}2^{-j}\theta_{x_j,x_j}. \tag{1.2}$ The series converges in norm and is positive compact. Put $r_\varepsilon=1-f_\varepsilon(h)$. Positivity and the rank-one norm identity give $\begin{aligned} \|r_\varepsilon x_j\|^2 &=\|\theta_{r_\varepsilon x_j,r_\varepsilon x_j}\|\\ &\leq2^j\|r_\varepsilon h r_\varepsilon\| \leq 2^{j-2}\varepsilon. \end{aligned} \tag{1.3}$ The last inequality is the scalar estimate $\sup_{t\geq0}\varepsilon^2t/(t+\varepsilon)^2=\varepsilon/4$. Thus $f_\varepsilon(h)x_j\to x_j$. Module linearity, density of the generated span and contractivity give convergence on every vector. Each $f_\varepsilon(h)x$ lies in $hE$, so Proposition 1.1 makes $h$ strictly positive.

Conversely, if $h$ is strictly positive, choose finite-rank operators $k_n$ within $1/n$ of $f_{1/n}(h)$. Write $k_n=\sum_{i=1}^{m_n}\theta_{v_{ni},w_{ni}}$. The preceding pointwise approximation gives $k_nx\to x$ for every $x$. Each $k_nx$ belongs to the $A$-span of the $v_{ni}$; their countable union generates $E$. The zero module is immediate. ∎

This proves the strict-positivity formulation directly. The preceding compact-operator lesson also proved its equivalent formulation: $E$ is countably generated exactly when $\mathcal K(E)$ is $\sigma$-unital. No $\sigma$-unitality of $A$ has been assumed.

2. The dense-range isometry

We need two elementary operator facts. They identify exactly which density statements replace the closed-range theorem in stabilization.

Lemma 2.1 (Positive domination). Let $b,d\in\mathcal L(M)^+$, with $b\geq d$ and $dM$ dense in $M$. Then $bM$ is dense.

Proof. Put $r_\varepsilon=\varepsilon(b+\varepsilon1)^{-1}$. Compression of $d\leq b$ gives $\|r_\varepsilon d^{1/2}\|^2 =\|r_\varepsilon d r_\varepsilon\| \leq\|r_\varepsilon b r_\varepsilon\| \leq\varepsilon/4. \tag{2.1}$ Consequently $r_\varepsilon y\to0$ on $d^{1/2}M$. That range is dense, because it contains $dM$. The contractions $r_\varepsilon$ therefore tend pointwise to zero on all of $M$. But $(1-r_\varepsilon)y=b(b+\varepsilon1)^{-1}y$ belongs to $bM$, proving its density. ∎

Lemma 2.2 (Dense polar unitary). Suppose $T\in\mathcal L(M,N)$, both $TM$ and $(T^*T)M$ are dense in their respective modules. There is a unitary $U:M\to N$ with $T=U(T^*T)^{1/2}. \tag{2.2}$

Proof. Set $B=T^*T$. On $B^{1/2}M$, define $U(B^{1/2}x)=Tx.$ This is well defined and preserves inner products, because $\langle Tx,Ty\rangle=\langle B^{1/2}x,B^{1/2}y\rangle$. In particular a zero representative has zero image. The domain is dense since it contains $BM$. The isometry extends to $M$; its range contains $TM$, so it is dense in $N$. A completed isometry has closed range, hence is onto. Its inverse satisfies $\langle Ux,z\rangle=\langle x,U^{-1}z\rangle$, proving that it is the adjoint. Thus $U$ is unitary and (2.2) holds. ∎

Injectivity alone would not justify either density hypothesis in this lemma. In a Hilbert module, zero orthogonal complement does not imply density. The proof below establishes the required density by explicit approximations.

3. Absorption of a countably generated module

Theorem 3.1 (Kasparov stabilization). For every C*-algebra $A$ and every countably generated Hilbert $A$-module $E$, $E\oplus H_A\cong H_A \tag{3.1}$ by a unitary module map.

Proof. First suppose $A$ is unital. Choose generators $\eta_j$ of norm at most one, repeating each member of an original countable generating family infinitely often. Let $e_j\in H_A$ be the column with unit in coordinate $j$. Define $T:H_A\to E\oplus H_A$ by the norm-convergent compact series $T=\sum_{j\geq1} \theta_{(2^{-j}\eta_j,\,4^{-j}e_j),\,e_j}. \tag{3.2}$ The $j$-th summand has norm at most $2^{-j}+4^{-j}$, so the series converges in the rectangular compact space. In particular $T(e_ja)=(2^{-j}\eta_ja,\,4^{-j}e_ja).$ Equivalently $Tx=(Rx,Dx)$, where $D=\operatorname{diag}(4^{-j})$ and $Rx=\sum_j2^{-j}\eta_jx_j$. Both maps are adjointable by (3.2), and $T^*T=R^*R+D^2\geq D^2. \tag{3.3}$ The range of $D^2$ contains every finite column: invert its finitely many nonzero scalar diagonal entries on the support of that column. Hence Lemma 2.1 shows that $T^*T$ has dense range.

We separately prove density of $TH_A$. Fix one original generator $\eta$. Along indices $j\to\infty$ for which $\eta_j=\eta$, $T(2^je_ja)=(\eta a,\,2^{-j}e_ja) \longrightarrow(\eta a,0) \tag{3.4}$ for every $a\in A$. Thus $\overline{TH_A}$ contains $E\oplus0$. Subtracting its vector $(2^{-j}\eta_ja,0)$ from $T(e_ja)$ shows that it contains $(0,4^{-j}e_ja)$, hence every finite column in the second summand. Their density proves $\overline{TH_A}=E\oplus H_A$. Lemma 2.2 now supplies the desired unitary $H_A\to E\oplus H_A$.

For nonunital $A$, let $A^+$ be its unitization. Extend the action on $E$ by $x(a+\lambda1)=xa+\lambda x$, retaining its $A$-valued inner product inside $A^+$. The norm is unchanged, so this is a complete Hilbert $A^+$-module. Its original countable generators still generate it densely. The unital argument gives $U:H_{A^+}\longrightarrow E\oplus H_{A^+}.$ Restrict to the closed submodules obtained by multiplying by the ideal $A$. Module linearity of $U,U^*$ ensures that they map onto one another. They are $\overline{H_{A^+}A}=H_A,\qquad \overline{(E\oplus H_{A^+})A}=E\oplus H_A. \tag{3.5}$ For the first equality, finite columns with entries in $A$ are products of finite unit columns by elements of $A$, and the closure of such columns is exactly $H_A$. The reverse inclusion follows because products have entries in $A$. The second equality also uses $\overline{EA}=E$. The restriction of $U$ is therefore a unitary implementing (3.1). ∎

This is the repeated-generator proof associated with Mingo and Phillips; the dense-range construction is explained in [Blackadar 1998, Theorem 13.6.2]. The diagonal term makes $T^*T$ have dense range, while the faster decay in that term makes the first-summand approximations (3.4) possible. Neither purpose follows just from choosing a bounded operator with dense range.

Corollary 3.2 (Multiplier corners). Every countably generated $E$ is unitarily isomorphic to $PH_A$ for a projection $P\in\mathcal L(H_A)=M(\mathbb K\otimes_{\min}A),$ and $\mathcal K(E)\cong P(\mathbb K\otimes_{\min}A)P. \tag{3.6}$

Proof. Let $V:E\oplus H_A\to H_A$ be the unitary from Theorem 3.1 and $Q$ the first-summand projection. Set $P=VQV^*$. Restricting $V$ identifies $E$ with $PH_A$. Conjugation sends each rank-one operator on that summand to the corresponding rank-one operator on $PH_A$.

Compression of a rank-one operator on $H_A$ gives $P\theta_{x,y}P=\theta_{Px,Py}$. Thus $P\mathcal K(H_A)P=\mathcal K(PH_A)$, by norm closure. The previous lessons proved $\mathcal K(H_A)=\mathbb K\otimes_{\min}A$ and $\mathcal L(H_A)=M(\mathcal K(H_A))$; substituting gives (3.6). ∎

The projection generally belongs to the multiplier algebra, not to the compact algebra. For $E=H_A$ with $A\ne0$, the identity is not compact. Also, over a non-$\sigma$-unital algebra, an arbitrary multiplier projection need not have countably generated range: $P=1$ already fails. The corollary supplies a projection from a countably generated module; it does not assert the converse without a condition on $A$.

4. Infinite amplification of a full module

The next consequence has a different hypothesis. We now require that $A$ be $\sigma$-unital and that $E$ be full. The additional fullness says that the module sees every coefficient of $A$.

Lemma 4.1 (The dual compact-map module). Put $C=\mathcal K(E)$ and $G=\mathcal K_A(E,A).$ Composition on the right and the inner product $\langle S,T\rangle_C=S^*T$ make $G$ a Hilbert $C$-module. If $E$ is full, multiplication on the output identifies $\mathcal K_C(G)\cong A,\qquad G\otimes_C E\cong A_A. \tag{4.1}$ The left $C$-action on $E$ in this tensor product is the canonical compact action.

Proof. Rectangular compact composition gives $S^*T\in C$; positivity and the norm identity $\|S^*S\|=\|S\|^2$ follow by regarding the maps as corners in $\mathcal L(E\oplus A)$. The operator-norm closed space $G$ is therefore complete in this Hilbert-module norm. Its other inner-product and right-action identities follow from operator composition.

For $x\in E$, let $S_x:E\to A$ be $S_xy=\langle x,y\rangle$, with adjoint $S_x^*a=xa$. This is bounded by Cauchy–Schwarz. It is compact even for nonunital $A$: for a positive approximate identity $e_\lambda$ of $A$, $e_\lambda S_x=\theta_{e_\lambda,x}:E\to A,\qquad \|S_x-e_\lambda S_x\|\leq\|x-xe_\lambda\|\longrightarrow0.$ Moreover, $\|S_x\|=\|x\|$, by the inner-product norm-duality formula. All rank-one maps $E\to A$ have the form $\theta_{a,x}=S_{xa^*}$, so the span of the $S_x$ is dense in $G$.

Left multiplication $L_aS=aS$ is adjointable on $G$, with adjoint $L_{a^*}$. For $S,T,V\in G$, $\theta^G_{S,T}(V)=S(T^*V)=(ST^*)V. \tag{4.2}$ Here $ST^*\in\mathcal K(A)=A$, under the earlier multiplication identification. In particular $S_xS_y^*a=\langle x,y\rangle a.$ Fullness makes the span of these coefficients dense in $A$. The representation $a\mapsto L_a$ is faithful: if $aS_x=0$ for all $x$, then $a\langle x,y\rangle=0$ for all $x,y$, hence $aA=0$, and an approximate identity gives $a=0$. It is thus isometric with closed range. Formula (4.2) and the dense coefficient span identify its range exactly with $\mathcal K_C(G)$.

Finally define $W(S\otimes x)=Sx$. Composition gives balancing, and $\langle x,S^*Ty\rangle_A=(Sx)^*(Ty)$ shows preservation of inner products on every finite sum. The image includes all $\langle z,x\rangle=S_zx$, whose span is dense in $A$ by fullness. Thus $W$ extends to an onto isometry, hence a unitary. ∎

Theorem 4.2 (Full amplification). If $A$ is $\sigma$-unital and $E$ is full and countably generated, then $E^\infty\cong H_A. \tag{4.3}$

Proof. If $A=0$ the assertion is immediate. Otherwise use the modules in Lemma 4.1. Since $\mathcal K_C(G)\cong A$ is $\sigma$-unital, the countable-generation theorem in the compact-operator lesson implies that $G$ is countably generated over $C$. Theorem 3.1 gives $G\oplus H_C\cong H_C.$ Tensor this unitary with $E$ using the contractive adjointable tensor construction. Its inverse tensors to the inverse, so the resulting map is unitary. Lemma 4.1 and the finite direct-sum tensor identification give $A\oplus E^\infty\cong E^\infty. \tag{4.4}$ Here the identification $H_C\otimes_C E\cong E^\infty$ is explicit: $(c_j)\otimes x\longmapsto(c_jx)_j. \tag{4.5}$ For finite columns, its inner-product identity follows by summing $\langle c_jx,d_jy\rangle=\langle x,c_j^*d_jy\rangle$. It extends by the simple-tensor estimate and density. Each coordinate image contains $CE$, dense in $E$, so its range is dense in $E^\infty$; the completed isometry is onto.

Take countably many orthogonal copies of (4.4). Countable direct sums of unitaries are unitary: the finite-coordinate map preserves the summed inner product, extends to the completions, and has the direct sum of the inverses as inverse. A bijection $\mathbb N^2\to\mathbb N$ likewise gives a unitary $(E^\infty)^\infty\cong E^\infty$. Consequently (4.4) yields $H_A\oplus E^\infty\cong E^\infty. \tag{4.6}$ The module $E^\infty$ is countably generated: place each member of a countable generating family of $E$ in each coordinate. Those countably many vectors generate the finite-coordinate subspace densely, and that subspace is dense in the completion. Applying Theorem 3.1 again gives $E^\infty\oplus H_A\cong H_A$. Together with the summand flip and (4.6), this proves (4.3). ∎

Fullness cannot be removed. For $A=\mathbb C\oplus\mathbb C$ and $E=(1,0)A$, every inner product of $E^\infty$ lies in the first summand, whereas $H_A$ is full. A unitary preserves the coefficient ideal, so these modules cannot be isomorphic. The proof of Theorem 4.2 uses $\sigma$-unitality of $A$ precisely to make the dual module $G$ countably generated.

5. Coordinates, bundles and variants

Proposition 5.1. The standard module $H_A$ is countably generated if and only if $A$ is $\sigma$-unital.

Proof. If $A$ has a sequential positive contractive approximate identity $u_n$, place $u_n$ in coordinate $j$, for all pairs $j,n$. These vectors generate: any coordinate $a$ is approximated by $u_na$, and finite columns are dense.

Conversely, take generators $z_n$ of $H_A$. Applying the bounded first-coordinate projection shows that their first coordinates generate $A$ densely as a right module. Indeed every first-coordinate vector $a$ is a limit of first coordinates of finite generated sums. Thus $A_A$ is countably generated, and $\mathcal K(A)=A$ is $\sigma$-unital by the earlier countable-generation theorem. ∎

There is nevertheless a unitary $A\oplus H_A\longrightarrow H_A,\qquad (a,(a_1,a_2,\ldots))\longmapsto(a,a_1,a_2,\ldots) \tag{5.1}$ for every $A$. Its inner product is the same norm-convergent coordinate sum, and its inverse separates the first coordinate. Thus this particular absorption identity does not require that $A_A$ be countably generated.

Example 5.2 (Bundle sections). Let $X$ be second countable, locally compact Hausdorff, and $V\to X$ a finite-rank Hermitian vector bundle. Then $\Gamma_0(V)$ is countably generated over $C_0(X)$. To see this, cover $X$ by countably many relatively compact trivializing neighborhoods. Choose a countable collection of compactly supported continuous cutoffs in these neighborhoods, with their positive sets covering $X$; multiply each cutoff by each member of a local orthonormal frame and extend by zero. This yields countably many global sections.

Any compactly supported section belongs to their module span. Cover its support by finitely many positive cutoff sets and use a subordinate partition of unity with supports inside those sets. In each term, local frame coordinates divided by the relevant cutoff give continuous scalar coefficients, extended by zero; their supports are compact. Summing the resulting frame expansions reconstructs the section. Compactly supported sections are dense in $\Gamma_0(V)$, by scalar cutoffs, so the claimed generation follows. Stabilization therefore realizes this module as a multiplier-projection range in $H_{C_0(X)}$. If every fibre is nonzero, it is full: near each point a local section has strictly positive squared norm, and a finite partition on any compact support expresses each compactly supported function using such inner products. Density gives the whole coefficient algebra. Theorem 4.2 then applies.

Example 5.3 (An obstruction from uncountability). For an uncountable set $I$, the Hilbert space $\ell^2(I)$, viewed as a Hilbert $\mathbb C$-module, is not countably generated. Each vector has countable support, since the coordinates exceeding $1/n$ in absolute value form a finite set for every $n$. The union of the supports of countably many vectors is countable. Their closed span is contained in the corresponding coordinate Hilbert space and misses every unit vector outside that union. In particular $\ell^2(I)\oplus\ell^2(\mathbb N)\not\cong\ell^2(\mathbb N)$: the latter is separable and the former contains an uncountable orthonormal family.

A grading on $A$ is an involutive *-automorphism $\gamma_A$. A graded Hilbert module has a linear isometric involution $\gamma_E$ satisfying $\gamma_E(xa)=\gamma_E(x)\gamma_A(a),\qquad \langle\gamma_E x,\gamma_E y\rangle=\gamma_A(\langle x,y\rangle).$ The standard graded module is $\widehat H_A=H_A\oplus H_A^{\mathrm{op}}$, with the natural grading on the first summand and its negative on the second.

Proposition 5.4 (Graded stabilization). A countably generated graded $E$ satisfies $E\oplus\widehat H_A\cong\widehat H_A$ by a grading-preserving unitary.

Proof. In the unital case, repeat a bounded generating sequence $\eta_j$ infinitely often and write $\eta_j=\eta_j^{(0)}+\eta_j^{(1)}$. In both degree sets, including zero components, use every standard coordinate $e_j^{(d)}$ of $\widehat H_A$, and set $T(e_j^{(d)}a)=(2^{-j}\eta_j^{(d)}a,\,4^{-j}e_j^{(d)}a), \qquad d=0,1.$ The corresponding rank-one series converges in norm and has degree zero. The diagonal-density and repeated-generator arguments work in both coordinate sets, giving the two hypotheses of Lemma 2.2. Its $U$ has degree zero: $T^*T$ and its square root commute with the grading, and the formula defining $U$ intertwines it on a dense subspace. For nonunital $A$, the grading extends to $A^+$ with the unit even. The ideal restriction (3.5) respects the grading. ∎

For a second countable locally compact group $G$, a strongly continuous compatible action on a Hilbert $A$-module satisfies $g(xa)=(gx)\beta_g(a),\qquad \langle gx,gy\rangle=\beta_g(\langle x,y\rangle).$ Strong continuity means norm continuity of $g\mapsto gx$ for every $x$. Conjugating an adjointable map by the actions need not give a norm-continuous orbit.

The equivariant variant has a carefully qualified conclusion. For a separable graded $G$-algebra $A$ and countably generated graded $G$-module $E$, set $\widehat H_A^G=L^2(G)\boxtimes\widehat H_A$ with the left regular action on $L^2(G)$ and the diagonal action on the exterior product. There is an even unitary $U:E\oplus\widehat H_A^G\to\widehat H_A^G$ whose orbit $g\mapsto gUg^{-1}$ is norm continuous. The exact programme dependency is Equivariant KK-theory and the Green–Julg theorem, Section 1 of Kasparov’s KK-theory. That assigned section owns this G-continuous even stabilization statement, under precisely the second countability, separability, grading and strong-continuity hypotheses just given. Its proof is planned; it is not presented as written or independently checked. Blackadar, Theorem 20.1.4, is the historical source. It does not assert that this unitary intertwines the group actions for arbitrary noncompact $G$. The theorem and its equivariant refinements are used in equivariant Kasparov theory.

6. Exercises with solutions

Exercise 6.1 (Basic: the shift). Prove $A\oplus H_A\cong H_A$ without imposing $\sigma$-unitality.

Solution. Use (5.1). For two pairs, the inner product is $a^*b+\sum_j a_j^*b_j$ on either side. Finite columns give the identity first, and the convergent series gives it on the completions. The inverse separates the first entry and shifts the others back. Preservation of inner products and invertibility make the inverse the adjoint. No unit column in $A$ is needed. ∎

Exercise 6.2 (Intermediate: both density statements). For the unital repeated-generator operator $T$ in (3.2), prove density of its range and of the range of $T^*T$. Explain why injectivity is not a replacement.

Solution. Each repeated generator $\eta$ occurs at indices tending to infinity. Formula (3.4) puts $(\eta a,0)$ in the closure of the range, and generation puts $E\oplus0$ there. Subtract these first-summand vectors from $T(e_ja)$; the closure also contains all $(0,e_ja)$, hence all of $E\oplus H_A$. Separately $T^*T\geq D^2$, and $D^2$ has every finite column in its range. Lemma 2.1 proves density of $(T^*T)H_A$. These statements make the isometry in Lemma 2.2 defined on a dense domain and onto a dense range. An injective module operator can have nondense range, so injectivity alone would not establish that extension. ∎

Exercise 6.3 (Intermediate: the compact corner). Deduce the compact-algebra identification in (3.6), including its surjectivity.

Solution. Conjugate the first-summand projection by a stabilization unitary, obtaining $P$. This identifies $E$ with $PH_A$. For any rank-one operator on $H_A$, $P\theta_{x,y}P=\theta_{Px,Py}$; conversely every rank-one operator on $PH_A$ has exactly this form with $x,y\in PH_A$. Taking finite sums and norm closures gives $\mathcal K(PH_A)=P\mathcal K(H_A)P$. Substitute $\mathcal K(H_A)=\mathbb K\otimes_{\min}A$. Thus both inclusion and surjectivity are proved; $P$ is allowed to be a multiplier projection. ∎

Exercise 6.4 (Intermediate: full amplification). Prove (4.3) and identify where fullness and $\sigma$-unitality enter.

Solution. Form $G=\mathcal K(E,A)$ over $C=\mathcal K(E)$. Fullness identifies $\mathcal K_C(G)$ with $A$ and $G\otimes_C E$ with $A_A$, using (4.2) and the dense span of $\langle E,E\rangle$. Since $A$ is $\sigma$-unital, $G$ is countably generated. Stabilize it and tensor with $E$; (4.5) gives $A\oplus E^\infty\cong E^\infty$. Infinite amplification and coordinate reindexing yield $H_A\oplus E^\infty\cong E^\infty$. The module $E^\infty$ is countably generated, so ordinary stabilization also gives $E^\infty\oplus H_A\cong H_A$. Combining the two unitaries proves $E^\infty\cong H_A$. Without fullness, the dual tensor range is the coefficient ideal instead of all of $A$; without $\sigma$-unitality the required countable generation of $G$ is not supplied. ∎

Exercise 6.5 (Advanced: when the standard module is countably generated). Prove Proposition 5.1, and explain why it does not restrict the generality of Theorem 3.1.

Solution. A sequential approximate identity $u_n$ gives the countable generating family consisting of $u_n$ in coordinate $j$, for all $j,n$, because $u_na\to a$. Conversely, first coordinates of any countable generating family of $H_A$ generate $A_A$. The compact algebra of this module is $A$, so the earlier countable-generation theorem makes $A$ $\sigma$-unital. Ordinary stabilization only assumes that the added module $E$ is countably generated; it never assumes that the absorbing module $H_A$ has this property. The nonunital proof constructs its unitary by restriction from the unitization even when $H_A$ is not countably generated. ∎

What this lesson does not prove

We use the earlier course results on compact rank-one norms and composition, nondegenerate coefficient and compact actions, countable generation versus $\sigma$-unitality, multiplier identification, and interior tensor products. Their relevant proofs are Theorem 3.1 and Lemma 1.3 of Compact operators, multipliers and the strict topology, and Theorems 1.2, 2.1 and Proposition 3.2 of Tensor products and C*-correspondences. Continuous functional calculus and C*-operator order are foundational prerequisites.

The bundle example uses locally compact cutoffs, local orthonormal frames and subordinate finite partitions on compact subsets; these are proved in the opening lemma of Section 5 of Finite projective modules, frames and K₀. Emerson, Section 5.2, gives the classical bundle treatment. It assumes second countability explicitly. We do not infer countable generation for arbitrary bundles on arbitrary locally compact spaces.

The $G$-continuous equivariant stabilization application has the exact planned programme provider Equivariant KK-theory and the Green–Julg theorem, Section 1; its required hypotheses and conclusion are stated above. This dependency does not enter the ordinary or graded stabilization proofs here. Graded stabilization is proved above. We do not construct foliated continuous fields or their geometric realization: [Connes 1982, the section on C*-modules and continuous fields on the leaf space] supplies that application context. The full-amplification proof here is algebraic and does not depend on a stable-isomorphism theorem.

References

[Blackadar 1998] Bruce Blackadar, K-Theory for Operator Algebras, second edition, Cambridge University Press, 1998, Proposition 13.6.1, Theorem 13.6.2, Corollary 13.6.3, Exercise 13.7.1(a), and Theorems 14.6.1 and 20.1.4. Author's corrected second edition.

[Blackadar 2006] Bruce Blackadar, Operator Algebras: Theory of C*-Algebras and von Neumann Algebras, Springer, 2006, II.7.1.11, II.7.2.6 and II.7.6.12. Author's revised edition.

[Connes 1982] Alain Connes, “A survey of foliations and operator algebras,” Operator Algebras and Applications, Part I, Proceedings of Symposia in Pure Mathematics 38, American Mathematical Society, 1982, 521–628, section “C* modules over C*(V,F) and continuous fields of Hilbert spaces on V/F.” Author's text.

[Emerson 2024] Heath Emerson, An Introduction to C*-Algebras and Noncommutative Geometry, Birkhäuser, 2024, Section 5.2 for local bundle tools.