HIGHER K-GROUPS FOR OPERATOR SYSTEMS

WALTER D. VAN SUIJLEKOM

arXiv:2411.02981v1 [math.OA] 5 Nov 2024

A BSTRACT. We extend our previous definition of K-theoretic invariants for op-

erator systems based on hermitian forms to higher K-theoretical invariants. We
realize the need for a positive parameter δ as a measure for the spectral gap of the
representatives for the K-theory classes. For each δ and integer p ≥ 0 this gives
operator system invariants V pδ (−, n), indexed by the corresponding matrix size.
The corresponding direct system of these invariants has a direct limit that pos-
sesses a semigroup structure, and we define the Kδp -groups as the corresponding
Grothendieck groups. This is an invariant of unital operator systems, and, more
generally, an invariant up to Morita equivalence of operator systems. Moreover,
there is a formal periodicity that reduces all these groups to either K0δ or K1δ . We
illustrate our invariants by means of the spectral localizer.

C ONTENTS
1. Introduction 1
Acknowledgements 2
2. Background on operator systems 2
3. δ-gapped K-theory for unital operator systems: K0δ and K1δ 2
3.1. The K0δ -groups 4
3.2. K1δ for unital operator systems 5
4. Higher δ-gapped K-groups and formal periodicity 7
5. Application to the spectral localizer 9
5.1. Example: spectral localizer on the circle 11
Appendix A. Clifford algebras 12
References 13

1. I NTRODUCTION
We continue our quest for a definition of K-theory for operator systems. In a
previous work, we have defined a K0 -group which we proved to be an invariant of
operator systems up to complete order isomorphism. Moreover, we showed that it
is stably equivalent, which by [9] amounts to invariance under Morita equivalence
of unital operator systems.
In this work, we extend this and define higher K-groups. At the same time, with
the applications to the spectral localizer of [12] in mind, we realize the need for a
parameter δ that indicates (and quantifies the size of) a spectral gap in the elements
that represent K-theory classes. This also solves the problem one is facing when

Date: November 5, 2024.

1
2 WALTER D. VAN SUIJLEKOM

considering the space of all invertible elements in a finite-dimensional operator

system, namely, that this space is contractible.
The definition of K0δ extends the group K0 considered in [21] which is obtained
as a special case by setting δ = 0. The groups K1δ ( E) are defined, and shown to
be stably equivalent as well. Moreover, we may introduce higher K-groups Kδp ( E)
by exploiting graded Clifford algebras, for which we show a formal periodicity to
hold, thus reducing them to K0δ and K1δ for p even and odd, respectively.
As an illustrative example, we obtain the spectral localizer as an index pairing
IndδD : Kδp ( E) → Z in the even ( p = 0) and odd ( p = 1) case, respectively.

Acknowledgements. I would like to thank Alain Connes and Kristin Courtney for
useful discussions. I am grateful to Yuezhao Li for his suggestions and comments.

2. B ACKGROUND ON OPERATOR SYSTEMS

We start by briefly recalling the theory of operator systems, referring to [8, 17,
18, 3] for more details.
A unital operator system ( E, e) is a matrix-ordered ∗-vector space E, equipped
with an Archimedean order unit e. A map φ : E → F between operator sys-
tems determines a family of maps φ(n) : Mn ( E) → Mn ( F ) given by φ(n) ([ xi j ]) =
[φ( xi j )]. A map φ : E → F between unital operator systems is called completely
positive if each φ(n) is positive (n ≥ 1). We also abbreviate completely positive by
cp, and unital completely positive by ucp.
A dilation of a ucp map φ : E → B(H) of a unital operator system is a ucp map
ψ : E → B(K), where K is a Hilbert space containing H such that PH ψ( x)|H =
φ( x) for all x ∈ E. The ucp map φ is called maximal if every dilation of φ is
obtained by attaching a direct summand.
A non-zero cp map φ : E → B(H) is said to be pure if the only cp maps satisfy-
ing 0 ≤ ψ ≤ φ are scalar multiples of φ.
We may view E as a concrete operator system in the C ∗ -algebra C ∗ ( E) it gen-
erates; in this case, we say that a ucp map φ : E → B(H) has the unique extension
property if it has a unique ucp extension to C ∗ ( E) which is a ∗-representation. If,
in addition, the ∗-representation is irreducible, it is called a boundary representation
[2]. The following result is well-known in the literature [2, 15, 1, 11, 6] (see also
[21]).
Proposition 1. Let φ : E → B(H) be a ucp map. Then
(1) φ is maximal if and only if it has the unique extension property.
(2) φ is pure and maximal if and only if it is a boundary representation.

3. δ- GAPPED K- THEORY FOR UNITAL OPERATOR SYSTEMS : K0δ AND K1δ

Even though in an operator system we cannot speak about invertible elements,
we may use the pure and maximal ucp maps to quantify the extent to which an
element is (non)singular.
Definition 2. Let ( E, e) be a unital operator system and let δ > 0. An element x ∈
Mn ( E) is called δ-singular if for all s ∈ (0, δ ) there exists a g > 0 such that for all pure
 HIGHER K-GROUPS FOR OPERATOR SYSTEMS 3

and maximal ucp maps φ : E → B(H) we have

  
0 x
(1) φ(2n) s · e2n + ≥ g · id⊕ 2n
x∗ 0 H

The largest real number g > 0 such that (1) holds is called the s-gap of x.

We will write Gδ ( E, n) for all δ-singular elements in Mn ( E), and Hδ ( E, n) for the
δ-singular self-adjoint elements in Mn ( E). We extend this definition of δ-singular
to the limiting case δ = 0 to be those elements for which (1) holds with s = 0; in the
self-adjoint case these are precisely the non-singular hermitian forms considered
previously in [21].
Proposition 3. An element x ∈ Mn ( E) is δ-singular if and only if
!
(n)
s ı E ( x)
(n)
ı E ( x∗ ) s
∗ ( E )) for all s ∈ (0, δ ), where ı : E → C ∗ ( E ) is
are invertible elements in M2n (Cenv E env
∗
the C -envelope.
Proof. In [6] the C ∗ -envelope of E is constructed as the direct sum of all boundary
representations (Hσ , σ ):
M
ıE : E → B(Hσ ).
σ

Now if x ∈ Mn ( E) then a matrix such as the above is invertible if and only if

(n)
its image ı E ( x) is bounded from below by g · id⊕σ (Hσ ) . But this holds if and
only if (1) holds for all boundary representations σ. Since by Proposition 1 a ucp
map is a boundary representation if and only if it is pure and maximal, the result
follows. □
(n) ∗ ( E ).
We can now translate δ-singularity into spectral properties of ı E ( x) in Cenv
For this consider the following spectrum
! 
(n)  (n)
 
0 ı E ( x) −λ ı E ( x)
Σ x : = σ (n) ≡ λ ∈ C | (n) ∗ is not invertible
ı E ( x∗ ) 0 ı E ( x ) −λ

This subset of R is symmetric around 0 since we may unitarily conjugate

! !
  (n)  (n)
1 0 −λ ı E ( x) 1 0 λ ı E ( x)
(n) = − (n)
0 −1 ı E ( x∗ ) −λ 0 −1 ı E ( x∗ ) λ
The notion of δ-singularity can now be captured by the property that ±s ∈
/ Σ x for
all s ∈ (0, δ ), or, equivalently:
Corollary 4. Let δ > 0. An element x ∈ Mn ( E) is δ-singular iff Σ x ⊆ (−∞, −δ ] ∪
{0} ∪ [δ, ∞).
Our notion of δ-singularity thus allows to detect the presence of a so-called
spectral gap of size δ. This terminology is borrowed from physics where the spectral
gap is the energy difference between the ground state and the first excited state.
4 WALTER D. VAN SUIJLEKOM

For this reason we also refer to such elements as being δ-gapped. Moreover, we can
find an explicit lower bound for the s-gap, since
!
(n)
s ı E ( x)
(2) (n) ≥ min{λ : λ ∈ s + Σ x } = min{s, δ − s}
ı E ( x∗ ) s
In the self-adjoint case, we have the following result.
(n)
Corollary 5. Let δ > 0. A self-adjoint element x ∈ Mn ( E) is δ-singular iff ı E ( x) has
(n)
spectral gap δ, i.e. σ (ı E ( x)) ⊆ (−∞, −δ ] ∪ {0} ∪ [δ, ∞).
Proof. This follows by a similarity transformation:
! !
(n) (n)
s ı E ( x) s + ı E ( x) 0
(n) ∼ (n) .
ı E ( x) s 0 s − ı E ( x)
(n)
Indeed, the matrix on the right-hand side is invertible iff ±s ∈
/ σ (ı E ( x)) for all
s ∈ (0, δ ). □
3.1. The K0δ -groups. The K0 -group of non-singular self-adjoint elements has been
treated in detail in [21]; let us consider here the changes when δ-singular, or δ-
gapped self-adjoint elements are considered.
Definition 6. Let δ ≥ 0 and x, x′ ∈ Hδ ( E, n). We say that x ∼n x′ if there exists
xe ∈ Hδ (C ([0, 1]) ⊗ E), n) such that
xe(0) = x; xe(1) = x′
We denote the equivalence class of x ∈ Hδ ( E, n) by [ x]n , and the set of all such equivalence
classes in Hδ ( E, n) by V0δ ( E, n), or, equivalently,
V0δ ( E, n) = Hδ ( E, n)/∼n .
All these sets V0δ ( E, n) for n ≥ 1 are invariants of unital operator systems:
Proposition 7. If E and F are completely order isomorphic then V0δ ( E, n) ∼
= V0δ ( F, n),
for any n ≥ 1.
∗ ( E ) ∼ C ∗ ( F ) via a
Proof. This follows directly from the fact that in this case Cenv = env
unital ∗-isomorphism. □
If we combine this structure at each level n with the direct sum of δ-gapped
self-adjoint elements, we obtain a semigroup structure as follows. Consider the
direct system of sets (V0δ ( E, n), ınm ) where for m ≥ n
(3) ınm : V0δ ( E, n) → V0δ ( E, m)
[ x ]n 7 → [ x ⊕ em−n ]m ,
We denote the direct limit of the direct system (3) by lim V0δ ( E, n). A more explicit
−→
description is given as follows: for x ∈ Hδ ( E, n) and x′ ∈ Hδ ( E, n′ ) we write
x ∼ x′ if there exists a k ≥ n, n′ such that x ⊕ ek−n ∼k x′ ⊕ ek−n′ in Hδ ( E, k). We will
write [ x] E , or simply [ x], for the equivalence class corresponding to x ∈ Hδ ( E, n)
and V0δ ( E) := ⨿n V0δ ( E, n)/∼ for the corresponding set of equivalence classes. The
following is then clear from the definition of the direct limit.
 HIGHER K-GROUPS FOR OPERATOR SYSTEMS 5

Proposition 8. The set V0δ ( E) is the direct limit lim V0δ ( E, n) of the direct system (3).
−→
Moreover, it is a semigroup when equipped with the direct sum [ x] + [ x′ ] = [ x ⊕ x′ ] and
identity element 0 = [e].

The K0δ -group K0δ ( E) for a unital operator system E is now defined to be the
Grothendieck group of V0δ ( E); we will refer to it as the δ-gapped K0 -group of E. In
[21] we have shown that for δ = 0, that is, for non-singular self-adjoint elements,
this group coincides with the C ∗ -algebraic K0 -group in the case of a C ∗ -algebra,
that it behaves well with respect to direct sums, and that it is invariant under
Morita equivalence of unital operator systems. For later use, we record the fol-
lowing result also in the case of general δ ≥ 0. The proof follows mutatis mutandis
from that of [21, Theorem 18]
Proposition 9 (Stability of K0δ ). Let E be a unital operator system, δ > 0, and let N be a
natural number. Then V0δ ( E) is isomorphic to V0δ ( MN ( E)) (and so are the corresponding
K0δ -groups).

3.2. K1δ for unital operator systems. For the analogue of K1 for operator systems
we drop the assumption of being self-adjoint, and consider δ-singularity in the
sense of Definition 2, or, in view of Corollary 4, the presence of a spectral gap of
size δ. Let us start with some motivating examples.
Example 10. (i) (Almost unitaries and quantitative K-theory) In [16] the quantitative
K1 -group of a filtered C ∗ -algebra A = ( Ar ) was defined in terms of ε-r-unitaries,
i.e. elements u ∈ Mn ( Ar ) such that ∥u∗ u − 1∥, ∥uu∗ − 1∥ < ε. This implies that
 2
0 u
> (1 − ε).
u∗ 0
In other words, the spectrum of the matrix on the left-hand side has spectrum con-
tained in (−∞, −δ ] ∪ [δ, ∞) for all δ < (1 − ε)1/2 . We conclude that u ∈
Gδ ( Ar , n) for all those δ (including δ = 0).
(ii) (Unitaries in operator spaces and systems) In [4] unitaries in operator spaces and
operator systems where characterized. These turn out to be unitaries in matrices with
entries in the C ∗ -envelope, so that we find that they define elements in Gδ ( E, n) as
long as 0 ≤ δ < 1.
(iii) (The odd spectral localizer) The odd spectral localizer [12] is defined in terms of a
spectral compression x = PTP of an invertible element T in a C ∗ -algebra A ⊆ B(H)
by a projection in H. We then have
2  ∗ 2 2
[ P, T ][ P, T ∗ ]
 
0 x TT
(4) =P P+P P
x∗ 0 0 T∗ T 0 [ P, T ∗ ][ P, T ]
Since T is invertible, it is bounded below by some g > 0; the same applies to T ∗ .
Consequently, we find from (4)
 2
0 x
≥ ( g2 − η2 ) P
x∗ 0

in terms of η = ∥[ P, T ]∥. So, provided δ < ( g2 − η2 )1/2 and η < g we find that
x ∈ Gδ ( PAP, 1).
6 WALTER D. VAN SUIJLEKOM

(iv) (Toeplitz matrices) Consider the unitary u given by the bilateral shift operator ℓ2 (Z).
If we compress this with a projection onto spanC {e1 , . . . en } in terms of the canonical
orthonormal basis of ℓ2 (Z), we obtain the following Toeplitz matrix:
 
0 1 ··· ··· 0
0 0 1 · · · 0
PuP =  ... ... . . . . . . 0
 
 
 
0 0 · · · 0 1
0 0 ··· 0 0
This is a candidate for index pairings in the spectral localizer (for the circle, [12, 10]),
and as such we expect it to represent an element in the Toeplitz operator system. Note
that even though u is invertible, PuP is not invertible (in fact, for this case we have
g = η = 1 in the previous example). However, the eigenvalues of
 
s PuP
Pu∗ P s
are shifted from those of PuP to become equal to −1 + s, s, 1 + s. Hence, PuP is
δ-singular provided 0 < δ < 1.
Definition 11. Let δ ≥ 0 and x, x′ ∈ Gδ ( E, n). We say that x ∼n x′ if there exists an
element xe ∈ Gδ (C ([0, 1]) ⊗ E), n) such that
xe(0) = x; xe(1) = x′
We will write V1δ ( E, n) = Gδ ( E, n)/∼n for the set of equivalence classes of elements in
Gδ ( E, n).
Remark 12. Let us come back to the parameter δ in the definition of δ-singularity (Def-
inition 2). In fact, we could have set δ = 0 from the start (as we did with hermitian
forms in [21]) and have simply considered all invertible elements in ı E ( E) ⊆ Cenv∗ ( E ) up
to homotopy equivalence. However, in most cases of interest —including the applications
to the spectral localizer (cf. Example 10.iv and Section 5 below)— such an invariant be-
comes trivial. Indeed, if E has a finite-dimensional C ∗ -envelope Md (C) then we claim that
ı E ( E) ∩ Md (C)× is contractible to the point ı E (e) = Id (and the same applies to Mn ( E)).
Namely, for any x ∈ ı E ( E) ∩ Md (C)× choose z ∈ S1 so that no eigenvalue of x lies on
the ray through the origin and z (since x has finitely many eigenvalues, this is always
possible). For any t ∈ [0, 1] we define γ (t) = tx + z(1 − t)Id . Then det γ (t) = 0 iff
z(1 − 1/t) is an eigenvalue of x. By our choice of z we find that det γ (t) ̸= 0 for all
t ∈ [0, 1]. Moreover, since complex linear combinations of x and Id are contained in ı E ( E)
it follows that the path γ (t) is contained in ı E ( E) ∩ Md (C)× . More generally, we may
(n)
conclude that ı E ( Mn ( E)) ∩ Mnd (C)× is contractible to a point for any n ≥ 1.
Again, we have the following invariance properties of the sets V1δ ( E, n):
Proposition 13. If E and F are unital operator systems. If φ : E → F is a ucp map for
which there exists a ∗-homomorphism φ
e that makes the following diagram commute,
φ
(5) E /F

ıE ıF
 φ

∗ ( E) / Cenv
∗ ( F)
e
Cenv
 HIGHER K-GROUPS FOR OPERATOR SYSTEMS 7

then there is an induced map φ∗ : V1δ ( E, n) → V1δ ( F, n) defined by

φ∗ ([ x]n ) = [φ(n) ( x)]n .
In particular, if E and F are completely order isomorphic then V1δ ( E, n) ∼
= V1δ ( F, n), for
any n ≥ 1.
Lemma 14. Let E and F be unital operator system and let δ ≥ 0. Then for any n we have
V1δ ( E ⊕ F, n) ∼
= V1δ ( E, n) × V1δ ( F, n).
Again, as with the group K0 discussed in the previous section, we may intro-
duce a semigroup structure on the direct limit V1δ ( E) := lim V1δ ( E, n) of a direct
−→
system (V1δ ( E, n), ınm ) with connecting morphisms defined as in 3. It is a semi-
group when equipped with the direct sum [ x] + [ x′ ] = [ x ⊕ x′ ] and identity ele-
ment 0 = [e].
Definition 15. Let ( E, e) be a unital operator system and let δ ≥ 0. We define the
K-theory group K1δ ( E) of E to be the Grothendieck group of V1δ ( E).
We will refer to this group as the δ-gapped K1 -group of E. Similar to Proposition
9 above we may show:
Proposition 16 (Stability of K1δ ). Let E be a unital operator system, δ ≥ 0, and let N be a
natural number. Then V1δ ( E) is isomorphic to V1δ ( MN ( E)) (and so are the corresponding
K1δ -groups).

4. H IGHER δ- GAPPED K- GROUPS AND FORMAL PERIODICITY

As in K-theory for C ∗ -algebras there are also higher-order invariants, which we
now introduce. We first need the following.
Definition 17. An operator system E is called Z2 -graded if there is a direct sum de-
composition of ∗-vector spaces E = E(0) ⊕ E(1) such that the corresponding grading
x 7→ (−1)| x| x is a complete order isomorphism.
Consequently, there is a ∗-isomorphism on the C ∗ -envelope that extends the
grading on E, turning Cenv∗ ( E ) into a Z -graded C ∗ -algebra.
2
Much in the same way as Van Daele defined K-theory for graded Banach alge-
bras [19, 20], we will consider odd δ-singular elements in Mn ( E) as our candidates
for K-theory representatives. More precisely, we set
Definition 18. Let ( E, e) be a Z2 -graded unital operator system. We define
b δ ( E, n) := Hδ ( E, n) ∩ Mn ( E(1) );
H (d = 0, 1).
b δ ( E, n) = H
and also set V b δ ( E, n)/∼ .
n

It would be interesting to develop this theory further for graded operator sys-
tems, similar to [19, 20], however, we will focus here on the ungraded case. For
this, we will exploit the graded Clifford algebras (cf. Appendix A) to connect to
our previous definitions of V0δ and V1δ . Let E be an ungraded unital operator sys-
tem, and consider the tensor product E ⊗ Cl p . Given the structure of the Clifford
algebras as matrix algebras, the operator system structure of this tensor product is
unambiguous, and so is the grading on it.
8 WALTER D. VAN SUIJLEKOM

Definition 19. Let ( E, e) be an (ungraded) unital operator system and let δ ≥ 0. We

define the higher invariants V pδ ( E, n) for p = 0, 1, . . . , by

V pδ ( E, n) := V
b δ ( E ⊗ Cl p , n )
p

The higher δ-gapped K-theory groups Kδp ( E) of a unital operator system E are defined
as the Grothendieck groups of the corresponding direct limits V pδ ( E) := lim V pδ ( E, n),
−→
equipped with direct sum as the additive operation.
Let us now check that this definition is consistent with that of the invariants
V0δ ( E, n) and V1δ ( E, n) (introduced in Definitions 6 and 11).
For the first case, p = 0, we can indeed identify using Equation (10):
  
(1) x 0
H δ ( E ⊗ Cl 1 , n ) ∼
= ∈ Hδ ( E, 2n) ∼= Hδ ( E, n).
0 −x
The last isomorphism follows from the fact that there is a similarity transformation
between the following matrices:
 
s 0 x 0  
0
 s 0 − x  ∼ s x ⊗ I2
x 0 s 0  x s
0 −x 0 s
Upon taking the homotopy equivalence classes, we find agreement with Definition
6. Note that such a doubling has also been considered in the definition of K-theory
of ungraded Banach algebras in [19, Remark 2.13(iv)].

For p = 1 we have, using Equation (11):

  
(1) 0 x
H δ ( E ⊗ Cl 2 , n ) ∼
= ∈ H δ
( E, 2n ) ∼
= Gδ ( E, n),
x∗ 0
again via a similarity transformation, this time:
 
s 0 0 x
 0 s x∗ 0 
 
  ∼ s∗ x
⊗ I2
 0 x s 0 x s
x∗ 0 0 s
We thus find agreement of the above definition of V1δ ( E, n) with Definition 11.
More generally, we have the following, which may be considered as an operator
system analogue of formal periodicity:
Proposition 20. Let ( E, e) be a unital operator system and let δ > 0. Then
δ
V2m ( E, n) ∼
= V0δ ( M2m ( E), n), δ
V2m ∼ δ
+1 ( E, n ) = V1 ( M2m ( E ), n )
for any m ≥ 0.
Proof. For the even-dimensional case, we use (10) to obtain
δ (1)
V2m ( E, n) = Hδ ( E ⊗ Cl2m+1 , n)/∼n
n o
∼ x 0 δ


= 0 −x ∈ H ( E ⊗ ( M 2 m (C) ⊕ M2m (C)) , n ) / ∼
n
n o
∼ δ δ
= x ∈ H ( M2m ( E), n) /∼n ≡ V0 ( M2m ( E), n).
 HIGHER K-GROUPS FOR OPERATOR SYSTEMS 9

For the odd-dimensional case, we instead use (11) to write

δ δ (1)
V2m +1 ( E, n ) = H ( E ⊗ Cl2m+2 , n )/∼n
n o
∼ 0 x δ


= x∗ 0 ∈ H ( E ⊗ M2m+1 (C), n ) /∼n
n o
∼
= x ∈ Gδ ( M2m ( E), n) /∼n ≡ V1δ ( M2m ( E), n). □

Proposition 20, in combination with the stability of K0δ and K1δ (Proposition 9
and 16), then yields the following periodicity result:
Theorem 21. Let ( E, e) be an ungraded unital operator system and let δ ≥ 0. Then
δ ( E ) ∼ Kδ ( E ) and Kδ ∼ δ
K2m = 0 2m+1 ( E ) = K1 ( E ).

5. A PPLICATION TO THE SPECTRAL LOCALIZER

In [13, 12] the spectral localizer was introduced as a powerful tool for index
pairings. They reduce the computation of a certain Fredholm index to the compu-
tation of the signature of a finite-dimensional matrix —the spectral localizer. We
will put it in the context of our notion of K-theory, realizing the spectral localizer
as a spectral flow formula (very much as in [13, 14]) which maps V pδ ( E, n) → Z.
For more details on spectral flow, including the spectral localizer, we refer to the
excellent textbook [7] and references therein.
In order to describe the index map from V pδ ( E, n), we need an operator system
spectral triple [5].
Definition 22. A (unital) operator system spectral triple is given by a triple ( E, H , D )
∗ ( E ) ⊆ B (H),
where E is a unital operator system realized concretely so that E ⊆ Cenv
and a self-adjoint operator D : Dom( D ) → H such that
• the commutators [ D, x] extend to bounded operators for all x ∈ E for a dense
∗-subspace E ⊆ E;
• the resolvent (i + D )−1 is a compact operator.
An operator system spectral triple is called even if in addition to the above, there is a
grading operator γ (so that γ ∗ = γ, γ 2 = 1H ) which commutes with all x ∈ E and
anti-commutes with D. Otherwise, it is called odd.
In the even case, we can decompose H = H+ ⊕ H− according to the eigenval-
ues of γ, and decompose accordingly
 
0 D0
D= .
D0∗ 0
In the odd case, there is no such grading, but for convenience we will then write
D0 ≡ D.
Given an operator system spectral triple ( E, H , D ), a parameter κ > 0 and an
element x ∈ Gδ (E , n) we now define the generalized spectral localizer for any s ∈
(0, δ ) as the following (unbounded) self-adjoint operator on H⊕4n :
κD0⊕n
 
s x 0
 x∗ s 0 κD0⊕n 
(6) Lκ ( D, x, s) = 
κ ( D ⊕n )∗
.
0 0 −s −x 
0 κ ( D0⊕n )∗ − x∗ −s
10 WALTER D. VAN SUIJLEKOM

We use the term ’generalized’ to distinguish it from the even and odd spectral
localizer consider in [7]. However, they are not unrelated:
Remark 23. Note that if we set s = 0 in the odd case (so that D0 = D) we obtain (twice)
the odd spectral localizer Lκodd introduced in [12]:
 ⊕n 
odd ⊕2 odd κD x
Lκ ( D, x, s = 0) ∼ ( Lκ ) ; Lκ = ,
x∗ −κD⊕n
that is, up to a similarity transformation.
In the even case, we will instead consider pairings with self-adjoint elements x ∈
Hδ (E , n) in which case we find for s = 0 the even spectral localizer Lκeven of [13]:
κ ( D0 ) ⊕ n
 
even ⊕2 even x
Lκ ( D, x, s = 0) ∼ ( Lκ ) ; Lκ =
κ ( D0∗ )⊕n −x
In general, the shift s ∈ (0, δ ) is crucial in order to define the spectral localizer
at the level of the operator systems, even though in some cases it may be set to
zero (cf. Section 5.1 below).
Proposition 24. Let ( E, H , D ) be a finite-dimensional operator system spectral triple.
Let x ∈ Gδ (E , n) and let s ∈ (0, δ ) and write gs = min{s, δ − s}. Then the signature of
the generalized spectral localizer,
Sig(Lκ ( D, x, s)),
is constant in the (κ, s)-region defined by 0 < κ < g2s ∥[ D, x]∥−1 and 0 < s < δ, and
invariant under homotopy equivalence. Consequently, we have an induced map IndδD :
V1δ (E , n) → Z given by
1
IndδD ([ x]) = lim Sig(Lκ ( D, x, s)),
κ,s→0 4
where the limit is taken in the above region.
In the even case, we take x ∈ Hδ (E , n) whence the induced map is IndδD : V0δ (E , n) →
Z.
Proof. Without loss of generality we take n = 1 and compute very similar to [13]
that
   2   
κ 2 D0 D0∗ 0 s x 0 −κ [ D0 , x]
+
0 κ 2 D0 D0∗ x∗ s −κ [ D0 , x∗ ] 0
Lκ ( D, x, s)2 = 
 
 ∗
    2 
0 κ [ D0 , x ] κ 2 D0∗ D0 0 s x 

+
κ [ D0 , x ] 0 0 2 ∗
κ D0 D0 x∗ s
 
(7) ≥ g2s − κ ∥[ D, x]∥ 1 M4 ( B(H)) .

Hence Lκ ( D, x, s) is invertible (and thus has well-defined signature) provided

0 < κ < g2s ∥[ D, x]∥−1 and 0 < s < δ. Moreover, the signature is constant in
this region since no eigenvalues will cross the origin. This also makes the limit
of Sig(Lκ ( D, x, s)) when (κ, s) approach the origin from within this region well-
defined.
Also, one may easily check that
Sig(Lκ ( D, x ⊕ x′ , s)) = Sig(Lκ ( D, x, s)) + Sig(Lκ ( D, x′ , s)).
 HIGHER K-GROUPS FOR OPERATOR SYSTEMS 11

In order to see that Sig(Lκ ( D, e, s)) = 0 consider (in the odd case) an orthonormal
eigensystem {vλ }λ of D in H so that Dvλ = λvλ . We may write the matrix of
Lκ ( D, x, s) in this eigensystem as
 
s 1 κλ 0
1 s 0 κλ 
(8) ⟨vλ , Lκ ( D, x, s)vλ′ ⟩ =δλλ′ 
κλ 0 −s −1 ,


0 κλ −1 −s
p
which has set of eigenvalues {± (1 ±′ s)2 + κ 2 λ 2 } so that the signature of the
spectral localizer vanishes.
−
In the even case, we consider instead an eigensystem {vλ = (v+ λ , vλ )}λ in H+ ⊕
− + ± ±
H− such that D0 vλ = λvλ and γvλ = ±vλ . Again, we find that the matrix of
Lκ ( D, x, s) in this eigensystem is given by (8) so that the signature of the spectral
localizer vanishes.
Consider now a homotopy xe in Gδ ( E, n) (or in Hδ ( E, n) in the even case) be-
tween x = xe(0) and x′ = xe(1) with s-gap ≥ gs . By a computation similar to the
one in Eq. (7) we find that
Lκ ( D, xe, s)2 ≥ g2s − κ sup ∥[ D, xe(t)]∥
t

So as long as κ < g2s / supt

∥[ D, xe(t)]∥ we find that Sig Lκ ( D, xe(t)) is constant in t.
We combine this with the fact that κ < g2s /∥[ D, xe(0)]∥ and κ < g2s /∥[ D, xe(1)]∥ to
conclude that limκ,s→0 Sig Lκ ( D, xe(0), s) = limκ,s→0 Sig Lκ ( D, xe(1), s). This com-
pletes the proof. □
We may now rephrase the main results of [13, 12]. For an invertible (self-adjoint)
element a ∈ A one considers the class [ a] in the K-theory group of the C ∗ -algebra
A. Given a spectral triple ( A, H , D ), there is an index pairing Ind D ([ a]). As shown
in loc.cit. this index may be computed in terms of a spectrally truncated (operator
system) spectral triple ( PAP, PH , PDP) for a spectral projection P (of D) of suffi-
ciently high rank. Indeed, we then have
1 1
(9) Ind D ([ a]) = Sig(Lκ ( PDP, x, 0)) = Sig( Lκeven/odd )
4 2
since in fact it is shown in [12] that Lκ ( D, x, 0) already has a positive gap, say
Lκ ( D, x, 0)2 ≥ g2 . Thus, adding the matrix with diagonal entries (s, s, −s, −s) to
obtain Lκ ( D, x, s) will not change this property, as long as s < g.
5.1. Example: spectral localizer on the circle. Let us illustrate the spectral local-
izer by spectral truncations of the circle. We consider a unitary u(t) = eimt with
winding number m and want to use the spectral localizer to compute the index
pairing with DS1 = −id/dt.
We take a spectral projection PN onto spanC {e− N , e− N +1 , . . . , e N } and realize
that PN uPN is a Toeplitz matrix with 1’s on the m′ th diagonal, and zeros elsewhere.
Note that PN uPN is not invertible, however the following self-adjoint 2 × 2 matrix
with values in the truncated operator system is non-singular for any s ∈ (0, δ ) for
some δ > 0:  
sPN PN uPN
PN u∗ PN sPN
Hence PN uPN ∈ Gδ ( PN C (S1 ) PN , 1).
12 WALTER D. VAN SUIJLEKOM

F IGURE 1. The negative and positive eigenvalues for the spectral

localizer Lκ,N = Lκ,N,s=0 on the circle for m = 1, κ = 1, N = 3
(left) and m = 2, κ = 0.1, N = 3 (right). The red (diamond-
shaped) dots indicate the surplus of positive eigenvalues as com-
pared to the negative ones.

The spectral localizer is given in this case by the following matrix:

s PN uPN κPN DPN 0 !
PN u∗ PN s 0 κPN DPN
Lκ,N,s := Lκ ( PN DPN , PN uPN , s) = κPN DPN 0 −s − PN uPN .
0 κPN DPN − PN u∗ PN −s

The main result of [12] (cf. Eq. (9) and the discussion following it) can now be used
to show that for suitable N, κ there exists g > 0 so that the generalized spectral
2
localizer satisfies Lκ,N,s ≥ g2 for any s < δ, including s = 0. As such, the signature
can be computed by setting s = 0, so that
1 1
Sig Lκ,N,s=0 = Sig Lκ,N ,
Ind PN DPN ([ PN uPN ]) =
4 2
where as in Remark 23 we have written Lk,N,0 = Lκ,N ⊕ Lκ,N in terms of
 
κPN DPN PN uPN
Lκ,N =
PN u∗ PN − PN DPN
The main result of [12] states that Ind( PD uPD ) = 21 Sig Lκ,N so that we deduce that
Ind( PD uPD ) = Ind PN DPN ([ PN uPN ])
In Figure 1 we illustrate the resulting signature by showing the negative and
positive eigenvalues of Lκ,N . We find already for low N that the spectral localizer
is equal to (twice) the winding number of u.

A PPENDIX A. C LIFFORD ALGEBRAS

Recall the definition of the Clifford algebras Cl p for p ≥ 1 associated to the
vector spaces C p with their standard quadratic form. These are Z2 -graded C ∗ -
(0) (1)
algebras, i.e. Cl p = Cl p ⊕ Cl p . For each p we have a grading operator Γ such
(d)
that Cl p = { a ∈ Cl p : Γ a = (−1)d aΓ } (if p is odd, this grading operator is an
element in the multiplier algebra of Cl p ).
 HIGHER K-GROUPS FOR OPERATOR SYSTEMS 13

In the odd case p = 2m + 1 we have Cl2m+1 ∼

= M2m (C) ⊕ M2m (C) and we may
take  
0 I2 m
Γ= ∈ M2m+1 (C).
I2 m 0
Consequently,
(0)
Cl2m+1 ∼
= {( x, x) : x ∈ M2m (C)} ⊂ M2m (C) ⊕ M2m (C).
(1)
(10) Cl2m+1 ∼
= {( x, − x) : x ∈ M2m (C)} ⊂ M2m (C) ⊕ M2m (C).
Instead, if p = 2m then Cl2m ∼ = M2m (C) and we may take
 
I m−1 0
Γ= 2 .
0 −I2 m − 1
Accordingly,
  
(0) ∼ x 0
Cl2m = : x, y ∈ M2m−1 (C) .
0 y
  
(1) 0 x
(11) Cl2m ∼= : x, y ∈ M2m−1 (C) .
y 0

R EFERENCES
[1] W. Arveson. The noncommutative Choquet boundary. J. Amer. Math. Soc. 21 (2008) 1065–1084.
[2] W. B. Arveson. Subalgebras of C ∗ -algebras. Acta Math. 123 (1969) 141–224.
[3] B. Blackadar. Operator algebras, volume 122 of Encyclopaedia of Mathematical Sciences. Springer-
Verlag, Berlin, 2006. Theory of C ∗ -algebras and von Neumann algebras, Operator Algebras and
Non-commutative Geometry, III.
[4] D. P. Blecher and M. Neal. Metric characterizations of isometries and of unital operator spaces and
systems. Proc. Amer. Math. Soc. 139 (2011) 985–998.
[5] A. Connes and W. D. van Suijlekom. Spectral truncations in noncommutative geometry and oper-
ator systems. Comm. Math. Phys. 383 (2021) 2021–2067.
[6] K. R. Davidson and M. Kennedy. The Choquet boundary of an operator system. Duke Math. J. 164
(2015) 2989–3004.
[7] N. Doll, H. Schulz-Baldes, and N. Waterstraat. Spectral flow—a functional analytic and index-theoretic
approach, volume 94 of De Gruyter Studies in Mathematics. De Gruyter, Berlin, [2023] ©2023.
[8] E. G. Effros and Z.-J. Ruan. Operator spaces, volume 23 of London Mathematical Society Monographs.
New Series. The Clarendon Press, Oxford University Press, New York, 2000.
[9] G. K. Eleftherakis, E. T. A. Kakariadis, and I. G. Todorov. Morita equivalence for operator systems,
arXiv:2109.12031.
[10] E.-M. Hekkelman. Truncated geometry on the circle. Lett. Math. Phys. 112 (2022) Paper No. 20, 19.
[11] C. Kleski. Boundary representations and pure completely positive maps. J. Operator Theory 71
(2014) 45–62.
[12] T. A. Loring and H. Schulz-Baldes. Spectral flow argument localizing an odd index pairing. Canad.
Math. Bull. 62 (2019) 373–381.
[13] T. A. Loring and H. Schulz-Baldes. The spectral localizer for even index pairings. J. Noncommut.
Geom. 14 (2020) 1–23.
[14] E. Lozano Viesca, J. Schober, and H. Schulz-Baldes. Chern numbers as half-signature of the spec-
tral localizer. J. Math. Phys. 60 (2019) 072101, 13.
[15] P. S. Muhly and B. Solel. An algebraic characterization of boundary representations. In Nonselfad-
joint operator algebras, operator theory, and related topics, volume 104 of Oper. Theory Adv. Appl., pages
189–196. Birkhäuser, Basel, 1998.
[16] H. Oyono-Oyono and G. Yu. On quantitative operator K-theory. Ann. Inst. Fourier (Grenoble) 65
(2015) 605–674.
[17] V. Paulsen. Completely bounded maps and operator algebras, volume 78 of Cambridge Studies in Ad-
vanced Mathematics. Cambridge University Press, Cambridge, 2002.
14 WALTER D. VAN SUIJLEKOM

[18] G. Pisier. Introduction to operator space theory, volume 294 of London Mathematical Society Lecture
Note Series. Cambridge University Press, Cambridge, 2003.
[19] A. Van Daele. K-theory for graded Banach algebras. I. Quart. J. Math. Oxford Ser. (2) 39 (1988)
185–199.
[20] A. Van Daele. K-theory for graded Banach algebras. II. Pacific J. Math. 134 (1988) 377–392.
[21] W. D. van Suijlekom. A generalization of K-theory to operator systems. arXiv:2409.02773.

I NSTITUTE FOR M ATHEMATICS , A STROPHYSICS AND PARTICLE P HYSICS , R ADBOUD U NIVERSITY

N IJMEGEN , H EYENDAALSEWEG 135, 6525 AJ N IJMEGEN , T HE N ETHERLANDS .
Email address: [email protected]