See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228389409

A new class of Banach spaces

Article  in  Journal of Physics A Mathematical and Theoretical · December 2008

DOI: 10.1088/1751-8113/41/49/495206

CITATIONS READS
17 557

2 authors, including:

Tepper L Gill
Howard University
126 PUBLICATIONS   487 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

CANONICAL PROPER-TIME THEORY OF GRAVITATION View project

Theoretical and Mathematical Physics View project

All content following this page was uploaded by Tepper L Gill on 28 October 2016.

The user has requested enhancement of the downloaded file.

 A NEW CLASS OF BANACH SPACES

T. L. GILL AND W. W. ZACHARY

Abstract. In this paper, we construct a new class of separable Banach spaces

KSp , for 1 ≤ p ≤ ∞, each of which contains all of the standard Lp spaces, as

well as the space of finitely additive measures, as compact dense embeddings.

Equally important is the fact that these spaces contain all Henstock-Kurzweil

integrable functions and, in particular, the Feynman kernel and the Dirac

measure, as norm bounded elements. As a first application, we construct the

elementary path integral in the manner originally intended by Feynman. We

then suggest that KS2 is a more appropriate Hilbert space for quantum the-

ory, in that it satisfies the requirements for the Feynman, Heisenberg and

Schrödinger representations, while the conventional choice only satisfies the

requirements for the Heisenberg and Schrödinger representations. As a second

application, we show that the mixed topology on the space of bounded con-

tinuous functions, Cb [Rn ], used to define the weak generator for a semigroup

T (t), is stronger than the norm topology on KSp . (This means that, when

extended to KSp , T (t) is strongly continuous, so that the weak generator on

Cb [Rn ] becomes a strong generator on KSp .)

1. Introduction

The standard university analysis courses tend to produce a natural bias and un-

ease concerning the use of finitely additive set functions as a basis for the general

1991 Mathematics Subject Classification. Primary (45) Secondary(46) .

Key words and phrases. Banach space, Henstock-Kurzweil integral, Feynman path integral.
1
2 GILL AND ZACHARY

theory of integration (despite the efforts of Alexandrov [AX], Bochner [BR], Du-

bins [DU], Dunford and Schwartz [DS], de Finetti [DFN] and Yosida and Hewitt

[YH]). (We should remember that the concept of measure was, and is, important

for geometry and some, but not all, parts of analysis. In other parts, the concept

of integral tends to dominate.)

Without denying an important place for countable additivity, Dubins ([DU],

[BD], and [DUK]) argues forcefully for the intrinsic advantages in using finite ad-

ditivity in the basic axioms of probability theory. (The penetrating analysis of the

foundations of probability theory by de Finetti [DFN] also supports this position.)

In a very interesting paper [DU], Dubins shows that the Wiener process has a num-

ber of ”cousins”, related processes with the same finite dimensional distributions

as the Wiener process. For example, there is one cousin with polynomial paths and

another with piecewise linear paths. Since the Wiener measure is unique, these

cousins must necessarily have finitely additive limiting distributions.

The most important of the finitely additive measures is the one generated by the

Henstock-Kurzweil integral (HK-integral), which generalizes the Lebesgue, Bochner

and Pettis integrals. (It was discovered independently by Henstock [HS1] and

Kurzweil [KW1].) The HK-integral is equivalent to the the Denjoy integral. How-

ever, it is much easier to understand (and learn) compared to the Denjoy and

Lebesgue integrals, and provides useful variants of the same theorems that have

made the Lebesgue integral so important. Furthermore, it arises from a simple

(transparent) generalization of the Riemann integral that is taught in elementary

calculus. Loosely speaking, one uses a version of the Riemann integral with the

interior points chosen first, while the size of the base rectangle around any interior
 A NEW CLASS OF BANACH SPACES 3

point is determined by an arbitrary positive function defined at that point (see

Section 2). For more detail and different perspectives, see Gordon [GO], Henstock

[HS], Kurzweil [KW], or Pfeffer [PF].

1.1. Purpose. Clearly, the most important factor preventing the widespread use

of the HK-integral in engineering, mathematics and physics has been the lack of

a natural Banach space structure for this class of functions (as is the case for the

Lebesgue integral). The purpose of this paper is to introduce a new class of Banach

spaces KSp (Ω), 1 ≤ p ≤ ∞, with Ω ⊂ Rn , (n = 1, 2, · · · ). These are all separable

spaces that contain the corresponding Lp spaces as dense, continuous, compact

embeddings. Our original interest was in the fact that each of these spaces contains

the Denjoy integrable functions, as well as all the finitely additive measures. These

spaces are perfect for the highly oscillatory functions that occur in quantum theory

and nonlinear analysis.

1.2. Summary. In Section 2, we give a brief introduction to the elementary HK-

integral, its properties and relationship to the Lebesgue integral. In Section 3,

we construct the KS-spaces and derive some of the important properties of these

spaces. We then prove that the Fourier transform and convolution operators have

bounded extensions to KS2 . These results are applied to the construction of the

elementary path integral in the manner originally intended by Feynman. We then

discuss our contention that KS2 is a more natural Hilbert space for quantum theory

as compared to the conventional choice.

4 GILL AND ZACHARY

2. HK-Integral

In this section, we introduce the HK-integral (in the simplest case) and present

some of its properties. Our purpose is to provide those researchers unfamiliar with

this integral a concrete sense of its simplicity. Proofs of all stated results can be

found in Gordon [GO]. The general case can be found in Henstock [HS] or Pfeffer

[PF].

Definition 1. Let [a, b] ⊂ R, let δ(t) map [a, b] → (0, ∞), and let P =

{t0 , τ1 , t1 , τ2 , · · · , τn , tn }, where a = t0 6 τ1 6 t1 6 · · · 6 τn 6 tn = b. We

call P an HK-partition for δ, if for 1 6 i 6 n, ti−1 , ti ∈ (τi − δ(τi ), τi + δ(τi )).

Remark 2. Gordon writes P = {(τi , [ti−1 , ti ]) : 1 6 i 6 n} and calls {τi } the tags

and {[ti−1 , ti ]} the collection of tagged intervals. Also, the phrase nearly everywhere

(n.e.) means except for a countable set.

Definition 3. The function f (t), t ∈ [a, b], is said to have a HK-integral if there is

an number F [a, b] such that, for each ε > 0, there exists a function δ from [a, b] →

(0, ∞) such that, whenever P is a HK-partition for δ, then (with ∆ti = ti − ti−1 )

Xn
∆ti f (τi ) − F [a, b] < ε.


i=1

Rb
In this case, we write F [a, b] = (HK)- a f (t)dt.

Theorem 4. Let f (t) : [a, b] → R.

(1) If f (t) is Lebesgue integrable on [a, b], then it is HK-integrable on [a, b] and
Rb Rb
HK- a f (t)dt =L- a f (t)dt.

(2) If f (t) is HK-integrable and bounded on [a, b], then it is Lebesgue integrable

on [a, b].
 A NEW CLASS OF BANACH SPACES 5

(3) If f (t) is HK-integrable and nonnegative on [a, b], then it is Lebesgue inte-

grable on [a, b].

(4) If f (t) is HK-integrable on every measurable subset of [a, b], then it is

Lebesgue integrable on [a, b].

Theorem 5. Let F : [a, b] → R be continuous. If F is differentiable nearly

Rt
everywhere on [a, b], then F 0 is HK-integrable on [a, b] and HK- a F 0 (s)ds = F (t) −

F (a) for each t ∈ [a, b].

The last result shows in what sense we can think of the HK-integral as the reverse

of the derivative. (The result is not true for Lebesgue integrals. The standard

example is F 0 (t) = 2tsin(π/t2 ) − (2π/t)cos(π/t2 ) for all non-rational numbers on

0 < t < 1 and equal to 0 at all rational points.)

3. KSp Spaces

In order to construct the spaces of interest, first recall that the HK-integral

is equivalent to the Denjoy integral (see Henstock [HS] or Pfeffer [PF]). In the

one-dimensional case, Alexiewicz [AL] has shown that the class D(R), of Denjoy

integrable functions, can be normed in the following manner: for f ∈ D(R), define

kf kD by
Z
 s 

kf kD = sup  f (r)dr .
s −∞

It is clear that this is a norm, and it is known that D(R) is not complete (see

Alexiewicz [AL]). Replacing R by Rn , for f ∈ D(Rn ) we have the following gener-

alization:
Z  Z 
   
(3.1) kf kD = sup  f (x)dx = sup  EBr (x)f (x)dx < ∞,
r>0 Br r>0 Rn
6 GILL AND ZACHARY

where Br is any closed cube of diagonal r centered at the origin in Rn , with sides

parallel to the coordinate axes, and EBr (x) is the characteristic function of Br .

Now, fix n, and let Qn be the set {x = (x1 , x2 · · · , xn ) ∈ Rn } such that xi is

rational for each i. Since this is a countable dense set in Rn , we can arrange it as

Qn = {x1 , x2 , x3 , · · ·}. For each l and i, let Bl (xi ) be the closed cube centered at

xi , with sides parallel to the coordinate axes and diagonal rl = 2−l , l ∈ N. Now

choose the natural lexicographical order which maps N × N bijectively to N, and

let {Bk , k ∈ N} be the resulting set of (all) closed cubes {Bl (xi ) |(l, i) ∈ N × N}

centered at a point in Qn . Let Ek (x) be the characteristic function of Bk , so that

Ek (x) is in Lp [Rn ] ∩ L∞ [Rn ] for 1 ≤ p < ∞. Define Fk ( · ) on L1 [Rn ] by

Z
(3.2) Fk (f ) = Ek (x)f (x)dx.
Rn

It is clear that Fk ( · ) is a bounded linear functional on Lp [Rn ] for each k, kFk k∞ ≤

1 and, if Fk (f ) = 0 for all k, f = 0 so that {Fk } is fundamental on Lp [Rn ] for

P∞
1 ≤ p ≤ ∞ . Fix tk > 0 such that k=1 tk = 1 and define a measure dP(x, y) on

Rn × Rn by:
hX ∞ i
dP(x, y) = tk Ek (x)Ek (y) dxdy.
k=1

We first construct our Hilbert space. Define an inner product ( · ) on L1 [Rn ] by

Z
(f, g) = f (x)g(y)∗ dP(x, y)
Rn ×Rn
(3.3) Z  Z ∗
X∞
= tk Ek (x)f (x)dx Ek (y)g(y)dy .
k=1 Rn Rn

We use a particular choice of tk in Gill and Zachary [GZ], which is suggested by

physical analysis in another context. We call the completion of L1 [Rn ], with the

above inner product, the Kuelbs-Steadman space, KS2 [Rn ]. Following suggestions

of Gill and Zachary, Steadman [ST] constructed this space by adapting an approach
 A NEW CLASS OF BANACH SPACES 7

developed by Kuelbs [KB] for other purposes. Her interest was in showing that

L1 [Rn ] can be densely and continuously embedded in a Hilbert space which contains

the HK-integrable functions. To see that this is the case, let f ∈ D[Rn ], then:

X∞ Z 2 Z 2
2     2
kf kKS2 = tk  Ek (x)f (x)dx 6 sup  Ek (x)f (x)dx 6 kf kD ,
k=1 Rn k Rn

so f ∈ KS2 [Rn ].

Theorem 6. For each p, 1 6 p 6 ∞, KS2 [Rn ] ⊃ Lp [Rn ] as a dense subspace.

Proof. By construction, KS2 [Rn ] contains L1 [Rn ] densely, so we need only show

that KS2 [Rn ] ⊃ Lq [Rn ] for q 6= 1. If f ∈ Lq [Rn ] and q < ∞, we have

"
X∞ Z  2qq #1/2
 
kf kKS2 = tk  Ek (x)f (x)dx
k=1 Rn

"
X∞ Z  q2 #1/2
q
6 tk Ek (x) |f (x)| dx
k=1 Rn

Z  q1
q
6 sup Ek (x) |f (x)| dx 6 kf kq .
k Rn

h i2n
1
Hence, f ∈ KS2 [Rn ]. For q = ∞, first note that vol(Bk )2 ≤ √
2 n
, so we have

"
X∞ Z 2 #1/2
 
kf kKS2 = tk  Ek (x)f (x)dx
k=1 R n

i1/2  n
hhX∞
2
i
2 1
6 tk [vol(Bk )] [ess sup |f |] 6 √ kf k∞ .
k=1 2 n

Thus f ∈ KS2 [Rn ], and L∞ [Rn ] ⊂ KS2 [Rn ]. 

The fact that L∞ [Rn ] ⊂ KS2 [Rn ], while KS2 [Rn ] is separable makes it clear in

a very forceful manner that whether a space is separable or not depends on the

topology. Before proceeding to additional study, we need to construct KSp [Rn ].

8 GILL AND ZACHARY

To construct KSp [Rn ] for all p and for f ∈ Lp , define:


 P∞ R p 1/p
k=1 tk Rn Ek (x)f (x)dx , 1 6 p < ∞,

  
kf kKSp =
 R 

 supk>1  Rn Ek (x)f (x)dx , p = ∞.

It is easy to see that k·kKSp defines a norm on Lp . If KSp is the completion of Lp

with respect to this norm, we have:

Theorem 7. For each q, 1 6 q 6 ∞, KSp [Rn ] ⊃ Lq [Rn ] as a dense continuous

embedding.

Proof. As in the previous theorem, by construction KSp [Rn ] contains Lp [Rn ]

densely, so we need only show that KSp [Rn ] ⊃ Lq [Rn ] for q 6= p. First, suppose

that p < ∞. If f ∈ Lq [Rn ] and q < ∞, we have

" Z  qp #1/p
X∞  q
kf kKSp = tk  Ek (x)f (x)dx
k=1 R n

"
X∞ Z  pq #1/p
q
6 tk Ek (x) |f (x)| dx
k=1 Rn

Z  q1
q
6 sup Ek (x) |f (x)| dx 6 kf kq .
k Rn

Hence, f ∈ KSp [Rn ]. For q = ∞, we have

X Z p 1/p
∞  
kf kKSp = tk  Ek (x)f (x)dx
k=1 Rn
hhX∞ i i1/p
6 tk [vol(Bk )]p [ess sup |f |]p 6 M kf k∞ .
k=1

Thus f ∈ KSp [Rn ], and L∞ [Rn ] ⊂ KSp [Rn ]. The case p = ∞ is obvious. 

Theorem 8. For KSp , 1 ≤ p ≤ ∞, we have:

(1) If f, g ∈ KSp , then kf + gkKSp 6 kf kKSp +kgkKSp (Minkowski inequality).

(2) If K is a weakly compact subset of Lp , it is a compact subset of KSp .

 A NEW CLASS OF BANACH SPACES 9

(3) If 1 < p < ∞, then KSp is uniformly convex.

(4) If 1 < p < ∞ and p−1 + q −1 = 1, then the dual space of KSp is KSq .

(5) KS∞ ⊂ KSp , for 1 ≤ p < ∞.

Proof. The proof of (1) follows from the classical case for sums. The proof of (2)

follows from the fact that, if {fn } is any weakly convergent sequence in K with

limit f , then
Z
Ek (x) [fn (x) − f (x)] dx → 0
Rn

for each k. It follows that {fn } converges strongly to f in KSp .

The proof of (3) follows from a modification of the proof of the Clarkson inequal-

ities for lp norms.

In order to prove (4), observe that, for p 6= 2, 1 < p < ∞, the linear functional

Z p−2 Z
2−p
X∞
Ek (y)f (y)∗ dy
 
Lg (f ) = kgkKSp tk  Ek (x)g(x)dx
k=1 Rn Rn

is a unique duality map on KSq for each g ∈ KSp and that KSp is reflexive from
R 
(3). To prove (5), note that f ∈ KS∞ implies that  Rn Ek (x)f (x)dx is uniformly
R p
bounded for all k. It follows that  Rn Ek (x)f (x)dx is uniformly bounded for each

p, 1 ≤ p < ∞. It is now clear from the definition of KS∞ that:

X Z p 1/p
∞  
tk  Ek (x)f (x)dx 6 kf kKS∞ < ∞
k=1 Rn

Note that, since L1 [Rn ] ⊂ KSp [Rn ] and KSp [Rn ] is reflexive for 1 < p < ∞, we
 ∗∗
see that the second dual L1 [Rn ] = M[Rn ] ⊂ KSp [Rn ]. Recall that M[Rn ] is

the space of bounded finitely additive set functions defined on the Borel sets B[Rn ].
10 GILL AND ZACHARY

This space contains the Dirac delta measure and the free-particle Green’s function

for the Feynman integral. We will return to M[Rn ] in the next section.

In many applications, it is convenient to formulate problems on one of the stan-

dard Sobolev spaces Wpm (Rn ). We close this section with an admittedly incomplete,

but most direct approach to the corresponding extension for the KSp spaces. First

recall that the space:

−α/2
Xm n
p (R ) = {Bα ∗ g = (I − 4) g : g ∈ Lp (Rn ), 0 < α < n, 0 < α < m}

coincides with Wpm (Rn ) when 1 < p < ∞ and m > 0, where Bα is the Bessel

potential of order α, 4 is the Laplacian and ∗ is the convolution operator.

p
Theorem 9. The completion of Xm n n
p (R ) relative to the KS (R ) norm defines

the space KSm n m n

p (R ), which contains Wp (R ) as a continuous dense and compact

embedding.

Proof. Since Bα ∈ L1 (Rn ), we see from Young’s inequality for convolutions that

Bα ∗ g ∈ Lp (Rn ) if g ∈ Lp (Rn ) and 0 < α < n. 

R p
In closing, we first recall that a function f such that K
|f (x)| dx < ∞ for

every compact set K in Rn is said to be in Lploc (Rn ). We can easily show that

Lploc (Rn ) ⊂ KSq (Rn ), 1 ≤ q ≤ ∞, for all p, 1 ≤ p ≤ ∞. This means that KSq (Rn )

contains a large class of distributions (see Adams [A]).

3.1. Extension of Fourier and Convolution Operators. Let L[B], L[H] denote

the bounded linear operators on B, H respectively, where we assume that the

separable Banach space B is a continuous dense embedding in the separable Hilbert

 A NEW CLASS OF BANACH SPACES 11

space H. The following is the major result in Gill et al [GBZS]. It generalizes the

well-known result of von Neumann [VN1] for bounded operators on Hilbert spaces.

Theorem 10. Let B be a separable Banach space and let A be a bounded linear

operator on B. Then A has a well-defined adjoint A∗ defined on B such that:

(1) the operator A∗ A ≥ 0 (maximal accretive),

(2) (A∗ A)∗ = A∗ A (selfadjoint), and

(3) I + A∗ A has a bounded inverse.

The proof depends on the fact that, given a separable Banach space B, there

always exist Hilbert spaces H1 and H2 such that H1 ⊂ B ⊂ H2 as continuous dense

embeddings, with H1 determined by H2 (see [GBZS]). If A is any bounded linear

operator on B, we define A∗ by

(3.4) A∗ x = J1−1 [(A1 )0 ]J2 |B (x),

where A1 is A restricted to H1 , J2 |B maps B into H0 2 and J1−1 maps H0 1 onto H1 .

It is not clear that A need have a bounded extension to H2 . On the other hand,

the theorem by Lax [LX] states that:

Theorem 11. If A is a bounded linear operator on B such that A is selfadjoint

(i.e., (Ax, y)H2 = (x, Ay)H2 for all x, y, ∈ B ), then A is bounded on H2 and

kAkH2 ≤ k kAkB with k constant.

Since A∗ A is selfadjoint on B, it is natural to expect that the same is true on

H2 . However, this need not be the case. To obtain a simple counterexample, recall

that, in standard notation, the simplest class of bounded linear operators on B is

12 GILL AND ZACHARY

B ⊗ B 0 , in the sense that:

B ⊗ B 0 : B → B, by Ax = (b ⊗ b0 )x = hx, b0 i b.

0 0
Thus, if b0 is in B 0 \H2 , then J2 {J1−1 [(A1 )0 ]J2 |B (x)} need not be in H2 , so that

A∗ A is not defined as an operator on all of H2 and thus, cannot have a bounded

extension. We can now state the correct extension of Theorem 11.

0
Theorem 12. Let A be a bounded linear operator on B. If B ⊂ H2 , then A has a

bounded extension to L[H2 ], with kAkH2 ≤ k kAkB with k constant.

Proof. The proof is now easy if, we observe that, with the stated condition,
0
J2 {J1−1 [(A1 )0 ]J2 |B (x)} is in H2 for all x ∈ B. It follows that, for any bounded linear

operator A defined on B, the operator T = A∗ A is selfadjoint on H2 . Thus, by

Lax’s theorem, T is bounded on H2 , with kA∗ AkH2 = kAk2H2 ≤ kA∗ AkB ≤ kkAk2B ,

where k = inf {M | kA∗ AkB ≤ M kAk2B }. 

We can now use Theorem 12 to prove that F and C, the Fourier (transform) op-

erator and the convolution operator respectively, defined on L1 [Rn ], have bounded

extensions to KS2 [Rn ]. It should be noted that this theorem also implies that both

operators have bounded extensions to all Lp [Rn ] spaces for 1 ≤ p < ∞. This is the

first proof based on functional analysis, while the traditional proof is obtained via

rather deep methods of (advanced) real analysis.

Theorem 13. Both F and C extend to bounded linear operators on KS2 [Rn ].

Proof. To prove our result, first note that C0 [Rn ], the bounded continuous functions

on Rn which vanish at infinity, is contained in KS2 [Rn ] . Now F is a bounded

linear operator from L1 [Rn ] to C0 [Rn ], so we can consider it as a bounded linear

 A NEW CLASS OF BANACH SPACES 13

operator from L1 [Rn ] to KS2 [Rn ]. Since L1 [Rn ] is dense in KS2 [Rn ] and L∞ [Rn ] ⊂

KS2 [Rn ], by Theorem 12 F extends to a bounded linear operator on KS2 [Rn ].

To prove that C has a bounded extension, fix g in L1 [Rn ] and define Cg on

L1 [Rn ] by:
Z
Cg (f )(x) = g(y)f (x − y)dy.

Once again, since Cg is bounded on L1 [Rn ] and L1 [Rn ] is dense in KS2 [Rn ], by

Theorem 12 it extends to a bounded linear operator on KS2 [Rn ]. Now use the

fact that convolution is commutative to get that Cf is a bounded linear operator

on L1 [Rn ] for all f ∈ KS2 [Rn ]. Another application of Theorem 12 completes the

proof. 

We now return to M[Rn ].

Definition 14. A uniformly bounded sequence {µk } ⊂ M[Rn ] is said to converge

w
weakly to µ (µn −
→ µ), if, for every bounded uniformly continuous function h(x),
Z Z
h(x)dµn → h(x)dµ.
Rn Rn

w s
→ µ in M[Rn ], then µn −
Theorem 15. If µn − → µ (strongly) in KSp [Rn ].

Proof. Since the characteristic function of a closed cube is a bounded uniformly

w
→ µ in M[Rn ] implies that
continuous function, µn −
Z Z
Ek (x)dµn → Ek (x)dµ
Rn Rn

for each k, so that limn→∞ kµn − µkKSp = 0. 

A little reflection gives:

Theorem 16. The space KS2 [Rn ] is a commutative Banach algebra with unit.
14 GILL AND ZACHARY

In closing, it is clear that all bounded linear operators on Lp [Rn ] have extensions

to KS2 [Rn ]. It is easy to see that they also have densely defined closed extensions

to KSp [Rn ] for p 6= 2. We have not been able to show that these extensions are

bounded.

4. Applications

4.1. Markov Processes. In the study of Markov processes, two of the natural

spaces on which to formulate the theory, Cb [Rn ], the space of bounded continuous

functions, or UBC[Rn ], the bounded uniformly continuous functions, do not have

the expected properties. It is well-known that the semigroups associated with

Markov processes, whose generators have unbounded coefficients, are not necessarily

strongly continuous when defined on Cb [Rn ]. This means that the generator of such

a semigroup does not exist in the standard sense. As a consequence, a number of

weaker (equivalent) definitions have been developed in the literature. For a good

discussion of this and related problems see Lorenzi and Bertoldi [LB].

Definition 17. A sequence of functions {fn } in Cb [Rn ] is said to converge to

f in the mixed topology, written τ M -lim fn = f , if supn∈N kfn k∞ 6 M and

kfn − f k∞ → 0 uniformly on every compact subset of Rn .

Theorem 18. If {fn } converges to f in the mixed topology on Cb [Rn ], then {fn }

converges to f in the norm topology of KSp [Rn ] for each 1 ≤ p ≤ ∞.

Proof. It is easy to see that both Cb [Rn ] and UBC[Rn ] are subsets of KSp [Rn ]

for 1 ≤ p ≤ ∞. Thus, it suffices to prove that τ M -lim fn = f implies that

limn→∞ kfn − f kKSp = 0. This now follows from the fact that each box, used

in our definition of the KSp [Rn ] norm, is a compact subset of Rn . 

 A NEW CLASS OF BANACH SPACES 15

Theorem 19. Suppose that T̂ (t) is a transition semigroup defined on Cb [Rn ], with

weak generator Â. Let T (t) be the extension of T̂ (t) to KSp [Rn ]. Then T (t) is

strongly continuous, and the extension A of  to KSp [Rn ] is the strong generator

of T (t).

Proof. First observe that the dual of Cb [Rn ] is M[Rn ], which is contained in

KSp [Rn ] for 1 ≤ p ≤ ∞. Thus we can apply Theorem 12 to show that T̂ (t)

has a bounded extension to KS2 [Rn ]. It is easy to show that the extended operator

T (t) is a semigroup. Since the τ M topology on Cb [Rn ] is stronger then the norm

topology on KS2 [Rn ], we see that the generator A, of T (t) is strong. 

4.2. Feynman Path Integral. Historically, the mathematics community has had

two responses to the introduction of a new mathematical idea or method into

physics. The first response has been to fit the idea or method into an existing

framework. The second and more exciting is when such an idea or method leads to

the development of a new branch of mathematics.

The most prevalent and successful response has been in finding an existing math-

ematical structure that will reasonably accommodate the physical theory and pro-

vide (at least) the framework for mathematical rigor. An excellent example of this

is the introduction of matrix algebra into the Heisenberg formulation of quantum

theory (e.g., matrix mechanics) by Born and Jordan [BHJ]. This made it possi-

ble for Schrödinger to show that, in the non-relativistic case, his wave mechanics

was equivalent to Heisenberg’s theory. This was later shown to be rigorously true

mathematically via the unitary equivalence between l2 and L2 as separable Hilbert

spaces (c.f., von Neumann [VN2]). However, even in this case, we should not con-

clude that this is the complete story. There have always been physical advantages
16 GILL AND ZACHARY

in looking at and working with some problems using the Heisenberg formulation.

In fact, in 1964, Dirac strongly suggested on physical grounds that, at the quan-

tum field level, Heisenberg’s formulation is much more fundamental (see [BO], page

130). Furthermore, recent studies strongly indicate that the mathematical concept

of isometric isomorphism is neither necessary nor sufficient for physical equivalence.

(For example, it is known, [GZA], that the Dirac operator is non-local in time, while

the square-root operator is non-local in space, but are unitarily equivalent.)

In some rare but important instances, there is no obvious mathematical struc-

ture which can completely accommodate the theory in the manner presented by

physicists. In this case, mathematicians have extended and/or adapted an existing

mathematical theory, developed new mathematical structures or suggested (in frus-

tration) that any conclusions derived from the use of these ideas or methods are at

least suspect. Over the last sixty years, all of the above positions have appeared in

response to Feynman’s introduction of his path integral into quantum theory.

Since his path integral is the object of this section, let us consider the simple

case of a free particle in non-relativistic quantum theory in R3 :

∂ψ(x, t) ~2
(4.1) i~ − ∆ψ(x, t) = 0, ψ(x, s) = δ(x − y).
∂t 2m

The solution can be computed directly:

−3/2 " #
 2
2πi~(t − s) im |x − y|
ψ(x, t) = K [x, t; y, s] = exp .
m 2~ (t − s)

Feynman wrote the above solution to equation (4.1) as

n  2 o
x(t)=x
(4.2) K [x, t; y, s] = ∫x(s)=y Dx(τ ) exp im
2~ ∫st  dx
dt
 dτ ,
 A NEW CLASS OF BANACH SPACES 17

where
n  2 o
x(t)=x
∫x(s)=y Dx(τ ) exp ∫st  dx
im
2~ dt
 dτ =:

(4.3)
 
h i3N/2 YN  X N h i
m 2
lim ∫R3 dxj exp ~i m
2ε(N ) (xj − xj−1 )  ,
N →∞ 2πi~ε(N ) 
j=1 j=1

with ε(N ) = (t − s)/N .

Problems

Feynman’s objective was to develop an approach to quantum theory which would

avoid the use of a Hamiltonian. Equation (4.2) can be viewed as an attempt

to ”apparently” define an integral over the space of all continuous paths of the

exponential of an integral of the classical Lagrangian on configuration space. Thus,

his objective was (partly) accomplished.

However, this approach (using the Lagrangian directly) has led to a new method

for quantizing physical systems, called the path integral method. It is now used

almost exclusively by large groups (in all branches of physics) and has also been used

(formally) by researchers in both mathematics and mathematical physics. Thus, we

must conclude that Feynman’s formulation (as he proposed it) is both physically

and mathematically distinct from those of Heisenberg and Schrödinger. (Feynman

showed that it was equivalent to the other two.)

From a mathematical point of view, this leads to a number of problems:

• The kernel K [x, t; y, s] and δ(x) are not in L2 [R3 ], the standard space

for quantum theory.

• The kernel K [x, t; y, s] cannot be used to define a measure.

Thus, a natural question is: Does there exist a separable Hilbert space containing

K [x, t; y, s] and δ(x) which also allows the convolution and Fourier transform as
18 GILL AND ZACHARY

bounded operators? A positive answer to this question is necessary if we are to

make sense of equation (4.3) and have a representation space for the Feynman

formulation of quantum theory (as presented).

The properties of KS2 [Rn ] derived earlier suggest that it may be a more ap-

propriate Hilbert space, compared to L2 [Rn ], for the Feynman formulation. It is

easy to prove that both the position and momentum operators have closed densely

defined extensions to KS2 [Rn ]. Furthermore, the extensions of F and C insure

that both the Schrödinger and Heisenberg theory have faithful representations on

KS2 [Rn ].

Since KS2 [Rn ] contains the space of measures, it follows that all the approx-

imating sequences for the Dirac measure converge strongly to it in the KS2 [Rn ]

topology. (For example, [sin(λ · x)/(λ · x)] ∈ KS2 [Rn ] and converges strongly to

δ(x).) Thus, the finitely additive set function defined on the Borel sets (Feynman

kernel [FH]): (with m = 1 and ~ = 1)

Z .
−n/2 2
Kf [t, x ; s, B] = (2πi(t − s)) exp{i|x − y| 2(t − s)}dy
B

is in KS2 [Rn ] and kKf [t, x ; s, B]kKS 6 1, while kKf [t, x ; s, B]kM = ∞ (the total

variation norm) and

Z
Kf [t, x ; s, B] = Kf [t, x ; τ, dz]Kf [τ, z ; s, B], (HK-integral).
Rn

Definition 20. Let Pn = {t0 , τ1 , t1 , τ2 , · · · , τn , tn } be a HK-partition for a function

δn (s), s ∈ [0, t] for each n, with limn→∞ ∆µn = 0 (mesh). Set ∆tj = tj −tj−1 , τ0 =

0 and, for ψ ∈ KS2 [Rn ], define

 
Z [[λt]] k k Z
X (λt)  Y 
Kf [Dλ x(τ ) ; x(0)] = e−λt Kf [tj , x(τj ) ; tj−1 , dx(τj−1 )] ,
Rn[0,t] k! j=1 Rn 
k=0
 A NEW CLASS OF BANACH SPACES 19

and

Z Z
(4.4) Kf [Dx(τ ) ; x(0)]ψ[x(0)] = lim Kf [Dλ x(τ ) ; x(0)]ψ[x(0)]
Rn[0,t] λ→∞ Rn[0,t]

whenever the limit exists.

Remark 21. In the above definition we have used the Poisson process. This is not

accidental but appears naturally from a physical analysis of the information that is

knowable in the micro-world (see [GZ]). In fact, it has been suggested by Kolokoltsov

[KO] that such jump processes often provide another effective way to give meaning

to the Feynman path integral and also offers a nice approach to Feynman diagrams.

The next result is now elementary, since KS2 [Rn ] is closed under convolution.

Theorem 22. The function ψ(x) ≡ 1 ∈ KS2 [Rn ] and

Z .
1 2
Kf [Dx(τ ) ; x(s)] = Kf [t, x ; s, y] = √ exp{i|x − y| 2(t − s)}.
Rn[s,t] [2πi(t−s)]n

The above result is what Feynman was trying to obtain without the appropriate

space. A more general (sum over paths) result, that covers almost all application

areas, will appear later, where these spaces have been used to provide a generaliza-

tion of the constructive representation theory for the Feynman operator calculus

(see [GZ1] and also [GZ] for other applications).

If we treat K [x, t; y, s] as the kernel for an operator acting on good initial

data, then a partial solution has been obtained by a number of workers. (See [GZ1]

for references to all the important contributions in this direction.)

A related approach to the Feynman path integral can be found in the work of

Fujiwara and Kumano-go (see [FK1], [FK2] and references therein). For a survey
20 GILL AND ZACHARY

of this approach, see [FK3]. They have systematically developed a time-slicing ap-

proximation method that covers a large portion of classical quantum theory. They

restrict themselves to scalar potentials with polynomial growth. However, their

method seems general enough to eventually include the additional cases. (They

show the power of their approach by providing an analytic formula for the second

term of the semi-classical asymptotic expansion of the Feynman path integral.)

4.3. Examples. A standard method is to compute the Wiener path integral for

the problem under consideration and then use analytic continuation in the mass to

provide a rigourous meaning for the Feynman path integral. The following example

provides a path integral representation for a problem that cannot be solved using

analytic continuation via a Gaussian kernel (see Gill and Zachary [GZ3]). It is

shown that, if the vector potential A is constant, µ = mc/~, and β is the standard

beta matrix, then the solution to the square-root equation for a spin 1/2 particle:
 q 
e

2
2 2 4
i~∂ψ(x, t)/∂t = β c p − c A + m c ψ(x, t), ψ(x, 0) = ψ0 (x),

is given by:
Z  
ie
ψ(x, t) = U[t, 0]ψ0 (x) = exp (x − y) · A K [x, t ; y, 0] ψ0 (y)dy,
2~c
R3

where
 h 1/2
i
(1)
 −H2 µ(c2 t2 −|| x−y ||2 )



 [c2 t2 −|| x−y ||2 ] , ct < −||x||,
2 

ictµ β
h i
2 2 2 1/2
−2iK2 µ(|| x−y || −c t )
K [x, t ; y, 0] = , c |t| < ||x||,
4π   π[|| x−y ||2 −c2 t2 ]
h i
(2) 2 1/2
 2 2

 H2 µ(c t −|| x−y || )


[c2 t2 −|| x−y ||2 ] , ct > ||x||.

The function K2 ( · ) is a modified Bessel function of the third kind of second or-
(1) (2)
der, while H2 , H2 are Hankel functions (see Gradshteyn and Ryzhik [GRRZ]).

Thus, we have a kernel that is far from the standard form. This example was first
 A NEW CLASS OF BANACH SPACES 21

introduced in [GZ2], where we only considered the kernel for the Bessel function

term. In that case, it was shown that, under appropriate conditions, that term

will reduce to the free-particle Feynman kernel and, if we set µ = 0, we get the

kernel for a (spin 1/2) massless particle. In closing this section, we remark that

the square-root operator is unitarily equivalent to the Dirac operator (in the case

discussed).

4.4. The Kernel Problem. Since any semigroup that has a kernel representa-

tion will automatically generate a path integral via the reproducing property, a

fundamental question is: Under what general conditions can we expect a given

(time-independent) generator of a semigroup to have an associated kernel? In this

section we discuss a class of general conditions for unitary groups. It will be clear

that the results of this section carry over to semigroups with minor changes.

Let A(x, p) denote a k × k matrix operator [Aij (x, p)], i, j = 1, 2, · · · , k, whose

components are pseudodifferential operators with symbols aij (x, η) ∈ C∞ (Rn ×

Rn ) and we have, for any multi-indices α and β,

 
 (α)
(4.5) aij(β) (x, η) 6 Cαβ (1 + |η|)m−ξ|α|+δ|β| ,


where

(α)
aij(β) (x, η) = ∂ α pβ aij (x, η),

with ∂l = ∂/∂ηl , and pl = (1/i)(∂/∂xl ). The multi-indices are defined in the

Pn
usual manner by α = (α1 , · · · , αn ) for integers αj ≥ 0, and |α| = j=1 αj , with

similar definitions for β. The notation for derivatives is ∂ α = ∂1α1 · · · ∂nαn and

pβ = pβ1 1 · · · pβnn . Here, m, β, and δ are real numbers satisfying 0 ≤ δ < ξ.

22 GILL AND ZACHARY

m
Equation (4.5) states that each aij (x, η) belongs to the symbol class Sξ,δ (see

[SH]).

Let a(x, η) = [aij (x, η)] be the matrix-valued symbol for A(x, η), and let

λ1 (x, η) · · · λk (x, η) be its eigenvalues. If | · | is the norm in the space of k × k

matrices, we assume that the following conditions are satisfied by a(x, η). For

0 < c0 < | η| and x ∈ Rn we have

 
 (α)
(1) a(β) (x, η) ≤ Cαβ |a(x, η)| (1 + |η)|)−ξ|α|+δ|β| (hypoellipticity),


(2) λ0 (x, η) = max Re λj (x, η) < 0,

16j6k
|a(x,η)|  
(3) |λ0 (x,η)| = O (1 + |η|)(ξ−δ)/(2k−ε) , ε > 0.

We assume that A(x, p) is a selfadjoint generator of a unitary group U (t, 0), so that

U (t, 0)ψ0 (x) = exp[(i/~)tA(x, p)]ψ0 (x) = ψ(x, t)

solves the Cauchy problem

(4.6) (i~)∂ψ(x, t)/∂t = A(x, p)ψ(x, t), ψ(x, t) = ψ0 (x).

Definition 23. We say that Q(x, t, η, 0) is a symbol for the Cauchy problem (4.6)

if ψ(x, t) has a representation of the form

Z
(4.7) ψ(x, t) = (2π)−n/2 ei(x,η) Q(x, t, η, 0)ψ̂0 (η)dη.
Rn

It is sufficient that ψ0 belongs to the Schwartz space S(Rn ), which is contained

in the domain of A(x, p), in order that (4.7) makes sense.

Following Shishmarev [SH], and using the theory of Fourier integral operators,

we can define an operator-valued kernel for U (t, 0) by

Z
−n/2
K(x, t ; y, 0) = (2π) ei(x−y,η) Q(x, t, η, 0)dη,
Rn
 A NEW CLASS OF BANACH SPACES 23

so that

Z
−n/2
(4.8) ψ(x, t) = U (t, 0)ψ0 (x) = (2π) K(x, t ; y, 0)ψ0 (y)dy.
Rn

The following results are due to Shishmarev [SH].

Theorem 24. If A(x, p) is a selfadjoint generator of a strongly continuous unitary

group with domain D, S(Rn ) ⊂ D in L2 (Rn ), such that conditions (1)-(3) are

satisfied. Then there exists precisely one symbol Q(x, t, η, 0) for the Cauchy problem

(4.6).

Theorem 25. If we replace our condition (3) in Theorem 24 by the stronger con-

dition

|a(x,η)|
(30 )
 
|λ0 (x,η)| = O (1 + |η|)(ξ−δ)/(3k−1−ε) , ε > 0, |η| > c0 ,

then the symbol Q(x, t, η, 0) of the Cauchy problem (4.6) has the asymptotic

behavior near t = 0:

Q(x, t, η, 0) = exp[−(i/~)ta(x, η)] + o(1),

uniformly for x, y ∈ Rn .

Now, using Theorem 24 we see that, under the stronger condition (3’), the kernel

K(x, t ; y, 0) satisfies
Z
dη
K(x, t; y, 0) = exp[i(x − y, η) − (i/~)ta(x, η)]
Rn (2π)n/2
Z
dη
+ exp[i(x − y, η)] o(1).
Rn (2π)n/2

In order to see the power of KS2 (Rn ), first note that A(x, p) has a selfadjoint

extension to KS2 (Rn ), which also generates a unitary group. This means that we
24 GILL AND ZACHARY

can construct a path integral in the same (identical) way as was done for the free-

m
particle propagator (i.e., for all Hamiltonians with symbols in Sα,δ ). Furthermore,

it follows that the same comment applies to any Hamiltonian that has a kernel rep-

resentation, independent of its symbol class. The important point of this discussion

is that no initial data nor Gaussian form for the kernel is required!

Discussion. A natural reaction to any suggestion that we replace the Lebesgue

integral by one based on a finitely additive measure would be negative. After all

we would lose all of the advantages of the powerful theorems (dominated conver-

gence theorem, monotone convergence theorem, etc) that depend on the countable

additivity of the measure. Those strongly vested in using L2 for the C ∗ -algebra

approach to quantum theory via the GNS construction may also feel obliged to

object to such a proposed change. These are all reasonable concerns. However, we

do not lose any of the powerful theorems found via countable additivity. First of all

the HK-measure is an extension of the Lebesgue measure so that all of its power is

still available to us. In fact, Henstock has extended each of the standard theorems

to the HK-integral (see [HS]). Those concerned with the C ∗ -algebra approach to

quantum theory need not be concerned since KS2 is a separable Hilbert space and

is also amenable to the GNS construction.

Conclusion

In this paper we have shown how to construct a natural class of separable Banach

spaces KSp which parallels the standard Lp spaces but contains them as dense

compact embeddings. These spaces are of particular interest because they contain

the Henstock-Kurzweil integrable functions and the HK-measure, which generalizes

 A NEW CLASS OF BANACH SPACES 25

the Lebesgue measure. We have also constructed the corresponding spaces KSm
p of

Sobolev type.

We have used KS2 to construct the free-particle path integral in the manner

originally intended by Feynman. We have suggested that KS2 has a claim as the

natural representation space for the Feynman formulation of quantum theory in that

it allows representations for both the Heisenberg and Schrödinger representations,

a property not shared by L2 .

In the analytical theory of Markov processes, it is well-known that, in general, the

semigroup T (t) associated with the process is not strongly continuous on Cb [Rn ],

the space of bounded continuous functions or UBC[Rn ], the bounded uniformly

continuous functions. We have shown that the weak generator defined by the mixed

locally convex topology on Cb [Rn ] is a strong generator on KSp [Rn ] (e.g., T (t) is

strongly continuous on KSp [Rn ] for 1 ≤ p ≤ ∞).

References

[A] R. A. Adams, Sobolev Spaces, Academic Press, New York,(1975).

[AL] A. Alexiewicz, Linear functionals on Denjoy-integrable functions, Colloq. Math., 1

(1948), 289-293.

[AX] A. D. Alexandroff, Additive set functions in abstract spaces, I-III, Mat. Sbornik N.

S., 8 ( 50) (1940), 307-348; Ibid. 9 ( 51) (1941), 563-628; Ibid. 13 ( 55) (1943),

169-238.

[BD] D. Blackwell and L. E. Dubins, On existence and nonexistence of proper, regular

conditional distributions, Ann. Prob., 3 (1975), 741-752.

[BHJ] M. Born, W. Heisenberg, and P. Jordan Sür Quantenmechanik II, Zeits. f. Phys.,

35 (1925), 557-615.
26 GILL AND ZACHARY

[BO] M. Born, Atomic Physics, eighth edition, Dover Publications, New York (1969).

[BR] S. Bochner, Integration von Funktionen, deren Werte die Elemente eines Vector-

raumes sind, Fund. Math., 20 (1933), 262-276.

[DFN] B. de Finetti, Theory of Probability, Vol. I, J. Wiley, New York, (1974).

[DS] N. Dunford and J. T. Schwartz, Linear Operators Part I: General Theory, Wiley

Classics edition, Wiley Interscience (1988).

[DU] L. E. Dubins, Paths of finitely additive Brownian motion need not be bizarre, Sem-

inaire de Probabilités XXXIII, J. Azéma, M. Émery, M. Ledoux and M. Yor (Eds.),

Lecture Notes in Math. 1709, 395-396, Springer-Verlag, Berlin-Heidelberg, (1999).

[DUK] L. E. Dubins and K. Prikry, On the existence of disintegrations, Seminaire de

Probabilités XXIX, J. Azéma, M. Émery, P. A. Meyer and M. Yor (Eds.), Lecture

Notes in Math. 1613, 248-259, Springer-Verlag, Berlin-Heidelberg, (1995).

[FH] R. P. Feynman and A. R. Hibbs, Quantum Mechanics and Path Integrals, McGraw-

Hill, New York, (1965).

[FK1] D. Fujiwara and N. Kumano-go, An improved remainder estimate of stationary

phase method for some oscillatory integrals over space of large dimension, Funkcialaj

Ekvacioj 49 (2006), 59-86.

[FK2] D. Fujiwara and N. Kumano-go, The second term of the semi-classical asymptotic

expansion for Feynman path integrals with integrand of polynomial growth, J. Math.

Soc. Japan 58 (2006), 837-867.

[FK3] D. Fujiwara and N. Kumano-go, Feynman path integrals and semi-classical approx-

imation, RIMS Kôkyûroku Bessatsu B5 (2008), 241-263.

[FW] I. Fujiwara, Operator calculus of quantized operator, Prog. Theor. Phys. 7 (1952),

433-448.

[GBZS] T. Gill, S. Basu, W. W. Zachary and V. Steadman, Adjoint for operators in

Banach spaces, Proceedings of the American Mathematical Society, 132 (2004),1429-

1434.
 A NEW CLASS OF BANACH SPACES 27

[GO] R. A. Gordon, The Integrals of Lebesgue, Denjoy, Perron and Henstock, Graduate

Studies in Mathematics, Vol. 4, Amer. Math. Soc., (1994).

[GRRZ] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products ,

Academic Press, New York, (1965).

[GS] J. A. Goldstein, Semigroups of Linear Operators and Applications, Oxford U. Press,

New York, (1985).

[GZ] T. L. Gill and W. W. Zachary, Foundations for relativistic quantum theory I: Feyn-

man’s operator calculus and the Dyson conjectures, Journal of Mathematical Physics

43 (2002), 69-93.

[GZ1] T. L. Gill and W. W. Zachary, Constructive representation theory for the

Feynman operator calculus, submitted for publication, J. Diff. Equations. (see

http://teppergill.googlepages.com/tepperlgill)

[GZ2] T. L. Gill and W. W. Zachary, Time-ordered operators and Feynman-Dyson alge-

bras, Journal of Mathematical Physics 28 (1987), 1459-1470.

[GZ3] T. L. Gill and W. W. Zachary, Analytic representation of the square-root operator,

Journal of Physics A: Math. and Gen. 38 (2005), 2479-2496.

[GZA] T. L. Gill, W. W. Zachary and M. Alfred, Analytic representation of the Dirac

equation, Journal of Physics A: Math. and Gen. 38 (2005), 6955-6976.

[HS] R. Henstock, The General Theory of Integration, Clarendon Press, Oxford, (1991).

[HS1] R. Henstock, A Riemann-type integral of Lebesque power, Canadian Journal of

Mathematics 20 (1968), 79-87.

[KB] J. Kuelbs, Gaussian measures on a Banach space, Journal of Functional Analysis 5

(1970), 354–367.

[KO] V. Kolokoltsov, ’A new path integral representation for the solutions of the

Schrödinger equation, Math. Proc. Cam. Phil. Soc. 32 (2002), 353-375.

[KW] J. Kurzweil, Nichtabsolut konvergente Integrale, Teubner-Texte zür Mathematik,

Band 26, Teubner Verlagsgesellschaft, Leipzig, (1980).

 28 GILL AND ZACHARY

[KW1] J. Kurzweil, Generalized ordinary differential equations and continuous dependence

on a parameter, Czechoslovak Math. Journal 7 (1957), 418-449.

[LB] L. Lorenzi and M. Bertoldi, Analytical Methods for Markov Semigroups, Monographs

and Textbooks in Pure and Applied Mathematics, Chapman & Hall/CRC, New York,

(2007).

[LX] P. D. Lax, Symmetrizable linear tranformations, Comm. Pure Appl. Math. 7 (1954),

633-647.

[M] V. P. Maslov, Operational Methods, Mir, Moscow, (1973). English translation 1976

(revised from the Russian edition).

[PF] W. F. Pfeffer, The Riemann Approach to Integration: Local Geometric Theory, Cam-

bridge Tracts in Mathematics 109, Cambridge University Press, (1993).

[SH] I. A. Shishmarev, On the Cauchy problem and T-products for hypoelliptic systems,

Math. USSR Izvestiya 20 (1983), 577-609.

[ST] V. Steadman, Theory of operators on Banach spaces, Ph.D thesis, Howard University,

1988.

[VN1] J. von Neumann, Über adjungierte Funktionaloperatoren, Annals of Mathematics

33 (1932), 294–310.

[VN2] J. von Neumann, Mathematical Foundations of Quantum Mechanics, translated by

R. T. Beyer. Princeton University Press, Princeton New Jersey (1955).

[YH] K. Yosida and E. Hewitt, Finitely additive measures, Trans. Amer. Math. Soc. 72

(1952), 46-66.

(Tepper L. Gill) Department of Electrical & Computer Engineering, and Mathematics,

Howard University, Washington DC 20059, USA, E-mail : [email protected]

(Woodford W. Zachary) Department of Electrical & Computer Engineering, Howard

University, Washington DC 20059, USA, E-mail : [email protected]

View publication stats