You are on page 1of 6

Letters in Mathematical Physics 17 (1989) 191-196.

191
9 1989by Kluwer Academic Publishers.

Dirac Combs
A. CORDOBA
Departamento de Matem~ticas, Universidad Aut6noma, Canto Blanco, 28049 Madrid, Spain

(Received: 23 March 1988)


Abstract. We consider tempered distributionsgiven by linear combinationsof delta functionsplaced at
differentpoints and whose Fourier transformis also a sum of the delta functions.We show that they can
be characterized as finite superpositionsof periodic structures.

AMS subject classifications (1980).Primary42B10; secondary44A15.

1. I n t r o d u c t i o n

An interesting application of the Fourier transform to crystallography appears in the


interpretation of the diffraction patterns of solid crystals: the diffracted wave function
is the Fourier transform of the electronic density function. We shall define the Fourier
transform by means of the integral

f (~) = ( exp(-27ffx. ~ ) f ( ~ ) d ~ ,
J~ n

for rapidly decreasing smooth functions fl A definition which can be extended by duality
to the space of tempered distribution 6a (R"). Among many important properties of this
transformation, we shall mention the so-called Poisson s u m m a t i o n f o r m u l a : T h e tempered
distribution ~ v~ z- 6v coincides with its own Fourier transform (here b~ denotes the
Dirac delta function at the lattice point v).
In crystallography, such a distribution is usually called a Dirac comb and provides
a natural model for a mono-atomic molecule. Its diffraction pattern is again a discrete
sum of delta functions (Bragg's peaks) placed also on the lattice Z".
One can obtain other lattices by composition with linear mappings:
Let A be a linear isomorphism of ~" and define
x~ = A . v , yt, = (A - ~ ) ' . # , v, # e Z " ,

then we have

6~, = det(A-1) ~ 5 , .

In reference [l], a simple and direct proof is given of the fact that those sets of { x , } ,
{Yv} described above are the only ones for which a Poisson summation formula holds.
This proof gives confirmation of an important crystallographic paradigm: whenever one
sees isolated Bragg's peaks in the Fourier spectrum, one must have a periodic structure,
i.e. a crystal.
192 A. CORDOBA

It then seems natural to ask what happens if our distribution gives different masses
to different deltas. Can these distributions be characterized as finite superpositions of
periodic structures?
Our theorem in this Letter gives a positive solution to this question, but our proof
needs some powerful tools of the harmonic analysis of locally compact Abelian groups,
namely the characterization obtained by P. Cohen for idempotent measures on such a
group G: It * # = # if and only if supp(/~) belongs to the coset ring of the dual group
(i.e. the smallest ring of subsets of G which contains all open cosets). We refer to the
work of P. Rosenthal [2] where this kind of argument has been used to describe the
translation invariant subspaces which are projections of LP(G).

2. Statement of Results
Let the set A = uJY= i A1 be given as a finite disjoint union of subsets of ~n and such
that there exists a positive lower bound for the distances between any two distinct points
in A. For the purpose of this Letter, we shall describe the situation by saying that the
set A is strongly discrete.
Given different complex numbers {aj}j ~ l ..... N we may consider the tempered dis-
tribution
N
It=Za, Z
j=l xeAj

Let us assume that its Fourier transform/~ can also be expressed in the form

and satisfies the property

[b~l ~< C < oo, for every unit cube Q c ~n. (,)
y~e Q

Then we have the following theorem.

THEOREM. Under the hypothesis considered above, each set Aj is a finite disjoint union
of n-dimensional lattices.

3. Proofs
Prior to the presentation of the proof, we shall consider various auxiliary lemmas.

LEMMA 1. Let {xj} c pn be a discrete subset ofgq" which is not relaavely compact, and
suppose that It = Z aj 3x, is a tempered distribution whose Fourier transform ~ can be
expressed in the form ~ = Z bk ~rk and satisfies condition (*). Then necessarily It -x-_O.
Proof. The hypothesis that {xj} is not relatively compact implies that for every positive
number t one can find a ball B(zt; t) disjoint with the support set {xj},
DIRAC COMBS 193

We choose a function V/e ~(R") satisfying supp(~k) = B(0; 1), (b(0) --- 1 and we
consider, for a fixed Yk, the function

We have,

0 = # ( ~ ) = ~ ( ~ ) = t" • b~ ~(t(y~ - Yk))" exp(2niz,-y=),


Ol

which yields

b~ = ~ b= ~(t(y~ - Yk)) exp(2rcizt" (Y~ - Yk))"


ot~k

In particular,

Ib~l ~ ~ Ib~l I ~ ( t ( y ~ - y~))l.


~t#k

Taking the limit when t goes to infinity, we get b k = 0 as a consequence of property (*).
[]

Before presenting our second lemma, let us recall the fact that the discrete subgroups
of R" are lattices and that H, subgroup of R", is not discrete if and only if each
neighbourhood of 0 in R~ contains infinitely many elements of H.

LEMMA 2. Let A = (x + H ) - (x 1 + H I ) . . . . . (x r + 1-1,) ~ fJ, be a strongly discrete


set, where H, H I , . . . , H, are subgroups of R n. Then H is a discrete group.
Proof. By induction on r.

C a s e r = I:A = ( x + H ) - ( x l + H 1 ) # J J .
If(x+H) n(x l+Hl)=~thenA =x+H
and the result is immediate.
If (x + H ) n (x I + H ~ ) ~ }J, then there exist elements h e l l , hi e l l 1 such that
x+h=x l+h~andwehave

(x + H ) n ( x I + Hl)= x + h + HnHl

that is

A = x + [ H - (h + H n H 1 ) ] .

Since H is a disjoint union ofcosets o f H n H1, it is clear thatA discrete implies H n H1


discrete.
Suppose that H is not discrete. Then each open ball B(0; r) centered at the origin in
R" must contain an infinite number of points in H. Since h + H n Hi is discrete,
B(0; r) n (h + H n H~) is a finite set and, therefore, A - x contains an infinity of points
in such a ball B(0; r). This is contrary to the hypothesis thatA is a strongly discrete set.
(The remainder of the induction argument is left to the reader.) []
194 A. CORDOBA

LEMMA 3. Suppose that { e l , . . . , e,,,} are linearly independent vectors in ~q". Then for
every r > 0 the set

D, = {v = k l e I + " " + k,,,e,,,/d(v, Z ' ) < r ; k j e Z , j = 1. . . . . m}

cannot be contained in a finite union of (m - 1)-dimensional planes.


Proof. By induction in m and using the theorem of Herman Weyl about the uniform
distribution modulo 1 of (kO 1. . . . . kOs), if the numbers {1, 01. . . . . 0s} are linearly
independent over the rationals. Observe that the conclusion of the lemma follows easily
if all the vectors ej have rational coordinates. The details are left to the reader.

Proof of the theorem


Let ~ by chosen as in L e m m a I and consider, for each positive integer M, the measure

Then condition (*) implies that for each M, VMis a measure of finite total variation in
uniformly on M: IIVu II c < oo.
The measures Vu have a natural extension as finite measures in ~ , the Bohr
compactification of R~, with uniformly bounded total variation. Therefore, by the
standard weak-star compactness argument, there is a subsequence, which we shall also
denote by { Vu}, converging in the weak star topology to a finite measure ~.
We have

M~cx~ j=l xcA/

Therefore,

=/aj, ifCeAj

Our next step is to use this fact to show that for each set Aj, we can find a finite
measure #j on the Bohr group ~n satisfying that

#j(0=l, if zAj, #j(0=0, ifqeAj.


We prove this by an induction argument on the number of sets Aj.
I f N = 1, then the result is obvious and we can take #l = (1/al)~.
Suppose then that the result holds for N - 1. Consider the measure ~k = V* V -- ak~.
DIRAC COMBS 195

Then we have

akaj , if~eAj, jCk,


vk(~ ) = ~(~)2 _ a~,~(~) ={[ a2

10, if~A-A k.
It is easy to see that there exists k such that there exists j with the property that

a f - akaj ~ a,,2 - aka,,, for m ~ j .


By the induction hypothesis, we can find/~j with the required properties with respect
to Aj. Then the measure 2 = ~ - ajl~j satisfies that its Fourier transform is given by

~(~)={a k-aj, if~eAk, k~ j,

0, ifr162 -Aj,
A
and the values a k - aj, k = 1. . . . , j . . . . . N are all different. We can use the induc-
tion hypothesis again to claim the existence of the required measures #~,, k = 1. . . . . N.
Therefore, we have conditions to apply the theorem of Paul Cohen, on idempotent
measures, to conclude that each Aj belongs to the coset ring of the group ~" provided
with the discrete topology, i.e. as a dual group to R'.
That is, each Aj is a finite union of sets of the form

A = (x + H ) - (x, + H , ) . . . . . (xr + Hr) ~

and by L e m m a 2, we know that H is a discrete subgroup of R" (with respect to the


ordinary topology), that is a lattice. Furthermore, there exist elements hje Hj such that

A =x+{H-(hl+HC~Hi) ..... (hr + H c ~ H r ) }.

We can write such a set in the form

A = x + {(21 + K , ) u " " u (Yk + Ks)} - a l ,

where yj + Kj are cosets of discrete subgroups Kj of R" and A 1 is contained in a finite


union of hyperplanes.
To see this, we merely observe that, in R', if G 1 is a subgroup of the discrete group
G2 and both have equal dimensions (as lattices in R'), then the index [G2: G~ ] must be
finite, i.e. G z = finite union of cosets of Gi, and therefore given g e G 2 ,
G2 - (g + G1) = finite union of cosets of G 1.
The identity

(xl- y~)u(x:- }'2)


= X 1 w X z - {(X~ c~X, ~ Y l ) w ( X f c~X2c~ Y2) u (Y, n Y:)}

valid for general sets Xt, Xz, Y~, Yz, yields the following:
Each set Aj can be written in the form

A s = (x 1 + H1) t..) ' ' ' k.) ( x s + H,) u B 1 - B 2


196 A. CORDOBA

where the Hj are n-dimensional discrete subgroups of ~n and the discrete sets B~, B 2
are contained in a finite union of hyperplanes.
Next we change the last expression in order to have disjoint lattices:
Suppose that (xj + Hj) n (x k + Hk) V~ fJ, then there exist elements hj~ Hi, h k 9 H k
such that xj + hj = x k + hk, that is:

(xj + Hi) c~ (x k + Ilk) = Xj + hj + H: n H k ,

(xj + Hi) u (Xk + Hk) = Xj + hj + (Hj w Ilk).

The hypothesis that As. is strongly discrete implies that Hj c~ H k has dimension n. If not,
after an application of a convenient linear mapping A, we can assume that Hj = Z "

H k = {mlel + ' ' " + mne, l m j 9 Z } ,


where ej = (0(, . . . , 0~) and, at least, one of the 0 / i s not rational.
But then an application of Lemma 3 yields that the set

{Xenkl3yeHj, O<d(x,y)<r}
cannot be contained in a finite union of hyperplanes and, therefore, Hj w H k - B cannot
be strongly discrete for any set B contained on such a union. This is a contradiction
with our hypothesis about Aj.
But if Hj c~ H k v6 ~J has dimension n, then Hj w H k = finite disjoint union of cosets
of Hj n H k. A finite number of applications of this procedure allows us to write each
set Aj in the form

Aj = (xy + Hi,) u . . . u ( x / ( . + H I ( . ) B( -

where the n-dimensional lattices xs + H~ are disjoint and the discrete sets B y, BeJ are
contained in a finite union of hyperplanes.
With the family of sets

A* = (x~ + Hy) w " " w (x/(j) + HI(j) )

we define the following measure

.1 = ~ aj E ~)x 9
j=l xeA 7

Then the Poisson summation formula implies that /~1 is a measure satisfying
condition (*). In that case # - #1 is also a measure whose Fourier transform satisfies
(*), but whose support is not relatively compact, being contained in a finite union of
hyperplanes, and by L e m m a 1 we must have # = #1 which implies that A* = Aj, for each
j = l . . . . . N. []

References
1. C6rdoba, A., La formule sommatoire de Poisson, Compt. Rend. Acad. Sci. Paris 306, S~rie I, 373-376
(1988).
2. Rosenthal, H. P., Projections onto translation invariant subspaces of LP(G), Mere. AMS No. 63, 1966.
3. Meyer, Y., Algebraic Numbers and Harmonic Analysis, North Holland Math., Vol. 2, 1972.
4. Zygmund, A., Trigonometric Series, Cambridge Univ. Press, 1968.

You might also like