1.

The Haar system

Let ψ(x) be the function
ψ(x) = χ[0, 1 ) (x) − χ[ 1 ,1) (x) ,
2 2
and for all k, j ∈ Z we set
ψj,k (x) = 2j/2 ψ(2j x − k) .
Theorem 1.1. The system {ψj,k : j, k ∈ Z} is a complete orthonormal system in L2 (R).
Proof. We begin with noticing that the function ψj,k has support in the interval
Ij,k = [k2−j , (k + 1)2−j ),
that has length 2−j , that is |Ij,k | = 2−j . Moreover,
Z Z
ψj,k dx = ψj,k dx = 0 .
R Ij,k
We make a few remarks and observations about the intervals Ij,k .
• The index j denotes the “generation”, while k denotes its “position” in the generation.
• If k 6= k 0 , Ij,k ∩ Ij,k0 = ∅.
• Given an index j, the intervals {Ij+1,k }k∈Z are obtained by bisecting the intervals
{Ij,k }k∈Z . In particular, bisecting the interval Ij,k gives rise to the intervals Ij+1,2k and
Ij+1,2k+1 , while the interval Ij−1,k is obtained by taking the union of the two adjacent
intervals Ij,2k and Ij,2k+1 in the subsequent generation.
• As a consequence, when j 0 > j and the intervals Ij,k and Ij 0 ,k0 are not disjoint, then
Ij,k ⊇ Ij 0 ,k0 and in fact Ij 0 ,k0 is contained in either the first or the second half of Ij,k .
We now prove the orthonormality of the Haar system {ψj,k }.
Let (j, k), (j 0 , k 0 ) be two pairs of indices.
Suppose first j = j 0 . If k 6= k 0 , so that Ij,k ∩ Ij,k0 = ∅, then
(ψj,k , ψj 0 ,k0 ) = 0 .
If k = k 0 , then Z
(ψj,k , ψj,k ) = kψj,k k2 = 2j dx = 1 .
Ij,k
If j < j 0 , then, by the last observation above,
Z
j/2
|(ψj,k , ψj 0 ,k0 )| = 2 | ψj 0 ,k0 dx| = 0 .
Ij 0 ,k0

Thus, the orthonormality follows.

Next, we wish to show that the system is complete, that is that span {ψk,j : j, k ∈ Z} is dense
in L2 (R).
We define
Vj = span {χIj,k : k ∈ Z} .
It is well known that simple functions are dense in L2 . Hence, finite linear combinations of
characteristic functions of intervals are also dense in L2 . It is easy to see that for every interval
I, the function χI can be approximated by functions in ∪j∈Z Vj (show this). Therefore, the
closure in L2 of span {Vj : j ∈ Z} is all of L2 .
1
2

Furthermore, we observe that:

(1) for every j ∈ Z, Vj−1 ⊆ Vj ;
(2) as a consequence, limN →+∞ VN = L2 (R) (notice that this means that given f ∈ L2 and
ε > 0 there exist N ∈ N and g ∈ VN such that kf − gkL2 < ε);
(3) moreover, ∩j∈Z Vj = {0}, that is, limM →−∞ VM = {0}, since if f ∈ ∩j∈Z Vj , then f is an
L2 function that is constant on intervals of arbitrarily long length, hence f must be 0.
We also notice that
• if we set ϕ = χ[0,1) = χI0,0 , and, as in the case of the ψ’s, ϕj,k (x) = 2j/2 ϕ(2j x − k), then
ϕj,k = 2j/2 χIj,k ;
• as a consequence, Vj = span {ϕj,k : k ∈ Z}.
Next, for j ∈ Z we define
Wj = span {ψj,k : k ∈ Z} .
Notice that we have shown that Wj ⊥ Wj 0 for j 6= j 0 and that our goal is to show that
2
L
j∈Z Wj = L .

Claim. For all N ∈ Z we have the orthogonal decomposition

VN = VN −1 ⊕ WN −1 .

Assume the claim for now and we finish the proof.

From the claim, by induction it follows that
 
VN = VN −1 ⊕ WN −1 = VN −2 ⊕ WN −2 ⊕ WN −1
N
M −1 
= VM ⊕ Wj
j=M

for any M ≤ N − 1. Hence, by (3) above, letting M → −∞,

N
M −1
VN = Wj .
j=−∞

Using (2), letting N → +∞ we obtain

M
L2 (R) = Wj ,
j∈Z

and, modulo the claim, we are done.

We prove the claim. Notice that
 
ϕ0,0 (x) ≡ χ[0,1) (x) = 21 χ[0,2) (x) + 1
2 χ[0,1) (x) − χ[1,2) (x)
1 1
= ϕ
21/2 −1,0
(x) + ψ
21/2 −1,0
(x)
∈ V−1 + W−1 .
By translation we also have
1 1
ϕ0,k (x) = ϕ
21/2 −1,k
(x) + ψ
21/2 −1,k
(x) ∈ V−1 + W−1 .
 3

Therefore, V0 ⊆ V−1 + W−1 . Since W−1 ⊆ V0 (as it is easy to check directly, or using the relation
above and the inclusion V−1 ⊆ V0 ), it follows that
V0 = V−1 + W−1 .
The fact that V−1 ⊥ W−1 implies that the above is an orthogonal sum.
From the equality
V0 = V−1 ⊕ W−1
it is easy to complete the proof of the claim, either repeating an analogous argument at the level
N , or using the fact that the dilation
f 7→ 21/2 f (2·)
maps Vj onto Vj+1 and Wj onto Wj+1 , with inverse
f 7→ 2−1/2 f (2−1 ·) .
(These maps preserve the orthogonality of the V ’s and W ’s.)
This completes the proof. 

2. The Hermite functions

Let D be the differential operator defined on C 1 functions by
1
Df (x) = √ xf (x) + f 0 (x) ,


2
and let D∗ its (formal) adjoint with respect to the L2 -inner product; that is the differential
operator defined by the equality Z Z
(Df )g = f (D∗ g) (2.1)

valid for all Cc1 (R) functions.

We now show that
1
D∗ f (x) = √ xf (x) − f 0 (x) .


(2.2)
2
In fact, if f, g ∈ Cc1 (R) (that is, they are C 1 -functions with compact support), integrating by
parts
Z Z Z
∗ 1


D f (x)g(x) dx = f (x)Dg(x) dx = 2 √ f (x) xg(x) + g 0 (x) dx
R R R
Z Z
1 1
= √2 xf (x)g(x) dx + √2 f (x)g 0 (x) dx
R R
Z  +∞ Z 
= √12 xf (x)g(x) dx + √12 f (x)g(x) − f 0 (x)g(x) dx

−∞
Z R R

√1 xf (x) − f 0 (x) g(x) dx ,

= 2
R
using the fact that f and g vanish at ±∞ (in fact, the vanishing of one of the two functions only
would have sufficed).
For simplicity of notation we write
T = D∗ .
4

The Hermite operator is, by definition,

 2
2 d
H=x − . (2.3)
dx
It is immediate to see that
H = 2T D + I = 2DT − I . (2.4)
Set
1 −x2 /2 1
h0 (x) = e , hk (x) = p T k h0 for k = 1, 2, . . . .
π 1/4 (k!)
We call hk the k-th normalized Hermite function.

Theorem 2.1. The functions {hk }, k = 0, 1, 2, . . . , form a complete orthonormal system for
L2 (R).

In the rest of these notes we prove this theorem.

Claim. For all N, m non-negative integer

 (m) 
sup (1 + |x|)N hk (x) < ∞ , (2.5)
x∈R

where hk is any normalized Hermite function, and h(m) denotes the m-th derivative of the
function h.
2
Disregarding the constants, it suffices to prove the statement for the functions e−x /2 and
2
T e−x /2 .
k

By induction it is easy to see that

 d m 2 2
e−x /2 = pm (x)e−x /2 ,
dx
2
where pm is a polynomial of degree m in x. Since (1 + |x|)N pm (x)e−x /2 is C ∞ , it is bounded
on all compact intervals, and since
2 /2
lim (1 + |x|)N pm (x)e−x = 0,
x→±∞
2
the Claim follows for h0 . For the general case, by induction it easily follows that T k e−x /2 =
2
qk (x)e−x /2 , where qk is a polynomial of degree k, so that
 d m 2 2
T k e−x /2 = pm+k (x)e−x /2 ,


dx
for some (other) polynomials pm+k of degree m + k.
The claim now follows easily.1
We remark that, as a consequence of the claim, the equality (2.1) is valid for when f is such
that f, f 0 ∈ L2 and g is one of the Hermite functions.2

Lemma 2.2. The Hermite functions {hk }, k = 0, 1, 2, . . . , satisfy the following relations:
√
(i) T hk = k + 1hk+1 ;
1We remark that C ∞ functions satisfying the condition (2.5) for all non-negative integers N, m, are called
rapidly decreasing functions, or Schwartz functions.
2Or, more generally, is a Schwartz function.
 5
√
(ii) Dhk = khk−1 ;
(iii) T Dhk = khk and Hhk = (2k + 1)hk .
In particular, the Hermite functions hk are eigenfunctions of the Hermite operator with eigen-
values 2k + 1.

Proof. It is a simple matter to show, by induction, that (i) holds true.

In order to prove (ii), notice that Dh0 = 0; hence (ii) is satisfied for k = 0. Assume by
induction that (ii) is satisfied for k − 1. Observe that from (2.4) it follows that3 DT − T D = I.
Then, using also (i),
1 1 1
Dhk = √ DT k h0 = √ DT hk−1 = √ T D + I hk−1


k! k k
√
k−1 1
= √ T hk−2 + √ hk−1
k k
k−1 1 √
= √ hk−1 + √ hk−1 = khk−1 .
k k
This proves (ii).
Condition (iii) now follows at once. 

Lemma 2.3. We have

(−1)k x2 /2  d k  −x2 /2 
T k f (x) = e e f (x) .
2k/2 dx
In particular,
(−1)k x2 /2
 d k 2
hk (x) =
1/2 e e−x .
π 1/2 2k k! dx

Moreover, if we set
2 /2
Hk (x) = ex hk (x) .
then Hk is a polynomial, k = 0, 1, . . . , and the linear span of {H0 , . . . , Hm } is the linear space
of polynomial of degree less or equal to m.

We mention in passing that the polynomials Hk are called the Hermite polynomials.

Proof. This is easy to prove by induction, and we leave it to the reader. 

We are now ready to prove Thm. 2.1.

Proof of Thm. 2.1. In order to prove that {hk }k=0,1,2,... , is an orthonormal system for L2 , we
first notice that kh0 k2 = 1. Next, we show that, for k > 0, we have that

[D, T k ] = DT k − T k D = kT k−1 . (2.6)

3The operator [A, B] = AB − BA is called the commutator of A and B.

6

When k = 1, we have already noticed that DT − T D = I. In general, for k ≥ 1 we have that4

k−1
 X
A, B k = B j [A, B]B k−1−j


j=0

the equality (2.6) now follows.

Then, let k, ` ≥ 1. We have that
Z Z
1
hk h` = T Dhk h`
k
Z
1
= Dhk Dh`
k
√ Z
`
=√ hk−1 h`−1 .
k
Now we proceed by the induction. If k = ` then khk k = 1.
If k 6= `, then the equality above (valid for all k, `)
Z Z
1
hk h` = T Dhk h`
k
R
it follows that if ` = 0 and k ≥ 1, then hk h0 = 0. Now we proceed by induction as before.
This proves that the Hermite functions form an orthonormal system in L2 (R).
In order to finish the proof, we need to show that the Hermite functions form a complete
system. Let f ∈ L2 be orthogonal to hk for all k = 0, 1, . . . . By Lemma 2.3 f is orthogonal to
2
xk e−x /2 for k = 0, 1, . . . . If f has compact support and I is a compact interval containing the
support of f , we have Z
2 /2
0 = (f, hk ) = f xk e−x ,
I
2
that is, the function f e−x /2 is orthogonal to all the polynomials, that are dense in L2 (I). Hence,
2
f e−x /2 = 0, so that f = 0.
Since the L2 -functions with compact support are dense in L2 , we can easily conclude the
proof. 2

4For,

[A, B k ] = AB k − B k A = ABB k−1 − B k A

= BAB k−1 + [A, B]B k−1 − B k A = BABB k−2 + [A, B]B k−1 − B k A
= B 2 AB k−2 + B[A, B]B k−2 + [A, B]B k−1 − B k A
k−1
X
= B j [A, B]B k−1−j .
j=0