In this section, if we do not specifically mention, we always assume thatX is aK-regular tree with measure and metric as in Section2.1.
Lemma 3.1 Let1≤p <∞. For everyf ∈ Lp(X), we have that Then it follows from Fubini’s Theorem that
exists forν-a.e.ξ ∈∂X and that the trace Trf satisfies the norm estimates.
To show that the limit in Eq. 3.1exists for ν-a.e. ξ ∈ ∂X, it suffices to show that the Forp >1, it follows from the H¨older inequality that
| ˜f∗(ξ )|p |f (0)|p + we obtain by means of Fubini’s theorem that
Here in the above estimates, the notationsIx andj (x)are the same ones as those we used in Lemma 3.1. Sinceν(Ix)≈ K−|x| andR1 <+∞, we further obtain that
Trace and Density Results on Regular Trees
Hence we obtain from estimates Eq.3.3and Eq.3.4thatf˜∗ is inLpν(∂X) for 1≤ p <
∞, which gives the existence of the limits in Eq.3.1forν-a.e.ξ ∈ ∂X. In particular, since
| ˜f| ≤ ˜f∗, we have the estimate
∂X
| ˜f|pdν|f (0)|p +
X
gfpdμ, and hence the norm estimate
˜fLpν(∂X) |f (0)| +
X
gfpdμ
1/p
= fN˙1,p(X).
Since every f ∈ N1,p(X)is locally absolutely continuous, a direct computation gives the estimate |f (0)| fN1,p(X). Hence we obtain the following result from the above theorem.
Corollary 3.3 Let1 ≤ p < ∞and assume that Rp < +∞. Then the trace Tr inEq.1.1 gives a bounded linear operator Tr :N1,p(X)→Lpν(∂X).
Next, we study non-existence of the traces when Rp = ∞. Before going to the main theorems, we introduce the following lemma.
Lemma 3.4([31]) Let(, d, μ)be aσ-finite metric measure space. Then the following conditions on(, d, μ)are equivalent:
(i) Lp()⊂Lq()for allp, q ∈ (0,∞)withp > q;
(ii) μ() <+∞.
Theorem 3.5 Let 1 ≤ p < ∞ and assume that Rp = +∞. Then there exists a function u∈ ˙N1,p(X)such that
[0,ξ )limx→ξu(x)= +∞, for all ξ ∈∂X. (3.5) Proof To construct the function u ∈ ˙N1,p(X) satisfying Eq. 3.5, it suffices to find a nonnegative measurable functiong : [0,∞)→ [0,∞]such that
+∞
0 g(t)λ(t) dt = +∞
+∞
0 g(t)pμ(t)Ktdt <+∞. (3.6) Given suchg, we may define the functionuby settingu(0)=0 andu(x)=|x|
0 g(t)λ(t) dt for anyx ∈ X. Then it follows from the definition of upper gradient thatgu :X → [0,∞]
defined bygu(x)=g(|x|)is an upper gradient ofu. Moreover, we obtain that gupLp(X) =
X
gupdμ≈ +∞
0
g(t)pμ(t)Ktdt <+∞. Hence the condition Eq.3.6impliesu∈ ˙N1,p(X)and that Eq.3.5holds.
Forp =1, sinceR1 =μ(t)Kλ(t)t
L∞([0,∞)) = ∞, the sets Ak :=
t ∈ [0,∞): λ(t)
μ(t)Kt ≥2k
, k ∈ N
form a nonincreasing sequence of subsets of[0,∞)and we have
Bkn λ(t) dt <+∞by replacingBkn with a suitable bounded subset if necessary. Then we defineg by setting
and from the definition ofBkn that +∞ it suffices to show the existence of a functionα satisfying
+∞
Remark 3.6 If additionallyμ(X) < ∞, instead of constructing the above increasing func-tion, we may easily modify the construction so as to obtain a piecewise monotone function
Trace and Density Results on Regular Trees
u ∈ N1,p(X) with values in[0,1] so thatu(x) = 1 when |x| = t2j and u(x) = 0 when
|x| = t2j+1, where tk → ∞ as k → ∞. Then this oscillatory function u belongs to N1,p(X), but has no limit along any geodesic ray. Hence we obtain the following result.
Proposition 3.7 Let 1 ≤ p < ∞ and assume that Rp = +∞. Ifμ(X) < ∞, then there exists a functionu∈N1,p(X)such thatlim[0,ξ )x→ξ u(x)does not exist for anyξ ∈∂X.
Remark 3.8 Since our weights only depend on the distance to the root, Theorem 3.2 and Theorem 3.5 boil down to embeddings on the positive real axis. One of the key properties is thatRp <∞if and only if
Lp(R+, Ktμ(t) dt) ⊂L1(R+, λ(t) dt) (3.9) where λ and μ are defined on [0,∞) as specified in the introduction. Consequently, if μ(X) < ∞, then Rp < ∞ implies Rq < ∞ whenever 1 ≤ p < q < ∞. However such an implication does not hold true ifμ(X) = ∞, but finiteness ofRp is still subject to interpolation, i.e. ifRp <∞andRr < ∞thenRq <∞for everyq ∈ [p, r].
The above results give the full answers to the trace results for the homogeneous New-tonian space N˙1,p(X)and also for the Newtonian space N1,p(X) when μ(X) < ∞. We continue towards the caseμ(X)= ∞.
Proposition 3.9 Let1≤p <∞and assumeμ(X)=∞. Then for everyf∈Lp(X), we have lim inf
[0,ξ )x→ξ|f (x)| =0, for a.e.ξ ∈ ∂X, (3.10) and hence Trf =0if Trf exists.
Proof Assume that Eq. 3.10 is false. Then there exist a function f ∈ Lp(X) and a set E ⊂∂X withν(E) >0 such that
lim inf
[0,ξ )x→ξ|f (x)| >0, for allξ ∈ E.
Hence for each ξ ∈ E, there exist a constant (ξ ) > 0 and an integer N (ξ ) := N ((ξ )) such that
|f (x)| ≥(ξ ) >0, for allx ∈ [0, ξ ) with |x| ≥N (ξ ).
It follows from Lemma 3.1 that fpLp(X) =
X
|f (x)|pdμ≈
∂X
[0,ξ )
|f (x)|pKj (x)dμ(x) dν(ξ )
≥
E
{x∈[0,ξ ):|x|≥N (ξ )}|f (x)|pKj (x)dμ(x) dν(ξ )
≥
E
{x∈[0,ξ ):|x|≥N (ξ )}(ξ )pKj (x)dμ(x) dν(ξ )
=
E
(ξ )p ∞
N (ξ )
Kj (t)μ(t) dt dν(ξ ),
wherej (t)is the largest integer such thatj (t) ≤t+1. Sinceμ(X)= ∞andμ ∈L1loc(X), for every integerN (ξ ), we have
∞
N (ξ )
Kj (t)μ(t) dt = ∞.
Since(ξ ) >0 for eachξ ∈E andν(E) >0, we obtain that fpLp(X) = +∞,
which contradicts the fact thatf ∈Lp(X). Thus Eq.3.10holds.
If Trf exists, then Tr|f| also exists. It follows from the definition of the trace Eq.1.1 and Eq.3.10that Tr|f| =0. Hence Trf = 0.
Proposition 3.10 AssumeR1 = +∞. Then there exists a functionu ∈ N1,1(X)such that lim[0,ξ )x→ξu(x)does not exist, for anyξ ∈∂X.
Proof It follows fromR1= μ(t)Kλ(t)t
L∞([0,∞)) = ∞that the sequence of sets Ek :=
t ∈ [0,∞): λ(t)
μ(t)Kt ≥2k
satisfies
|Ek|>0 for anyk ∈N. Hence we may choose a sequence{tk :tk ∈ [0,∞)}k∈N with
tk → ∞ as k → ∞ and |Ek ∩ [tk−1, tk]|> 0 for any k ∈N. (3.11) Sinceμ∈L1loc([0,∞)), we have that for eachk ∈ N,
0<
tk
tk−1
μ(t)Ktdt =: Mk <∞.
By the absolute continuity of integral with respect to measure, we may divide the interval [tk−1, tk]into2kMksubintervals{Ij}j whose interiors are pairwise disjoint such that
2kMk j=1
Ij = [tk−1, tk] and 0<
Ij
μ(t)Ktdt ≤2−k. (3.12) Since|Ek ∩ [tk−1, tk]| > 0 from Eq. 3.11, we obtain there is at least one subintervalIk ∈ {Ij}j such that|Ek ∩Ik|>0. Then we define a functiongby setting
g(t)=
2
Ek∩Ikλ(t) dt, if t ∈Ek ∩Ik, k ∈N;
0, otherwise.
Since λ(t) is always positive and λ ∈ L1loc([0,∞)), the above definition is well-defined.
Next we construct the functionu. For anyk ∈ N, since we have tk
tk−1
g(t)λ(t) dt =
Ek∩Ik
2λ(t)
Ek∩Ikλ(t) dt dt =2, (3.13) we may apply the same idea of construction as in Remark 3.6 on{x ∈X :tk−1 ≤ |x| ≤tk} to obtain a piecewise monotone function u with upper gradient gu(x) = g(|x|)and with values in [0,1]so thatu(x) = 0 when |x| = tk−1, tk andu(x) = 1 when |x| = tk where tk−1 < tk < tk. Then the functionuhas no limit along any geodesic ray.
Trace and Density Results on Regular Trees
Thus it remains to show that u ∈ N1,1(X). We first estimate theL1-norm of the upper gradientguofu. By the definitions of functiongand ofEk, it follows from estimate Eq.3.13
that Given[0, ξ )satisfying the above inequality, we have
klim→∞
Ak
|f (x(t))|pKtμ(t)dt =0.
Sincef is continuous and
AkKtμ(t) dt =1 for eachk, there exists atk ∈Ak such that
sinceRp < ∞. Since∞
2.⇒3.This implication is trivial.
3.⇒1.Fix p > 1, and suppose that Rp = ∞. Then, for the sequence of subintervals finiteness of the total measure with
kμ(Ik) < ∞ so as to obtain a piecewise monotone functionu ∈ N1,p(X)with upper gradientgu(x) =g(|x|),whereg is from above, and so thatuhas no limit along any geodesic ray. This contradicts 3.
Forp = 1, we have R1 = R1. For p > 1, it is easy to check that Rp < ∞ implies Rp <∞, while the inverse does not hold true. Furthermore, we will show that the finiteness ofRp will not imply the finiteness ofRq for any 1≤q <∞.
For simplicity, we consider the special case whereλandμare piecewise constant. More precisely, assume that
λ(t)=λj, μ(t)=μj, fort ∈ [j, j +1), j ∈ N,
Trace and Density Results on Regular Trees
where{λj}j∈N and{μj}j∈N are two sequences of positive and finite real numbers. Then ds =d λ(z)=λjd|z| and d μ(z)=μjd|z|, for j ≤ |z|< j +1, j ∈ N. (3.15) We begin with easily checkable conditions.
Lemma 3.12 Let 1 ≤ p < ∞. Let Xbe a K-regular tree with measure and metric as in Eq.3.15. Then the following hold:
(i) Rp <∞if there exists a constantM >0 such that
sup
forp >1, and
Remark 3.13 The conditions in Lemma 3.12 to determine whetherRp is finite or not are only sufficient conditions but not necessary ones. Towards this:
For (i), pickμj =K−j(1+j )−1,(λj)p =2−j(1+j )−1. ThenRp <∞but
Remark 3.15 The above examples show that Rp < ∞ does not guarantee that Rq < ∞ for some 1 ≤ q < ∞ when μ(X) = ∞ and p > 1. Hence the existence of the trace Tr :N1,p →Lpν(∂X)is not equivalent to the finiteness ofRq for some 1≤q < ∞.
4 Density
In this section, we focus on the density properties of compactly supported functions in N1,p(X)and in N˙1,p(X), 1≤ p < ∞. The function1is defined by1(x) = 1 for allx in Xand we abuse the notation by using∇uto denoteguif needed for convenience.
Our first result is an analog of the corresponding result for infinite networks [32], also see [24].
Lemma 4.1 Let1≤p <∞and assume thatμ(X) <∞. Then we have that N01,p(X)=N1,p(X) ⇐⇒ 1∈ N01,p(X).
Proof Since it follows from μ(X) < ∞ that 1 ∈ N1,p(X), we obtain that N01,p(X) = N1,p(X)implies1∈ N01,p(X).
Towards the other direction, the hypothesis 1 ∈ N01,p(X)gives a family of compactly supported functions{1n}n∈NinN1,p(X)such that1n →1 inN1,p(X)asn→ ∞. Recall
Trace and Density Results on Regular Trees
thatXm := {x ∈ X : |x| < m}for anym ∈ N. Without loss of generality we may assume that1n is nonnegative for anyn∈ Nand that
1n−1pN1,p(X) < 1
4pμ(X1), for alln∈N.
We claim that for anyn∈N, there exists a pointxn withxn ∈ X1such that|1−1n(xn)|<
1/4. If not, then we have|1−1n(x)| ≥ 14 for anyx ∈X1. Hence we obtain that 1−1npN1,p(X) ≥ 1−1npN1,p(X1) ≥ 1
4pμ(X1),
which is a contradiction. By the triangle inequality, we have 1−1n(xn)≤ |1−1n(xn)| < 14, and hence1n(xn) > 34.
Next, we claim that we may assume1n(x) > 1/2 for allx ∈ Xn by selecting a subse-quence of{1n}n∈Nif necessary. Assume that this claim is not true. Then there existsN ∈ N such that for any n ∈ N, there exists a point yn ∈ XN with1n(yn) ≤ 1/2. Hence for any n∈ N, we have found two pointsxn ∈X1andyn ∈ XN such that|1n(xn)−1n(yn)| ≥1/4.
Letγ = [xn, yn]be the geodesic connectingxn andyn. Then
γ
|∇(1n)|ds ≥1/4 for any n∈ N.
By an argument similar to that for the estimate Eq.2.1and Eq.2.2, we have that there exists a constantC(N, p, λ, μ) >0 such that
X
|∇(1−1n)|pdμ=
X
|∇(1n)|pdμ≥C(N, p, λ, μ) >0 for any n∈ N, which is a contradiction to1n →1inN1,p(X).
Thus, from the arguments above, we may assume that there exists a family of compactly supported functions{1n}n∈N inN1,p(X)such that
1n →1inN1,p(X)asn→ ∞, 1n(x)≥ 12 for anyx ∈Xn.
We define1¯n :=min{2·1n,1}for alln∈ N. Then the family(1¯n)n∈N satisfies
⎧⎪
⎨
⎪⎩
1¯n →1inN1,p(X)asn→ ∞, 1¯n ≡1 inXn,
1¯nis a function with compact support.
(4.1)
Given a function uin N1,p(X), let us show thatun1¯n → uin N1,p(X) whereun(x)is a truncation ofuwith respect toan := 1¯n−1−1/2N1,p(X), namely
un(x)= u
|u|an if|u| ≥an u if|u| ≤an .
From the basic properties of truncation (see for instance [11, Section 7.1]), we have that
⎧⎪
⎨
⎪⎩
un →uinN1,p(X)asn→ ∞,
|un(x)| ≤an,
|∇un| ≤3|∇u|.
(4.2)
We first show thatun1¯n →uinLp(X)asn→ ∞. By the triangle inequality, it follows from Eq.4.1and Eq.4.2that
un1¯n −uLp(X) ≤ un1¯n−unLp(X)+ un−uLp(X)
≤ an1¯n−1N1,p(X)+ un−uLp(X)
= 1¯n−11/2N1,p(X)+ un−uLp(X) →0 as n→ ∞.
Recall that every function inN1,p(X)is locally absolutely continuous, see Section2.2. By the product rule of locally absolutely continuous functions, we obtain that
|∇(un1¯n−u)| = |∇(un1¯n−un +un−u)|
≤ |un||∇(1¯n−1)| + |1¯n−1||∇un| + |∇(un−u)|
≤ an|∇(1¯n−1)| + |∇un|χX\Xn + |∇(un −u)|.
Hence we obtain from the triangle inequality and Eq.4.2that
∇(un1¯n−u)Lp(X) ≤ an∇(¯1n−1)Lp(X)+ ∇unLp(X\Xn) + ∇(un −u)Lp(X)
≤ 1¯n−11/2N1,p(X)+3∇uLp(X\Xn)+ ∇(un −u)Lp(X), which tends to 0 as n → ∞. Therefore, un1¯n → u in N1,p(X) as n → ∞. Since the support ofun1¯n is compact, it follows from the definition ofN01,p(X) thatu ∈ N01,p(X), and henceN01,p(X)=N1,p(X).
Notice that1 ∈ ˙N1,p(X)no matter ifμ(X) is finite or not. By slightly modifying the previous proof, we obtain the following result.
Corollary 4.2 Let1 ≤p <∞. Then the following statements are equivalent N˙01,p(X)= ˙N1,p(X) ⇐⇒ 1∈ ˙N01,p(X)
Applying Lemma 4.1, we obtain our first density result.
Proposition 4.3 Let1 ≤ p < ∞and assume thatμ(X) < ∞. Suppose additionally that Rp = ∞. Then we have that
N01,p(X)=N1,p(X).
Proof It follows from Lemma 4.1 that it suffices to construct a sequence of compactly supportedN1,p-functions which converges to1inN1,p(X).
Forp >1 and
Rp = ∞
0
λ
p
p−1(t)μ1−1p(t)K1−tpdt = ∞,
we define the family of functions {ϕn}n∈N as follows. For eachn ∈ N, let rn > n be an integer such that rn
n
λp−1p (t)μ1−p1 (t)K1−pt dt ≥2n. (4.3) We setϕn(x)=1 for allx ∈Xn,ϕn(x)=0 for allx ∈ X\Xrn and
ϕn(x)=1− |x|
n λp−1p (t)μ1−p1 (t)K1−pt dt rn
n λ
p
p−1(t)μ1−1p(t)K1−tpdt
Trace and Density Results on Regular Trees
for all x ∈ Xrn \Xn. Since λp/μ ∈ L1/(ploc −1)([0,∞)) and λ,1/μ > 0, then ϕn is well-defined. It is easy to check thatϕn is compactly supported.
By the construction ofϕn, an easy computation shows that
∇(ϕn(x)−1)=0 (4.4)
Thanks to Eq.4.4and Eq.4.5, we obtain the estimate Sincep >1 and Eq.4.3holds, we obtain that
rn is a nonincreasing sequence of subset of[0,∞)and that we have
|Ek|>0 for any k ∈ N.
We define a sequence{ϕk}of functions by setting ϕk(x)=1− 1
It follows directly from the definition ofϕk that eachϕk has compact support and that
|ϕk −1| ≤2χX\Xk, |∇(ϕk)(x)−1)| ≤ χEkn(x) λ(x)|Ekn|. Hence, thanks toμ(X) < ∞and the definition ofEkn, we obtain that
ϕk −1N1,1(X) = ϕk −1L1(X)+ ∇(ϕk −1)L1(X)
≤ 2χX\XkL1(X)+
X
χEkn(t)
λ(t)|Ekn|dμ(t)
2μ(X\Xk)+ 1
|Ekn| kn
k
μ(t)Kt
λ(t) χEkn(t) dt
≤ 2μ(X\Xk)+ 1
2k →0 ask → ∞. Henceϕk →1inN1,1(X)ask → ∞.
By using the same construction of the sequence of compactly supportedN1,p-functions as the one in the above proof, we obtain the following corollary immediately from Corollary 4.2.
Corollary 4.4 Let1≤p <∞. AssumeRp = ∞. Then we obtain that N˙01,p(X)= ˙N1,p(X).
Proposition 4.5 Let1 ≤ p < ∞and assume thatμ(X) < ∞. Suppose additionally that Rp <∞. Then we have
N01,p(X)N1,p(X).
Proof SupposeN01,p(X)= N1,p(X). Since 1∈ N1,p(X), it follows that for everyε > 0, there exists a functionu∈N1,p(X)with compact support such that
1−uN1,p(X) < ε. (4.6)
Letξ ∈ ∂X be arbitrary, andxj := xj(ξ )be the ancestor ofξ with |xj| = j andx0 = 0.
Let 0< < 12μ1/pL1([0,1]). By repeating the argument in the beginning of Proof of Lemma 4.1 with the change that we replace μ(X1)/4p andX1 by p and [0, x1(ξ )], respectively, we obtain the existence of xξ ∈ [0, x1(ξ )] for which the function u in Eq. 4.6 satisfies
|1−u(xξ)| < 12. By the triangle inequality, we have 1− |u(xξ)| ≤ |1−u(xξ)| < 12, and hence|u(xξ)|> 12.
Notice thatuhas compact support. Then for anyξ ∈∂X, we have limn→∞u(xn(ξ ))=0 and that
1
2 < lim
n→∞|u(xξ)−u(xn(ξ ))| ≤
[0,ξ )
guds =
[0,ξ )
guλ(x)
μ(x)dμ. (4.7)
Trace and Density Results on Regular Trees
By choosing small enough, the above estimate yields a contradiction, and hence N01,1(X)=N1,1(X).
Forp >1, by Eq.4.7and the H¨older inequality, we have that 1
By choosing small enough, the above estimate gives a contradiction, and hence N01,p(X)=N1,p(X)forp >1.
Since N01,p(X) ⊂ N1,p(X)by definition, we obtainN01,p(X) N1,p(X) for allp ≥ 1.
Corollary 4.6 Letp ≥1and assume thatRp <∞. Then we have N˙01,p(X)N˙1,p(X).
Proof SupposeN˙01,p(X) = ˙N1,p(X). Since 1∈ ˙N1,p(X), it follows that for every > 0, there exists a functionu∈ ˙N1,p(X)with compact support such that
1−uN˙1,p(X) < .
Then by the definition of ourN˙1,p-norm, we have|u(0)−1|< and hence|u(0)|> 1−. Then using a similar argument to the one in the Proof of Proposition 4.5 (replaceu(xξ) withu(0)), we obtain a contradiction. The claim follows.
The above results give a full picture for the density properties for homogeneous Newto-nian spacesN˙1,p(X)and for Newtonian spacesN1,p(X)whenμ(X) <∞. When the total measure is infinite, the density results for the Newtonian spaceN1,p(X)are quite different.
Lemma 4.7 LetK = 1, i.e.,Xbe a 1-regular tree and assume thatμ(X) = ∞. Then for anyf ∈ N1,p(X), there exists a sequence of compactly supported N1,p-functions{fn}n∈N
such thatfn →f inN1,p(X).
Proof Notice that we may compose any f ∈ N1,p(X) as f = f+ − f− where f+ = f ·χ{f≥0} ≥0 andf−= −f ·χ{f≤0} ≥0. Hence we may assume thatf ≥0.
SinceK =1,∂Xcontains only one pointξ0and there is a unique geodesic ray. It follows from Proposition 3.9 that
lim inf
[0,ξ0)x→ξ0f (x)=0. (4.8) Denote byxn the vertex ofXwith|xn| =nwhenn∈N. Then it follows from Eq.4.8that
f (xn)−
[xn,ξ0)
gf ds ≤0, ∀n∈N. (4.9)
We define functionsfn by setting fn(x):=
f (x), if|x| ≤n;
max{0, f (xn)−2
[xn,x]gf ds}, if|x|> n.
Then it is easy to check that fn ∈ N1,p(X), since 0 ≤ fn ≤ f and gfn ≤ 2gf. Next, we check thatfn is compactly supported. Assume not. Sincefn is non-increasing for |x| > n by definition, we have thatfn(x) >0 for any|x|> nand hence that
xlim→ξ0fn(x)=f (xn)−2
[xn,ξ0)
gf ds ≥0.
Combining this with Eq.4.9, we conclude that
[xn,ξ0)
gf ds =0.
Thengf = 0 for|x| > nand it follows from Eq.4.8thatf andfn has to be identically 0 for|x| ≥n, which is a contradiction. Hencefn is compactly supported.
Trace and Density Results on Regular Trees
At last, we estimate theN1,p-norm offn −f. By the fact that 0 ≤ fn ≤f andgfn ≤ 2gf, we obtain the estimate
fn −fN1,p(X) = fn−fN1,p(X∩{|x|≥n})
≤ fnN1,p(X∩{|x|≥n})+ fN1,p(X∩{|x|≥n})
≤ 3fN1,p(X∩{|x|≥n}) →0 as n→0,
since f ∈ N1,p(X). Thus {fn}n∈N is a sequence of compactly supported N1,p-functions withfn →f inN1,p(X), which finishes the proof.
If∞
0 λ(t) dt = ∞, thenXis complete and unbounded with respect to distanced and it follows by using suitable cutoff functions thatN01,p(X)= N1,p(X). Our next result shows that this is also the case whenXis bounded and not complete if we assumeμ(X)= ∞. Theorem 4.8 Let1≤p <∞and assume thatμ(X)= ∞. Then we have that
N01,p(X)=N1,p(X).
Proof IfK = 1, the result follows directly from Lemma 4.7. Hence we assumeK ≥2 in the ensuing proof.
For any f ∈ N1,p(X), by the same argument as in Lemma 4.7, we may assume that f ≥0. It suffices to construct a sequence {fn}n∈N of compactly supportedN1,p-functions such thatfn →f inN1,p(X).
For each n ∈ N, we denote by {xn,j}Kj=1n the vertices of n-level, i.e., |xn,j| = nfor all j = 1,· · · , Kn. For any xn,j, we study the subtreexn,j which is a subset of Xwith root xn,j. More precisely,
xn,j := {x ∈X:xn,j < x}.
Since every vertex has exactlyK children, we may dividexn,j intoK subsets, where each subset contains a subtree whose root is a child of xn,j and an edge connecting this child withxn,j. We denote by{ix
n,j}Ki=1theseK subsets.
Fixf ∈ N1,p(X). We first study the function u := f|xn,j. If fN1,p(xn,j) > 0, we first modify the functionuto a functionvwithv(x)=v(|x|)for anyx ∈xn,j, i.e., for any x, y ∈xn,j with|x| = |y|, thenv(x)=v(y). The modification procedure is as follows:
Step 1 Sincexn,j =K
i=1xi
n,j, without loss of generality, we may assume
uN1,p(1xn,j) =min{uN1,p(xn,ji ) :i =1,2, . . . , K}. (4.10) Then we define a functionu1by identically copying the minimalN1,p-energy subtree of u(here isu|1
xn,j), to the otherk−1 subtreesix
n,j,i=2,· · ·K. More precisely, u1(x):=
u(x), ifx ∈x1
n,j; u|1xn,j(y)withy ∈x1
n,j,|y| = |x|, ifx ∈xi
n,j. It follows from Eq.4.10that
u1N1,p(xn,j) ≤ uN1,p(xn,j).
Then for any x, y ∈ xn,j ∩ {x ∈ X : n ≤ |x| ≤ n +1} with |x| = |y|, we have u1(x)= u1(y).
Step 2 Denote by{xn+1,t}Kt=1theKchildren ofxn,j. We repeat the Step 1 by replacing the functionuandxn,j withu1andxn+1,t, respectively. Here we repeat the Step 1 for allK subtreesxn+1,t,t = 1,· · · , K. Hence we obtain a functionu2 onxn,j by additionally lettingu2(x)= u1(x)ifx ∈ xn,j withn ≤ |x| ≤ n+1. Moreover, it is easy to check that
u2N1,p(xn,j) ≤ u1N1,p(xn,j) ≤ uN1,p(xn,j)
and thatu2(x)=u2(y)for anyx, y ∈xn,j ∩ {x ∈ X:n≤ |x| ≤n+2}with|x| = |y|.
Continuing this procedure, we obtain a sequence of functions{uk}k∈N. We define v = limk→∞uk. Then we know from induction that
vN1,p(xn,j) ≤ uN1,p(xn,j) = fN1,p(xn,j) (4.11) and thatv(x)=v(y)for anyx, y ∈xn,j with|x| = |y|.
The value of functionv(x)only depends on the distanced(xn,j, x). We may regardvas a function on a 1-regular tree with rootxn,j and infinite measure, sinceμ(xn,j)= ∞. Hence, from the Proof of Lemma 4.7, we are able to choose a compactly supportedN1,p-function fn,j onxn,j with
fn,j −vN1,p(xn,j) ≤3uN1,p(xn,j) =3fN1,p(xn,j). (4.12) Then it follows from Eq.4.11and Eq.4.12that
fn,j −fN1,p(xn,j) ≤ fn,j −vN1,p(xn,j)+ v−fN1,p(xn,j)
≤ fn,j −vN1,p(xn,j)+ vN1,p(xn,j)+ fN1,p(xn,j)
≤ 5fN1,p(xn,j). (4.13)
IffN1,p(xn,j) =0, thenf =0 onxn,j and we just definefxn,j = f|xn,j. At last, we define a functionfn by setting
fn(x):=
f (x), if|x| ≤n; fn,j(x), ifx ∈xn,j.
Then it is easy to check that fn ∈ N1,p(X)and thatfn is compactly supported, sincefn,j are compactly supported for anyj =1,· · · , Kn. It follows from estimate Eq.4.13that
fn −fN1,p(X) = fn−fN1,p(X∩{|x|≥n}) =
Kn
j=1
fn −fN1,p(xn,j)
=
Kn
j=1
fn,j −fN1,p(xn,j) ≤5
Kn
j=1
fN1,p(xn,j)
= 5fN1,p(X∩{|x|≥n}) →0, asn→0,
sincef ∈N1,p(X). Thus we have found a sequence{fn}n∈Nof compactly supportedN1,p -functions withfn →f inN1,p(X), which finishes the proof.
Trace and Density Results on Regular Trees