Higher order coercive inequalities on nilpotent Lie groups

Given the following Logarithmic-Sobolev inequality:

for \beta\in(0,1). Then, for all m\in\mathbb{N},

where \nabla^{\alpha}f=(X_{1}^{\alpha_{1}}X_{2}^{\alpha_{2}}...X_{n}^{\alpha_{n}}f) such that |\alpha|={\displaystyle \sum_{i=1}^{n}\alpha_{i}.}

Let p>0. Given the following Logarithmic Sobolev inequality:

for \beta\in(0,1). Then, for all m\in\mathbb{N},

where \nabla^{\alpha}f=(X_{1}^{\alpha_{1}}X_{2}^{\alpha_{2}}...X_{n}^{\alpha_{n}}f) such that |\alpha|={\displaystyle \sum_{i=1}^{n}\alpha_{i}.}

Suppose outside a ball B_R, there exists an \varepsilon\equiv \varepsilon(R)\in(0,\infty) and a constant E_{\varepsilon}\in(0,\infty) such that

If \varepsilon a\frac{n^2}{2}<1, then for every m\leq n

there exists a constant a_m\in(0,\infty)

From [BDZ], we have

Consider a differentiable function F\equiv F(N)\geq 0. Then we have

If we assume that

and

then the iteration is finite and we get

Let d\mu\equiv e^{-U}d\lambda and let

Suppose exist C\in(0,\infty) such that

and for some a\in(0,\infty)

Then we have

whence

Special case n+1=2:

Potential bound: Let U\equiv U(N). Then we have

Based on homogeneity considerations, one should be able to claim that there exists a small constant \varepsilon\in(0,\infty) and a constant E_{\varepsilon}\in(0,\infty) such that

Then we get

and hence

Thus using our inductive assumption (5), we arrive at

Now for the general case, using (C\star), we have

and hence

Suppose outside a ball B_R there exists an \varepsilon\equiv \varepsilon(R)\in(0,\infty) and a constant E_{\varepsilon}\in(0,\infty) such that

If \varepsilon a\frac{n^2}{2}<1, then

there exists a constant a_n\in(0,\infty)

Suppose, for d\mu \equiv e^{-U}d\lambda, the following quadratic form bounds are satisfied

with some constants b_m,c_m\in(0,\infty) independent of f, then the following Logarithmic Sobolev inequality holds

with some constants \tilde C,\tilde D\in(0,\infty).

Let \mathbb{H}^n be Heisenberg group and let {N=\left(|x|^{4}+16z^{2}\right)^{\frac{1}{4}}} be the corresponding Kaplan norm. Then, the interaction U\equiv U(N) given as smooth function of the Kaplan norm N does not satisfy Adams’ regularity condition.

If the nilpotent Lie algebra is of second order, then

and hence we have the following relation of seminorms

If the nilpotent Lie algebra is of second order, then

Let \mathfrak{G} be a nilpotent Lie algebra is of second order. Suppose the interaction U satisfies the following quadratic form bound

with some constant C'\in(0,\infty) independent of f. Then, the following global regularity bound holds

with a constant D\in(0,\infty) independent of f.

Consider the Heisenberg group \mathbb{H}_1 with generating fields

and the corresponding Kaplan norm

With \kappa\in(0,\infty), let

From [HZ] it satisfies the following U-Bound

We have shown in Section ? that this interaction is not Adams’ regular.
For quadratic form estimate with potential \mathcal{V}_Z, with Z \equiv [X,Y]= \partial_z\equiv Z, we note that

Thus

Using z^2=\frac1{16}\left(N^4-|\mathbf{x}|^4\right), we can represent the above as follows

For q\geq 2, in [BDZ] we got the following

In case when U(N)\equiv N^\kappa, we have \frac{U'(N)}{N^2}=\kappa N^{\kappa-3}, which is not sufficient for us. Fortunately, using a technique employing Hardy’s inequality, [YaoThPhD], one can improve that to obtain

In particular, in the case when U(N)\equiv N^\kappa, we have \frac{U'(N)}{N}=\kappa N^{\kappa-2}.

Thus, iterating the quadratic form bound, we arrive at the following desired regularity estimate

Let \mathfrak{G}=\mathbb{H}_1. Suppose the interaction U=N^\kappa, with \kappa>2. Then the following global regularity bound is true.

For any s\in\mathbb{N}, there exists \tilde C\in(0,\infty) such that

For the third-order case, we have

Since the fourth-order fields are zero and the third-order fields are in the center, in the third-order case this simplifies as follows

In terms of graphs, we admit vertices of at most third order.

For \varepsilon\in(0,1), let

and let

In particular, for \varphi=0, we have

with (1+\varepsilon)/(1-\varepsilon)\geq 2 for \varepsilon \geq 1/3, while for \varphi=\pi, we have

The gradient of U in polar coordinates is as follows

Hence,

In particular, for \varphi=\pi, we have

For the Laplacian, we have that

Hence,

Thus,

It is interesting to see the values for particular \phi. In particular, at \varphi=\pi, we have

Thus, for \varphi=\pi, we have

which is strictly positive for large r, but does not grow to infinity with r\to\infty.
On the other hand, for \varphi=0, we have

which is positive for \epsilon<1 and r large enough,

and in fact growing to infinity as r\to\infty.
Finally, for \varphi=\pm \pi/2, we have

which is positive for \epsilon<1 and r large enough, and diverges to infinity with r\to \infty.

In general, we have

which is strictly positive for sufficiently large r, but not necessarily diverging to infinity as r\to\infty.
Using a partition of unity \chi_1, \chi_2\equiv 1-\chi_1, with twice differentiable and bounded derivatives of the function \chi_1, equalling to one outside a ball of radius R\in(1,\infty) and zero inside a ball of radius R/2, we can decompose the interaction as follows

With this decomposition, we have U_1, an Adams’ regular potential, for which we can choose R\in(0,\infty) so that

for some constant m\in(0,\infty). Then, for the measure d\mu_1\equiv \frac{1}{Z_1}e^{-U_1}d\lambda with Z_1\equiv \int e^{-U_1}d\lambda, we have the following bound, [HZ],

Given the lower bound \mathcal{V}_1\geq m>0, this immediately implies Poincaré inequality,

Since U_2 is bounded, we can now apply the bounded perturbation lemma to arrive at

There exists a constant \tilde m\in(0,\infty) such that

for any function f for which the right-hand side is well-defined.

For \varepsilon\in(0,1) and \alpha\in(0,\infty), let

Let

We have

and hence,

For the Laplacian of the interaction, we have

Suppose that \nabla U=0, which is the case when

and for large r, that allows for (infinitely many ) solutions close to \omega r^{\kappa}\sim 2n\pi with \cos(\omega r^{\kappa})>0, as well as for solutions close to \omega r^{\kappa}\sim (2n+1)\pi with \cos(\omega r^{\kappa})<0. Then, we have

which can be positive or negative depending on the sign of \cos(\omega r^\kappa), with the modulus growing to infinity.

If \kappa>0, there does not exist a constant C\in(0,\infty) such that the following Adams’ regularity condition holds

Let d\mu\equiv \frac1{Z}e^{-\mathcal{U}}d\lambda where

and Z is a normalisation constant.

Then we have

and

Hence

and

In general, the last formula cannot be dominated by the square of the gradient for small |\mathbf{x}|^2, and hence, Adams’ regularity condition does not hold.
Let

and we have

On the other hand, for Hardy’s inequality, we have

and hence

Thus, in our example, we obtain

Combining (I) and (II), with some A^2 \in(0,\infty) to be chosen later, we obtain

After rearrangement, this yields

If n is large enough, so that

then there exists a constant B\in(0,\infty) such that we have

and hence

At this point, we note that the following inequality holds

This, together with (IV), yields

Since there exists a constant D\in(0,\infty) such that

the following result is true.

Let d\mu\equiv \frac1{Z}e^{-\mathcal{U}}d\lambda where

and Z is a normalisation constant.
Suppose Hardy’s inequality for the Haar measure and subgradient holds with the constant \frac{C}{(n-2)^2} and for a given constant A\in(0,\infty)

then there exists a constant D\in(0,\infty) such that

We remark that we have encountered the functions W_\pm before. Namely, in Example 20 with n=1 we have

Hence

and hence

In the Heisenberg group \mathbb{H}^n, the Kaplan norm N_{Kaplan} satisfies