On the dimension drop for harmonic measure on uniformly non-flat Ahlfors-David regular boundaries. Aritro Pathak Abstract We extend earlier results of Azzam on the dimension drop of the harmonic measure for a domain Ω ⊂ ℝ n \Omega\subset\mathbb{R}^{n} with n ≥ 3 n\geq 3 , with Ahlfors-David regular boundary ∂ Ω \partial\Omega of dimension s s with n − 1 − δ 0 ≤ s ≤ n − 1 n-1-\delta_{0}\leq s\leq n-1 , that is uniformly non flat. Here δ 0 \delta_{0} is a small positive constant dependent on the parameters of the problem. Our novel construction relies on elementary geometric and potential theoretic considerations. We avoid the use of Riesz transforms and compactness arguments, and also give quantitative bounds on the δ 0 \delta_{0} parameter. 1 Introduction. The study of the harmonic measure in relation to the geometric properties of the boundary of a domain \Omega\subset\mathbb{R}^{n} , is a classical subject in analysis. The harmonic measure arises in an important way in the Dirichlet problem. Previous studies, going back to the work in [ Rie23 ] , study the mutual absolute continuity of the harmonic measure with respect to the surface measure in various settings. More recently, there has been a focus on the harmonic measure for boundaries \partial\Omega which have codimension larger than 1 , through the conjectures in [ Vol22 ] , and the recent work of [ DJJ23 , Tol24 ] for the special situation of boundaries that are a subset of a codimension one hyperplane or a manifold. In [ Tol24 ] , it is shown that in \mathbb{R}^{n} , for such a boundary lying on a codimension one hyperplane, with Hausdorff dimension greater than or equal to n-3/2-\epsilon for some small enough \epsilon , there is a dimension drop of the harmonic measure. In [ DJJ23 ] , it is shown that in \mathbb{R}^{2} , a boundary with a small enough Hausdorff dimension (less than 0.2497 ) in \mathbb{R}^{3} , and lying on a line, the harmonic measure is actually absolutely continuous with the surface meausure. Prior work by [ Car85 , JW88 , Vol93 , Vol92 , UZ02 , Bat06 , Bat00 , Bat96 ] focused on the case of dynamically defined boundaries which satisfy some kind of totally disconnected properties, for which the dimension drop phenomenon was established. This enabled the use of some ergodic theory. Extensions of this work to the case of boundaries that are not dynamically defined or totally disconnected, but satisfying a uniformly non flatness condition was undertaken in [ Az20 ] , for the case of boundaries with codimension less than 1 . Here we work on extending these results. Here we give a new argument, that ostensibly avoids the use of Riesz transforms, and the techniques stemming from the work of [ AHM+16 , Tol15 ] . We employ a double integral averaging argument with a Riesz type kernel, in Lemma 10 , and use in Lemma 12 a change of pole argument for the harmonic measure, to localize the problem with poles of harmonic measures close to a specific surface cube under question, by closely following the corresponding argument of Lemma 4.3 of [ Az20 ] . We also give explicit bounds on the dimension of the boundary, in terms of the parameters of the problem, for which the dimension drop of the harmonic measure is admissible. In future work, we plan to explore applications of the techniques developed here to related problems concerning harmonic measure on rough domains. In particular, these methods open up alternative approaches to questions of uniform rectifiability of the boundary and to the absolute continuity of harmonic measure with such boundaries, which we intend to investigate further. For prior work on related questions on the harmonic measure, and the dimension drop phenomenon, as well as related questions in uniform rectifiability of the boundary and absolute continuity of the harmonic measure, we refer the reader to, [ Wol95 , Tol15 , Pop98 , MV86 , Mey09 , Mat95 , Mak85 , LVV05 , KT99 , HMM+17 , HKM06 , GM08 , Bou87 , BE17 , AAM16 , AHM+16 , AMT17 , AM15 ] . In this paper we, prove the following result. Theorem 1 . Given any integer n\geq 3 , C_{1}>1 , and 0<\beta<1 , there is a constant \kappa\in(0,1) , so that the following holds. Define, N_{\beta}=\lceil\frac{1}{\beta}\rceil and \beta_{1}:=\frac{1}{4N_{\beta}+2} . Suppose \delta_{0}=\delta_{0}(\beta,C_{1}):=\beta_{1}^{\frac{4n\log C_{1}}{\beta_{1}^{n}}}. Suppose \Omega\subseteq\mathbb{R}^{n} is a connected domain. Further, for any point x\in\mathbb{R}^{n} and any r<\text{diam}(\Omega) , we have, b\beta_{\omega_{\Omega}}(x,r):=\inf_{V}\left[\sup_{y\in B(x,r)\cap\partial\Omega}\frac{\operatorname{dist}(y,V)}{r}\;+\;\sup_{y\in V\cap B(x,r)}\frac{\operatorname{dist}(y,\partial\Omega)}{r}\right]\geq\beta>0, (1) where the infimum is over all (n-1) -dimensional planes V\subseteq\mathbb{R}^{n} . Further, assume that the boundary \partial\Omega is Ahlfors-David s -regular, ( s,C_{1} AD-regular) meaning if \sigma=\mathcal{H}^{s}\!\!\mid_{\partial\Omega} , then C_{1}^{-1}r^{s}\leq\sigma(B(x,r))\leq C_{1}r^{s}\quad\text{for all }x\in\partial\Omega,\;0n-1 from the earlier results. Remark 4 . We have not carried out the calculations in the plane \mathbb{R}^{2} , where with some modifications to the argument and of the fundamental solution which becomes logarithmic, one would expect a result of a similar structure to be true. 2 Basic estimates and overview of the proof. The Green function is constructed by setting G(X,Y):=-\mathcal{E}(X,Y)\;+\;\int_{\partial\Omega}\mathcal{E}(X,z)\,d\omega^{Y}(z), (3.4) where \mathcal{E}(X):=c_{n}|X|^{1-n} is the usual fundamental solution for the Laplacian in \mathbb{R}^{n+1} . Note that we have chosen to take a negative sign for the contribution of the pole, and a positive sign for the contribution of the potential due to the harmonic measure. Also, G(X,Y) is positive everywhere inside the domain, by construction, and it vanishes at the boundary. See for example, section 2 of [ HMM+17 ] . The basic idea behind our proof is to first isolate some specific boundary ball B and consider the harmonic measure due to some pole located within this boundary ball and with the pole being a controlled distance away from the boundary. This can be phrased in terms of the corkscrew point relative to this boundary ball. We isolate ‘corner’ regions, which are sub-balls of B centered on a point on the boundary that is at a minimum distance from the given corkscrew point. If the harmonic measure in such a sub-ball centered on such a ’corner’ point were to be approximately constant, one shows that at the corner point, there is a non zero and non uniform gradient of the potential due to this harmonic measure in this sub-ball. Now, the entire surface must remain an equipotential, precisely at the value of 0 . This non uniform gradient due to the local harmonic measure has to be compensated by the contribution to the potential, coming from the harmonic measure located further away, along with the contribution due to the pole. In effect, one shows that the gradient due the harmonic measure further away, and the second order contribution from the harmonic measure located nearby and further away, both can be controlled in a manner that this compensation to the local non-uniform contribution cannot happen exactly. This is a contradiction, and we can thus quantify that the harmonic measure cannot be nearly constant, in subballs centered on this corner point. Given any arbitrarily large M>1 , we find some l(M) dependent on M and a dyadic cube C(M) of length l(M) , by a careful repetition of this argument, so that the average Poisson kernel over the cube C(M) is at least a factor M bigger than the average Poisson kernel over B : \displaystyle k^{p_{B}}(C(M))\geq Mk^{p_{B}}(B). (3) Next, by a Markov chain argument effectively outlined in Lemma 4.3 of [ Az20 ] , we compare the harmonic measure due the pole p_{B}\in B , that is a bounded distance away from the boundary relative to the radius of B , to the harmonic measure due to the pole at the a-priori chosen point p which is located at an arbitrary point in the domain. This leads us to a lower bound of the same form as in Eq. 3 , but with a different constant M_{2} . This happens in every dyadic cube in \partial\Omega . This eventually leads to a dimension drop of the harmonic measure. 3 Preliminaries The following lemma is originally due to Bourgain( [ Bou87 , Lemma 1] ). Lemma 2 . If \Omega\subseteq\mathbb{R}^{n} be a domain that has Ahlfors-David s -regular boundary, with constant c_{1} and s>n-2 , then there is b\in(0,1) so that \omega_{\Omega}^{x}(B(\xi,r))\geq c_{0}\;\;\mbox{ for }\xi\in\partial\Omega,\;\;00 so that for all 00 , depending only on d , n , and the regularity constant for E , such that we can associate to each cube Q\in\Delta a “center” c(Q)\in Q which satisfies \operatorname{dist}(c(Q),\,E\setminus Q)\;\geq\;C_{1}^{-1}\,\operatorname{diam}Q. (9) Thus we have the surface ball B_{Q}:=\Delta(c(Q),C_{1}^{-1}\text{diam}Q)\subset Q . Henceforth, it should be understood from the context that we mean the harmonic measure with respect to the domain \Omega and we simply write \omega in place of \omega_{\Omega} . In certain situations, we will use the harmonic measure in certain subdomains, by eventually adopting the argument in Lemma 4.3 of [ Az20 ] . Note that the condition of Eq. 1 is taken to hold over any ball centered over points on the boundary. One easily sees that this is equivalent to having the condition hold over cubes instead of balls. Assume Eq. 1 . In this case, given any arbitrary ball B(x,r) with x\in\partial\Omega , consider the cube C(x,r) of minimal volume that contains B(x,r) . Here, C(x,r) is the Euclidean cube given by \{(y_{1},\dots,y_{n}):x_{i}-r\leq y_{i}\leq x_{i}+r,1\leq i\leq n\} , where x=\{x_{1},\dots,x_{n}\} . Then obviously we have, \displaystyle\sup_{y\in C(x,r)\cap\partial\Omega}\frac{\operatorname{dist}(y,V)}{r}\geq\sup_{y\in B(x,r)\cap\partial\Omega}\frac{\operatorname{dist}(y,V)}{r},\sup_{y\in C(x,r)\cap V}\frac{\operatorname{dist}(y,\partial\Omega)}{r}\geq\sup_{y\in B(x,r)\cap V}\frac{\operatorname{dist}(y,\partial\Omega)}{r} (10) Conversely, once we have the condition of Eq. 1 for cubes in place of balls, by considering the balls of minimal volume containing any given cube C(x,\mathscr{r}) , we get the converse statement. Here, C(x,\mathscr{r}) is the cube with center x and diameter r . By a similar argument, the condition of Eq. 2’ can also be written in terms of cubes in place of balls. Without loss of generality, we keep the same constants for the case of the cubes, in Eq. 2’ . Thus, henceforth we use the following two conditions; for any point x\in\mathbb{R}^{n} and any r<\text{diam}(\Omega) , we have, b\beta_{\omega_{\Omega}}(x,r):=\inf_{V}\left[\sup_{y\in C(x,r)\cap\partial\Omega}\frac{\operatorname{dist}(y,V)}{r}\;+\;\sup_{y\in V\cap C(x,r)}\frac{\operatorname{dist}(y,\partial\Omega)}{r}\right]\geq\beta>0, (1’) where the infimum is over all (n-1) -dimensional planes V\subseteq\mathbb{R}^{n} . Further, assume that the boundary \partial\Omega is Ahlfors-David s -regular, ( s,C_{1} AD-regular) meaning if \sigma=\mathcal{H}^{s}\!\!\mid_{\partial\Omega} , then C_{1}^{-1}r^{s}\leq\sigma(C(x,r))\leq C_{1}r^{s}\quad\text{for all }x\in\partial\Omega,\;00 for any i\geq 1 , denotes a constant whose value may change from line to line. When necessary, we indicate the dependence of C on ambient parameters. For any Q\in\Delta_{j} , we denote by l(Q)=2^{j} , with the value of j from Theorem 5 , and we denote by r(B) the radius of a ball B(x,r) of radius r centered at x . We denote by \frac{1}{C}B the ball withthe same radius as B but with 1/C times the radius of B . By S(x,r) we denote the sphere of radius r centered at x . We borrow the notation from [ Az20 ] , for a Radon measure \mu , s\geq 0 , and a ball B , we define \displaystyle\Theta_{\mu}^{s}(B)=\frac{\mu(B)}{r(B)^{s}}. (11) We also use the following notation to define the density of the harmonic measure for any one of the cubes of Theorem 5 , as, \displaystyle k^{t}(Q):=\frac{\omega^{t}_{\Omega}(Q)}{\sigma(Q)}. (12) When necessarily, we also use a subscript, for the harmonic measure, to denote the domain in which we are working, and we also use a superscript when necessary to denote the pole of the harmonic measure. This should be understood from context. When no subscript is used, it is understood that the domain in consideration is \Omega . We also distinguish between Euclidean cubes with faces parallel to the coordinate planes, such as \{(x_{1},\dots,x_{n}):b_{i}\leq x_{i}\leq a_{i},\text{for all}\ 1\leq i\leq n.\} , and the dyadic cubes of David and Semmes [ DS1 , DS2 ] mentioned in Theorem 4. As mentioned earlier, C(x,r) is the Euclidean cube given by C(x,r):=\{(y_{1},\dots,y_{n}):x_{i}-r\leq y_{i}\leq x_{i}+r,1\leq i\leq n\} , where x=\{x_{1},\dots,x_{n}\} . Depending on the context, we will also consider (n-1) -dimensional cubes contained it hyperplanes. When 2^{j}\leq C_{2}<2^{j+1} for some j\in\mathbb{Z} , we write C_{1}\approx C_{2} with C_{1}=2^{j} . Throughout this paper, by a covering of any set E , we mean a finite set of elements indexed by J , \{E_{j}\}|_{j\geq J} so that we have the disjoint union E=\sqcup_{j\geq 1}E_{j} (i.e. for any two E_{j},E_{k} with j\neq k , j,k\in J , we have E_{j}\cap E_{k}=\phi ). Finally, we denote the null set by \phi . 5 Proof of the theorem. Without loss of generality, in this paper we consider the pole p of the harmonic measure to be located as some corkscrew point A(x,r_{0}) . All the balls and dyadic cubes under subsequent consideration have diameters less than this fixed r_{0} . For the following argument, for any point p_{1}\in\Omega , we consider a point q_{1}=\widehat{p_{1}} on the boundary \partial\Omega that is closest to p_{1} , and |p_{1}-q_{1}|=r_{1} We consider this ’corner’ point q_{1} , and a ball of radius t_{1} sufficiently small in comparison to r_{1} so that for a fixed large positive integer w , we have, \displaystyle\frac{t_{1}}{r_{1}}=c(\beta_{1})^{w(1+s)}0 , so that the following holds. Let n-1-\deltaM\Theta(U), (41) then the lemma is proved by an elementary covering of B(q_{1},r) by a set of dyadic cubes at the length scale j with 2^{j}\leq r\leq 2^{j+1} , since we get that for M large enough dependent on C,c_{0} , one of these dyadic cubes must satisfy the condition of Eq. 35 in which case the lemma is proved. More precisely, consider the covering of C(q_{1},r) by a set of disjoint dyadic cubes \{Q_{i}:l(Q)\approx r\} . By construction, the total diameter of the union of these disjoint dyadic cubes is bounded by a uniform multiple of r . If we have for each i , that, \displaystyle\Theta(Q_{i})\theta_{1}>\theta_{0} , and \displaystyle\theta_{0}=\beta_{1}^{2w}. (47) We consider the plane passing through q_{1} and perpendicular to the gradient due to the harmonic measure in (C(q_{1},t_{1})\setminus C(q_{1},\kappa^{N}t_{1}))\cap H^{-}_{1}\cap\partial\Omega which we write as \nabla A_{N} . \displaystyle\nabla A_{N}=\sum_{m=0}^{N-1}\nabla A_{m}. (48) Call this plane \Sigma_{2} , and denote by \hat{u}_{2} the unit normal to it (we choose the direction \hat{u}_{2} so that \hat{u}_{2}\cdot\hat{v}>0 ). By employing the argument of Lemma 8 exactly, we can ensure that the point q_{2} is at a distance at least \beta_{1}^{w}\gamma\kappa^{N}t_{1} from \Sigma_{2} , inside the cube C(q_{1},\gamma\kappa^{N}t_{1}) . Next, consider the gradient at q_{1} due to the harmonic measure in \Omega\setminus B(q_{1},t_{1}) , and the plane passing through q_{1} perpendicular to this gradient, which we call \Sigma_{3} . By repeating the argument of Lemma 8 now with respect to the plane \Sigma_{3} , we conclude that q_{2} is a distance at least \theta_{0}\gamma\kappa^{N}t_{1}=\bigl(\beta_{1}^{2w}\gamma\kappa^{N}t_{1}\bigr) from \Sigma_{3} , inside the cube C(q_{1},\gamma\kappa^{N}t_{1}) . Now, the difference in the potential values between q_{1} and q_{2} due to the harmonic measure in C(q_{1},t_{1})\setminus C(q_{1},\kappa^{N}t_{1}) is given by using the Taylor theorem up to the second order; as in Eq. 8 . Note that due to the above considerations, the first term on the left in ( 8 ) is at least \beta_{1}^{3w}\,\gamma\,k^{N}t_{1}\,C_{3}\,M^{-1}\,N\,\Theta(U). The second term on the right in ( 8 ) is bounded from above by \displaystyle M\gamma^{2}\kappa^{2N}t_{1}^{2}\,\Theta(U)\frac{1}{\kappa^{N}t_{1}}\;\lesssim\;M\gamma^{2}\kappa^{N}t_{1}\,\Theta(U). (49) This follows by truncating the sum in Eq. 46 to only take into account the harmonic measure in the domain C(q_{1},t_{1})\setminus C(q_{1},\kappa^{N}t_{1}) . Now, consider the difference in potential between the points q_{1} and q_{2} due to the harmonic measure in the domain \Omega\setminus C(q_{1},t_{1}) , i.e. the contribution in I_{\text{far}} in the expression in Eq. 20 . We call the magnitude of the gradient at q_{1} , due to the harmonic measure in \Omega\setminus C(q_{1},k^{N}t_{1}) , by T_{N} . Note that the second derivative contribution from I_{\text{far}} is bounded crudely by the same expression as in Eq. 46 . We choose N large enough so that, \displaystyle\beta_{1}^{3w}\,\gamma\,k^{N}t_{1}\,C_{3}\,M^{-1}\,N\,\Theta(U)\;\gg\;M\gamma^{2}k^{N}t_{1}\,\Theta(U). (50) Thus, we get \beta_{1}^{3w}C_{3}N\;\gg\;M^{2}\gamma\,C_{5}. It is enough to choose \gamma=\frac{1}{2} (5) (5) (5) Note that in the case of s\theta_{0} , which was defined in Eq. 47 . Now we consider the number of iterations till we reach the cube C(q_{1},k^{N+N_{2}}t_{1}) . Here we have chosen N_{2} many further steps so that we have that the lower bound on the gradient term from I_{\text{mid}} when considering the cube C(q_{1},\kappa^{N+N_{2}}t_{1}) , is greater than the right hand side of Eq. 61 . \displaystyle c\,\beta_{1}^{3w}\,M^{-1}\Theta(U)\,(N+N_{2})\;>\;\left(\frac{1}{\beta_{1}}\right)^{2w}MN\Theta(U)C_{2}. (62) So, (N+N_{2})>\;C\left(\frac{1}{\beta_{1}}\right)^{5w}M^{2}N. So, N_{2}>\;C\left(\frac{1}{\beta_{1}}\right)^{5w}\cdot M^{2}N-N\;\simeq\;C\left(\frac{1}{\beta_{1}}\right)^{5w}M^{2}N. Thus, repeating the entire calculation within the cube C(q_{1},k^{N+N_{2}}t_{1}) , we get a contradiction since \displaystyle T_{N+N_{2}}\theta\gamma\kappa^{N+N_{2}}t_{1}=T_{N}\theta\gamma\kappa^{N+N_{2}}t_{1}>\;c\,\beta_{1}^{3\omega}\,\gamma\kappa^{N+N_{2}}t_{1}\,M^{-1}(N+N_{2})\Theta(U). (63) So, this means that, \displaystyle T_{N}\geq\frac{1}{\theta}c\beta_{1}^{3w}M^{-1}\Theta(U)(N+N_{2})\geq c\beta_{1}^{3w}M^{-1}\Theta(U)(N+N_{2}), (64) which combined with Eqs. 61 and 62 , gives us a contradiction. Here again, within the Euclidean cube C(q_{1},\kappa^{N+N_{2}}t_{1}) we need to repeat the argument of Lemma 10 so that we consider all the dyadic cubes of length scale \kappa^{N+N_{2}+N_{1}}t_{1} , to ensure that the potential contribution I_{\text{1,near}} within the cube C(q_{1},\kappa^{N+N_{2}}t_{1}) as well as the second order contribution in Eq. 8 from I_{\text{mid}},I_{\text{far}} , are negligible in comparison to the gradient contribution from I_{1,\text{mid}} which we now define to be the contribution from the region C(q_{1},t_{1})\setminus C(q_{1},\kappa^{N+N_{2}}t_{1}) . Thus, the argument gives that there exists some dyadic cube contained in Q , of length scale at least 2^{-j_{0}}l(Q) for some j_{0}\geq 1 where l(U) is the length scale of U so that its density is either greater than M\Theta(U) , (7) (7) (7) Recall that we have considered q_{1} as the corkscrew point relative to the ball \tfrac{1}{3}B_{U} . or smaller than M^{-1}\Theta(U) . In the former case, for M sufficiently large, we also get that the density is greater than MC\Theta(Q) and the argument is finished, leading to the use of Lemma 12 . In the latter case, we need the following lemma. Lemma 11 . We can find some M_{1}>1 depending on M and the ambient parameters so that, if there is a cube A_{1}\subset U with \frac{\omega(A_{1})}{\sigma(A_{1})}M_{1}\Theta(U). Moreover, it suffices to take \displaystyle M_{1}=1+\frac{1}{2}\Big(\frac{1-M^{-1}\eta_{2}}{1-\eta_{1}}-1\Big). Proof. Consider that there is some M_{1}>1 so that, \frac{\omega(A_{1})}{\sigma(A_{1})}1. (65) \displaystyle\Leftrightarrow 1-M^{-1}\eta_{2}>1-\eta_{1} (66) \displaystyle\Leftrightarrow\eta_{1}>M^{-1}\eta_{2} (67) \displaystyle\Leftrightarrow M>\frac{\eta_{2}}{\eta_{1}}>1. (68) then we can choose some fixed 11 constructed above, which means \Theta D_{1})>M_{1}\Theta(U) . We then consider all the dyadic subcubes in D_{1} and the corkscrew point relative to the ball \tfrac{1}{3}B_{D_{1}} , and repeat the argument of Theorem 9 . Note that in the process, considering the second order contribution from I_{mid},I_{1,mid},I_{far} , we might have cubes whose density increases by a factor of at least M\Theta(D_{1})>MM_{1}\Theta(U) , but again with a basic covering argument, and noting that D_{1}\subset U and \omega^{x}(U)\geq c_{0} with x_{0}\in\tfrac{1}{3}B_{U}\setminus\partial\Omega and a crude bound of 1 for the harmonic measure of any Euclidean cube, we get that such cubes necessarily lie within Q(\supset U) . In the process, we find either a dyadic cube of length scale at least 2^{-2j_{0}}l(Q) and whose density increases by a factor of M\Theta(D_{1}) in which case we are done, or we find by use of Lemma 13 restricted to D_{1} , a dyadic subcube D_{2} where the density increment is at least M_{1}\Theta(D_{1})\geq M_{1}^{2}\Theta(U) . We next repeat the argument within the Euclidean cube D_{2} , and continue the process, and upon running this argument at most O(k) many times where M_{1}^{k}=M , we find a subcube Q_{1}\subset Q where the average density has increased by a factor of M over the average density in U , and whose length scale is at least 2^{-Ckj_{0}}l(Q)=2^{-Cj_{0}\log\frac{M}{M_{1}}}l(Q) , and we conclude the proof of Theorem 9 by replacing \Theta(U) with \Theta(Q) , up to a multiplicative factor. This concludes the proof of Theorem 9 for the case of s=n-1 . (2)The case n-1-\delta_{0}\leq sw\beta_{1}^{3nw}\log\frac{1}{\beta_{1}}+\log 2+2\log C. (91) For all \beta_{1}<1 , since we have \displaystyle-\log(1-\beta_{1}^{n-1}) \displaystyle\geq\beta_{1}^{n-1}, (92) and also \displaystyle w\beta_{1}^{3nw} \displaystyle=\frac{\log C}{\beta_{1}^{n}}\,\beta_{1}^{3n\log C/\beta_{1}^{n}} \displaystyle=(\log C)\,\beta_{1}^{\frac{4n\log C}{\beta_{1}^{n}}-n}\ll 1, (93) we obtain that the above inequality holds. Furthermore, we have \displaystyle\delta_{0} \displaystyle=\beta_{1}^{3nw} \displaystyle=\beta_{1}^{\frac{3n\log C}{\beta_{1}^{n}}}\ll\beta_{1}^{\frac{(n+2-\delta_{1})\log C}{\beta_{1}^{n}}}, (94) since \beta_{1}<1 and 3n>(n+2-\delta_{0}) when n\geq 3 . Thus, the dimension drop argument works when \displaystyle\delta_{0}\approx\beta_{1}^{\frac{3n\log C}{\beta_{1}^{n}}}. (95) It remains to show that the contribution due to the second order term in the Taylor expansion, from \Omega\setminus C(q_{1},t_{1}) , and from C(q_{1},t_{1})\setminus C(q_{1},\kappa^{N}t_{1}) is negligible in comparison to the gradient contribution from I_{\text{mid}} . We see that either of these contributions is bounded from above by \displaystyle\sum_{j=0}^{\infty}\frac{MC_{4}\Theta(U)}{(2^{j}k^{N}t_{1})^{1+\delta}} \displaystyle\leq\frac{MC_{4}\Theta(U)}{(k^{N}t_{1})^{1+\delta}}. (96) Thus the contribution of the second order term is bounded from above by \displaystyle\gamma^{2}(k^{N}t_{1})^{2}\cdot\frac{MC_{4}\Theta(U)}{(k^{N}t_{1})^{1+\delta}} \displaystyle=\gamma^{2}MC_{4}\Theta(U)(k^{N}t_{1})^{1-\delta}. (97) We recall that the gradient contribution from I_{mid} is bounded from below by, \displaystyle\frac{M^{-1}\,\Theta(U)\,\beta_{1}^{w(n+2-\delta)}}{\kappa^{\delta-\delta_{2}}\,\log\!\left(\tfrac{1}{\kappa}\right)\,\delta}\,(\kappa^{N}t_{1})^{1-\delta_{1}}. (98) Thus, if \displaystyle\frac{M^{-1}\gamma\,\Theta(U)\,\beta_{1}^{\omega(n+2-\delta)}}{\kappa^{\delta}\,\log\!\left(\tfrac{1}{\kappa}\right)\,\delta}\gg\gamma^{2}MC_{4}\Theta(U) (99) then we are done. Thus to satisfy Eq. 116 , it is enough to take, \displaystyle\frac{M^{-1}}{\kappa^{\delta}\,\log\!\left(\tfrac{1}{\kappa}\right)}\gg\gamma MC_{4}, (9) since we have already also considered earlier the condition, \displaystyle\frac{\beta_{1}^{\omega(n+2-\delta)}}{\delta}\gg 1. (100) Thus, it is enough to consider a value of \gamma small enough so that \displaystyle\gamma\ll\frac{M^{-2}}{C_{4}\,\kappa^{\delta}\,\log\!\left(\tfrac{1}{\kappa}\right)}. (101) Recall that \kappa is explicitly taken as \displaystyle\kappa=C_{1}^{-2/(n-1-\delta)}. (102) Note that since the second order contribution from I_{\text{mid}} is negligible in comparison to the gradient contribution from I_{mid} , this enables us, as in the previous section, to a-priori consider the points q_{1},q_{2} to belong to the centers of respective dyadic cubes, and so the argument analogous to Lemma 10 also goes through exactly, for this case. Next, for the upper bound on the gradient contribution from I_{\text{mid}} to the difference in the potential between q_{1} and q_{2} , we have, \displaystyle\leq\frac{C\,\Theta(U)\,M\,\gamma\,\kappa^{N}t_{1}}{(\kappa^{N}t_{1})^{\delta}}\left(1+\kappa^{\delta}+\kappa^{2\delta}+\cdots\right) (103) \displaystyle=C_{1}\,\Theta(U)\,M\,\gamma\,(\kappa^{N}t_{1})^{1-\delta}\frac{1}{1-\kappa^{\delta}} (104) \displaystyle=C_{1}\,\Theta(U)\,M\,\gamma\,(\kappa^{N}t_{1})^{1-\delta}\frac{1}{\kappa^{\delta}}\frac{1}{(\tfrac{1}{\kappa})^{\delta}-1} (105) \displaystyle\leq C\,\Theta(U)\,M\,\gamma\,(\kappa^{N}t_{1})^{1-\delta}\frac{1}{\kappa^{\delta-\delta_{2}}}\frac{1}{\delta\,\log(\tfrac{1}{\kappa})}, (106) for \delta small enough, and the \delta_{2} parameter coming from Eq. 80 as before. Thus, similar to Eq. 60 for the case of s=n-1 , here we have, in the cube C(q,\kappa^{N}t_{1}) , \displaystyle\frac{M^{-1}\gamma\Theta(U)\beta_{1}^{w(n+2-\delta)}}{\kappa^{\delta-\delta_{2}}\log\frac{1}{\kappa}}(\kappa^{N}t_{1})^{1-\delta}\frac{1}{\delta}\leq T_{N}\theta\gamma\kappa^{N}t_{1}\leq C\,\Theta(U)\,M\,\gamma\,(\kappa^{N}t_{1})^{1-\delta}\frac{1}{\kappa^{\delta-\delta_{2}}}\frac{1}{\delta\,\log(\tfrac{1}{\kappa})} (107) Recall that we have, \beta_{1}^{2w}\leq\theta\leq 1 . In particular, we have from here, that, \displaystyle T_{N}\leq\frac{1}{\theta}C\,\Theta(U)\,M\,\gamma\,(\kappa^{N}t_{1})^{-\delta}\frac{1}{\kappa^{\delta-\delta_{2}}}\frac{1}{\delta\,\log(\tfrac{1}{\kappa})}\leq\Big(\frac{1}{\beta_{1}^{2w}}\Big)C\,\Theta(U)\,M\,\gamma\,(\kappa^{N}t_{1})^{-\delta}\frac{1}{\kappa^{\delta-\delta_{2}}}\frac{1}{\delta\,\log(\tfrac{1}{\kappa})} (108) Thus, if we further go down to a radial length \kappa^{N+N_{2}}t_{1} , obtain the same bounds on the second derivative contributions from I_{\text{mid}},I_{\text{far}} , the potential contribution from I_{\text{near}} , by going down to a length scale of \kappa^{N+N_{2}+N_{1}} , so that only the gradient contributions from I_{\text{mid}} and I_{\text{far}} dominate these previous contributions, and we require that \displaystyle\frac{\beta_{1}^{\,w(n+2-\delta)}M^{-1}\Theta(U)\gamma}{\,\delta\kappa^{\delta-\delta_{2}}\log\frac{1}{\kappa}\,}\left(\kappa^{N+N_{2}}t_{1}\right)^{-\delta}\;\geq M\cdot\frac{1}{\beta_{1}^{2w}}\frac{M\Theta(U)\gamma}{\delta\kappa^{\delta-\delta_{2}}\log\frac{1}{\kappa}\,}\left(\kappa^{N}t_{1}\right)^{-\delta}, (109) then making the same argument within the cube C(q_{1},\kappa^{N+N_{2}}t_{1}) , we obtain a contradiction. (9) (9) (9) Note that \gamma is chosen to satisfy Eq. 101 This is because within the cube C(q_{1},\kappa^{N+N_{2}}t_{1}) , we must have analogous to Eq. 63 , and using Eq. 107 , that, \frac{M^{-1}\gamma\Theta(U)\beta_{1}^{w(n+2-\delta)}}{\kappa^{\delta-\delta_{2}}\log\frac{1}{\kappa}}(\kappa^{N+N_{2}}t_{1})^{1-\delta}\frac{1}{\delta}\leq T_{N+N_{2}}\theta\gamma\kappa^{N+N_{2}}t_{1}\\ \Rightarrow T_{N+N_{2}}=T_{N}\geq\frac{M^{-1}\gamma\Theta(U)\beta_{1}^{w(n+2-\delta)}}{\delta\kappa^{\delta-\delta_{2}}\log\frac{1}{\kappa}}(\kappa^{N+N_{2}}t_{1})^{-\delta} (110) To satisfy Eq. 109 , it is enough to require that, \beta_{1}^{\,w(n+4-\delta)}M^{-1}\kappa^{-N_{2}\delta}\;\gtrsim\;M^{2}, so \left(\frac{1}{\kappa}\right)^{N_{2}\delta}\;\gtrsim\;M^{3}\left(\frac{1}{\beta_{1}}\right)^{w(n+4-\delta)}. Hence, N_{2}\delta\log\frac{1}{\kappa}\;\gtrsim\;3\log M+w(n+4-\delta)\log\frac{1}{\beta_{1}}. Therefore, when \displaystyle N_{2}\;\gtrsim\;\frac{3\log M+\bigl(\frac{\log C}{\beta_{1}^{n}}\bigr)(n+4-\delta)\log\frac{1}{\beta_{1}}}{\delta\log\frac{1}{\kappa}}. (111) we have the contradiction. Lastly by an argument identical to the one for the case of (1) with s=n-1 , we also iterate this argument \sim\log M/M_{1} many times within the dyadic cube U , to conclude Theorem 9 . ∎ 5.0.1 Change of poles argument. Now we adopt the proof of Lemma 4.3 of [ Az20 ] , with modifications, to complete the argument. Lemma 12 . Let Q be a dyadic cube centered on \partial\Omega and let p\in\Omega\setminus Q , and the boundary \partial\Omega is (s,C_{1}) - Ahlfors-David regular, with n-1-\delta_{0}\leq s\leq n-1 . Consider the center c(Q) of the dyadic cube Q , and as in Theorem 9 , consider the dyadic cube U containing c(Q) so that l(U)=\frac{1}{C^{\prime}}l(Q) with C^{\prime}\approx C and C is large enough depending on the parameters of Theorem 5 (3). Set \omega=\omega^{p}_{\Omega} . Then for every M_{1}>1 large enough dependent on C^{\prime} , there exists \eta>0 dependent on M_{1} , and a dyadic cube Q_{1}\subset\tfrac{1}{2}B_{Q} such that l(Q_{1})\geq\eta l(Q) and \Theta^{s}_{\omega}(Q_{1})\notin\bigl[M_{1}^{-1}\,\Theta^{s}_{\omega}(Q),\,M_{1}\,\Theta^{s}_{\omega}(Q)\bigr]. Proof. Consider the ball B_{0}=\frac{1}{5}B_{U} and the pole p\in\Omega\setminus aB_{0} , with a=10b^{-1}>b^{-1} , and b as in the Bourgain lemma. We assume that \omega^{p}\!\left(\tfrac{1}{10}B_{0}\right)\geq M_{2}^{-1}\,\omega^{p}(Q), (112) with M_{2} large enough to be chosen momentarily. Note that the surfaces measures of \frac{1}{10}B_{0} and Q are comparable depending on C^{\prime} . Assuming that Eq. 112 does not hold, we consider the minimal disjoint covering \mathcal{C}_{1} of \frac{1}{10C^{\prime}}B_{0} by dyadic cubes of diameter comparable to l(\frac{1}{10C^{\prime}}B_{0}) . If the Poisson kernel of each of the dyadic cubes in this cover \mathcal{C}_{1} is greater than M^{-1}_{1}\Theta^{s}_{\omega}(Q) , we then easily get that, \displaystyle\omega^{p}(\frac{1}{10}B_{0})>\omega^{p}(\cup_{S\in\mathcal{C}_{1}}S)>M^{-1}_{1}\sigma(\cup_{S\in\mathcal{C}_{1}}S)\Theta^{s}_{\omega^{p}}(Q)>M^{-1}_{1}C_{3}\sigma(\frac{1}{10}B_{0})\Theta^{s}_{\omega^{p}}(Q)=M_{1}^{-1}C_{4}\omega^{p}(Q). (113) Thus if we choose M^{-1}_{1}C_{4}>M^{-1}_{2} , we get a contradiction. Thus, at least one of the dyadic cubes in \mathcal{C}_{1} must have Poisson kernel less than or equal to M^{-1}_{1}\Theta^{s}_{\omega}(Q) and we would be done. Thus assume that Eq. 112 holds, and consider, E=\{x\in\partial B_{0}\cap\Omega:\delta_{\Omega}(x)>\varepsilon\}, and let \widetilde{\omega}=\omega_{\Omega\setminus B_{0}} . By the Markov property for harmonic measure, and the Bourgain lemma, we get, for \varepsilon>0 small enough, \displaystyle\omega^{p}\!\left(\tfrac{1}{10}B_{0}\right) \displaystyle=\int_{\partial B_{0}\cap\Omega}\omega^{x}\!\left(\tfrac{1}{10}B_{0}\right)\,d\widetilde{\omega}^{p}(x) \displaystyle\leq C\varepsilon^{\alpha}\,\widetilde{\omega}^{p}(\partial B_{0}\cap\Omega)+\int_{E}\omega^{x}\!\left(\tfrac{1}{10}B_{0}\right)\,d\widetilde{\omega}^{p}(x). By the Bourgain lemma, we have \omega^{x}(Q)\gtrsim 1 on \partial B_{0}\cap\Omega , and thus by the maximum principle we get \widetilde{\omega}^{p}(\partial B_{0}\cap\Omega)\leq\omega^{p}(Q)\leq M_{2}\,\omega^{p}\!\left(\tfrac{1}{10}B_{0}\right). The last inequality follows from Eq. 112 . Thus, with \varepsilon>0 small enough depending on M_{1} and a , we get from the above two inequalities that \omega^{p}\!\left(\tfrac{1}{10}B_{0}\right)\leq 2\int_{E}\omega^{x}\!\left(\tfrac{1}{10}B_{0}\right)\,d\widetilde{\omega}^{p}(x). (114) Let B_{j} be a covering of E by bounded number of balls centered on E , of radius \varepsilon/4 , so that 2B_{j}\subset\Omega . We show there is t>0 (depending on \varepsilon and M_{1} ), and there is j so that if B^{\prime}=B_{j} , we get \widetilde{\omega}^{p}(B^{\prime})\geq t\,\omega^{p}\!\left(\tfrac{1}{10}B_{0}\right). If not, then for each j we have \widetilde{\omega}^{p}(B_{j})M\,\frac{\sigma(Q_{1})}{\sigma(Q)}\omega^{x}(Q)\geq Mc_{0}\frac{\sigma(Q_{1})}{\sigma(Q)}. In the last step above, we have again used the Bourgain estimate. So, from Eq. 116 above, we get, \omega^{p}(Q_{1})\gtrsim M_{2}^{-1}Mc_{0}\omega^{p}(Q)\frac{\sigma(Q_{1})}{\sigma(Q)}. Thus if we choose M large enough compared to M_{2} , we get the desired result. ∎ This concludes the result and leads to the following lemma. 6 Dimension drop of the harmonic measure Recall that the dimension of a Borel measure \mu in \mathbb{R}^{n} is defined as follows: \dim(\mu)=\inf\bigl\{\dim(G):G\subset\mathbb{R}^{n}\text{ Borel},\ \mu(G^{c})=0\bigr\}. Lemma 13 . For n\geq 1 , s>0 , C_{1}>1 , for any large M=M(n,s,C_{0})>1 , the following holds. Let \partial\Omega\subset\mathbb{R}^{n} be an (s,C_{1}) –AD regular set with n-1-\delta_{0}\leq s\leq n-1 . Let \omega:=\omega^{p_{0}} be the harmonic measure with pole p_{0} , supported on \partial\Omega and c_{1}\in(0,1) such that, for any dyadic cube C_{1}\subset\partial\Omega , 0