--- title: Moser's Harnack Inequality url: https://www.emergentmind.com/topics/moser-s-harnack-inequality type: topic --- # Moser's Harnack Inequality Moser’s Harnack inequality is the interior comparison principle asserting that a nonnegative weak solution of a uniformly elliptic or parabolic equation cannot oscillate arbitrarily on smaller domains. In the standard elliptic divergence-form setting, \[ (a_{ij}u_i)_j=0 \quad \text{in } Q_6, \] with uniformly elliptic coefficient matrix \((a_{ij})_{n\times n}\) and ellipticity constants \(\lambda,\Lambda\), the classical statement is \[ \sup_{Q_1} u \le C \inf_{Q_1} u, \] where \(C\) depends only on \(n,\lambda,\Lambda\) [1901.06128]. In the parabolic setting, the same principle compares the supremum on an earlier cylinder with the infimum on a later one, reflecting forward-in-time propagation [2212.07977]. Within De Giorgi–Moser theory, Harnack’s inequality sits alongside local boundedness and weak Harnack estimates as a cornerstone of regularity theory for uniformly elliptic divergence-form equations, and it is one of the central tools behind interior continuity, positivity propagation, and compactness arguments [1901.06128]. ## 1. Historical position and classical meaning In the chronology recalled by Li and Zhang, De Giorgi proved Hölder continuity for elliptic equations in divergence form in 1957, Moser gave a new proof in 1960, and in the following year Moser obtained the Harnack inequality [1901.06128]. That sequence is historically important because the later note argues that the Harnack inequality was already “hidden” in De Giorgi’s 1957 work. For the parabolic theory, the 2022 trajectorial paper states that in 1971 Moser published a simplified version of his proof of the parabolic Harnack inequality, with the fundamental lemma of Bombieri and Giusti as the core new ingredient [2212.07977]. The classical content is quantitative rather than merely qualitative. In the elliptic case, the inequality compares the maximum and minimum of a nonnegative solution on a smaller cube. In the parabolic case, it compares values on separated subcylinders, with the supremum taken earlier in time and the infimum later in time. This temporal asymmetry is not an incidental technicality: it is built into the formulation because diffusion propagates forward in time [2212.07977]. A common oversimplification is to identify Moser’s contribution only with one final sup–inf estimate. The literature summarized here presents a more structured picture: Harnack’s inequality emerges from a chain consisting of local boundedness, weak Harnack control, and a logarithmic mechanism that links upper and lower estimates [1901.06128] [2212.07977]. ## 2. Elliptic divergence-form theorem and the “hidden” De Giorgi structure For the standard linear uniformly elliptic equation in divergence form, \[ (a_{ij}u_i)_j=0 \quad \text{in } Q_6, \] Moser’s recalled theorem states that if \(u\ge 0\) is a weak solution, then \[ \sup_{Q_1} u \le C \inf_{Q_1} u, \] with \(C\) depending only on \(n,\lambda,\Lambda\) [1901.06128]. Li and Zhang identify two ingredients already present in De Giorgi’s 1957 work. The first is the local maximum estimate for nonnegative weak subsolutions, \[ \|u\|_{L^\infty(Q_1)} \le C \|u\|_{L^2(Q_3)}, \] and, in the more general form used in the note, \[ \|u\|_{L^\infty(Q_1)} \le C \|u\|_{L^{p_0}(Q_3)} \] for any \(p_0>0\). The second is a weak Harnack-type measure-to-point estimate for nonnegative weak supersolutions: for any \(c_0>0\), there exists \(c>0\), depending only on \(n,\lambda,\Lambda,c_0\), such that \[ m\{x\in Q_1:\ u(x)>c\}>c_0 \quad \Longrightarrow \quad u>1 \text{ in } Q_3. \] From this, the note derives a weak Harnack inequality \[ \|u\|_{L^p(Q_1)} \le C \inf_{Q_3} u \] for some \(p>0\) [1901.06128]. The proof architecture is short once these two estimates are available. One normalizes by assuming \(\inf_{Q_3}u=1\), constructs a decreasing sequence \(c_k\), proves the decay \[ m\{x\in Q_1:\ u(x)>c_k\} \le 2^{-k}m(Q_1), \] and uses Theorem 1.3 together with a Calderón–Zygmund cube decomposition to propagate the measure decay from level \(k-1\) to level \(k\). The full Harnack inequality then follows by combining the local maximum principle with the weak Harnack estimate. The note explicitly frames this as showing that De Giorgi’s 1957 argument already contains the Harnack inequality in implicit form, and attributes that historical observation to DiBenedetto [1901.06128]. ## 3. Parabolic formulation and the Bombieri–Giusti mechanism In Moser’s parabolic setting, the equation is \[ \partial_t u = \nabla \cdot (A(t,x)\nabla u) \] on a space-time cylinder \(Q_T=(0,T)\times \Omega\), where \(A(t,x)\in L^\infty(Q_T;\mathbb R^{d\times d})\) is uniformly elliptic: \[ \lambda |\xi|^2 \le (A(t,x)\xi,\xi), \qquad |a_{ij}(t,x)|\le \Lambda. \] For a nonnegative weak solution in a parabolic cylinder \(Q\), the Harnack inequality compares the supremum on a backward subcylinder \(Q_-\) with the infimum on a later forward subcylinder \(Q_+\): \[ \sup_{Q_-} u \le C\, \inf_{Q_+} u, \] with \(C=C(d,\lambda,\Lambda,T)\) [2212.07977]. The 2022 reinterpretation isolates the two quantitative inputs in Moser’s 1971 streamlined proof. The first is an \(L^p\!-\!L^\infty\) estimate for positive solutions, obtained by iteration. The second is a weak \(L^1\) estimate for \(\log u\): for a positive weak supersolution \(u>0\), \[ \bigl|\{(t,x)\in K_-:\log u(t,x)>c+s\}\bigr| \le C\,r^2|B|\,s^{-1},\qquad s>0, \] and similarly on \(K_+\) for lower tails of \(\log u\). These two inputs are then glued by the Bombieri–Giusti lemma, which is the abstract device converting an \(L^p\!-\!L^\infty\) estimate and weak \(L^1\) logarithmic control into the Harnack comparison [2212.07977]. The paper’s new contribution is geometric. It gives a trajectorial proof of the weak \(L^1\) estimate by following parabolic curves \[ \gamma(r)=\bigl(t+r^2(\eta-t),\,x+r(y-x)\bigr), \qquad r\in[0,1], \] and integrating \(g=\log u\) along these paths. This replaces the classical Poincaré-based argument and interprets Moser’s logarithmic estimate as positivity propagation along parabolic trajectories. The authors emphasize that the new estimate is slightly weaker than Moser’s classical one because the cylinders must be separated by a positive time gap, but that this is still enough for Harnack since the Bombieri–Giusti mechanism only needs the estimate on well-separated sets [2212.07977]. ## 4. Weak Harnack, logarithmic control, and the structure of the method Across the papers summarized here, Moser’s Harnack inequality appears not as an isolated endpoint but as the output of a recurring analytic pattern. In the elliptic note, Moser’s 1961 proof is described as an iteration scheme involving \(L^{p_0}\) bounds for \(u\) and \(u^{-1}\), followed by the John–Nirenberg inequality to combine upper and lower estimates [1901.06128]. In the parabolic reinterpretation, the analogous bridge is the Bombieri–Giusti lemma rather than John–Nirenberg [2212.07977]. This distinction clarifies the role of weak Harnack inequalities. A weak Harnack estimate is an \(L^p\)-to-inf inequality for supersolutions; it is not yet the full sup–inf comparison. The full Harnack inequality emerges only after weak Harnack control is paired with local boundedness of subsolutions. The logarithmic estimate is the mechanism that allows these two directions to communicate. A related misconception is to treat the logarithmic step as a technical ornament. In the papers considered here, it is indispensable. Whether formulated through weighted BMO and John–Nirenberg, through Bombieri–Giusti, or through a direct trajectorial estimate for \(\log u\), the logarithmic control is the oscillation estimate that turns integral information into pointwise comparison [1901.06128] [2212.07977]. ## 5. Nonlocal, fractional, and metric-measure extensions The same Harnack architecture persists in several nonclassical settings, although the statement often changes form or acquires auxiliary terms. | Setting | Representative statement | Main mechanism | |---|---|---| | Fractional Laplace with lower order terms | \(\sup_{B_{1/2}}u \le C\big(\inf_{B_{1/2}}u+\|b\|_{L^\infty(B_1)}\big)\) | Caffarelli–Silvestre extension, Moser iteration, John–Nirenberg [1101.0013] | | Mixed local/nonlocal parabolic equation | weak Harnack with tail term \(\left(\fint_{V^-(r)} (u+T)^q\right)^{1/q} \le C(\essinf_{V^+(r)}u+T)\) | energy estimates, reverse Hölder, logarithmic estimate, Bombieri–Giusti [2105.15016] | | Time-fractional diffusion | weak Harnack \(\left(\fint_{Q^-} u^p\right)^{1/p} \le C\,\essinf_{Q^+}u\) with optimal critical exponent | Moser iteration, Yosida approximation, Bombieri–Giusti [1009.4852] | | Parabolic minimizers in metric spaces | \(\operatorname*{ess\,sup}_{\delta Q^-} u \le C\,\operatorname*{ess\,inf}_{\delta Q^+} u\) | variational method, doubling, weak \((1,p)\)-Poincaré, Bombieri–Giusti [1301.3765] | In the fractional Laplace case, the nonlocal equation \[ (-\Delta)^\sigma u(x)=a(x)u(x)+b(x) \] is converted into a degenerate local elliptic extension in one higher dimension, and the authors emphasize that no sign condition is assumed on \(a(x)\) [1101.0013]. In the mixed local/nonlocal parabolic setting, the nonlocal tail is intrinsic and cannot be discarded; it records the influence of values of the solution outside the spatial cylinder [2105.15016]. In the time-fractional problem, the inequality is weak rather than full, and the paper states that the full parabolic-style Harnack inequality remains open there [1009.4852]. In geodesic metric measure spaces, the argument becomes strictly variational and extends the sufficiency direction of the Grigor’yan–Saloff-Coste theorem from the linear heat equation to \(p>1\) [1301.3765]. ## 6. Quasilinear, anisotropic, subelliptic, and hypoelliptic variants Several papers recast Moser’s Harnack philosophy in settings where the operator is quasilinear, anisotropic, degenerate, or only hypoelliptic. For the quasilinear equation \[ -\Delta_pu+a(x,u)|\nabla u|^q=f(x,u), \] the 2016 paper proves a Harnack comparison inequality for positive \(C^1\) weak solutions by adapting Moser iteration in the style of Trudinger; the conclusion is a comparison estimate for \(v-u\), not the classical single-function theorem [1601.03863]. The 2025 anisotropic extension replaces \(|\nabla u|\) by a Finsler norm \(H(\nabla u)\) and studies \[ -\Delta_p^H u+a(x,u)H^q(\nabla u)=f(x,u), \] again via Moser iteration, obtaining a Harnack type comparison inequality and a Harnack type inequality for the linearized operator [2501.06593]. In the Grushin setting, the subelliptic nondivergence-form equation \[ Lu=\sum a_{ij}X_jX_i u=0 \] with \(X_1=\partial_{x_1}\) and \(X_2=x_1\partial_{x_2}\) is treated by a completely different route: weighted ABP, critical density, double ball property, power decay property, and then the axiomatic Harnack theory of Di Fazio–Gutiérrez–Lanconelli [1406.7311]. For the hypoelliptic operator \[ \Delta_y u + B(y)u_x + \gamma(x,y)u=0, \] the 2011 paper proves a Moser-style Harnack inequality for bounded nonnegative distributional solutions, but by a probabilistic proof based on the Feynman–Kac formula and hypoelliptic density estimates rather than Moser iteration [1112.3200]. These examples show that “Moser-style” may refer to the form of the conclusion and the regularity principle, even when the proof mechanism is no longer iterative. A further variation appears in the one-dimensional singular parabolic \(p\)-Laplacian, \(1