---
title: Cheng-type Eigenvalue Comparison Theorems
url: https://www.emergentmind.com/topics/cheng-type-eigenvalue-comparison-theorems
type: topic
---

# Cheng-type Eigenvalue Comparison Theorems

Cheng-type eigenvalue comparison theorems are spectral comparison results in which eigenvalues of Laplace-type operators on a manifold or a domain are estimated against model eigenvalues determined by constant-curvature space forms, one-dimensional Sturm–Liouville problems, or synthetic curvature-dimension models. In the classical Riemannian setting, Cheng’s 1975 estimates control Laplace eigenvalues of closed manifolds by diameter or volume under a lower Ricci curvature bound; subsequent work extends the same comparison paradigm to geodesic balls, convex and non-convex domains, Robin and mixed boundary problems, weighted and nonlinear operators, differential forms, Kähler geometry, weak curvature hypotheses, and non-smooth metric measure spaces [1510.07281][2607.04588][2507.23671].

## 1. Classical formulation and model spaces

In the closed-manifold setting, Cheng’s comparison principle appears as an upper bound on Laplace eigenvalues under a lower Ricci curvature bound. If \((M^n,g)\) is closed and connected with
\[
\mathrm{Ricci}\ge -(n-1)K,\qquad K\ge 0,
\]
then there exist constants \(C_6(n)\) and \(C_7(n)\), depending only on \(n\), such that for every \(k\ge 1\),
\[
\lambda_k(M)\le \frac{(n-1)^2}{4}K + C_6(n)\Bigl(\frac{k}{d}\Bigr)^2,
\qquad
\lambda_k(M)\le \frac{(n-1)^2}{4}K + C_7(n)\Bigl(\frac{k}{V^{1/n}}\Bigr)^2,
\]
where \(d=\operatorname{diam}(M)\) and \(V=\operatorname{Vol}_g(M)\) [1510.07281]. In this formulation, the term \(\frac{(n-1)^2}{4}K\) reflects the bottom of the spectrum on the simply connected space form of constant curvature \(-K\), while the remaining term encodes the global size parameter.

A more local version compares the first Dirichlet eigenvalue of a geodesic ball with that of the corresponding ball in a model space. In the weighted framework of Chen–Ma–Mao, the model eigenvalue \(\lambda_1^m(K,R)\) is defined by the radial Sturm–Liouville problem
\[
\phi''(r)+(m-1)\frac{s_K'(r)}{s_K(r)}\phi'(r)+\lambda_1^m(K,R)\phi(r)=0,
\qquad
\phi'(0)=0,\ \phi(R)=0,
\]
and if \(f\equiv\mathrm{const}\) and \(m=n\), the bound
\[
\lambda_1(B_o(R))\le \lambda_1^n(K,R)
\]
is exactly Cheng’s original result for the Laplacian [2607.04588]. The same model-ball viewpoint also underlies the \((p,q)\)-Laplacian comparison of Abolarinwa–Azami, where \(\lambda_{1,p,q}(B(x_0,r_0))\le \lambda_{1,p,q}(V_n(k,r_0))\) under \(\mathrm{Ric}_M\ge (n-1)k\), with equality precisely in the isometric case [2007.06303].

The classical model spaces are geodesic balls in simply connected space forms. For \(K=0\), Cheng’s comparison reduces to Euclidean balls; for \(K<0\), the hyperbolic radial ODE replaces the Euclidean Bessel description; and for \(K>0\), the admissible radius is restricted by the model injectivity scale [1510.07281][2607.04588].

## 2. Comparison mechanisms

The standard proof architecture begins with a radial ground state on the model ball and transplants it to the target manifold. In the \((p,q)\)-system, one takes the radial first eigenfunctions \((\hat u,\hat v)\) on the model ball \(V_n(k,r_0)\), defines \(\tilde u(x)=\hat u(\operatorname{dist}(x,x_0))\) and \(\tilde v(x)=\hat v(\operatorname{dist}(x,x_0))\), and inserts the pair into the Rayleigh-type variational characterization of \(\lambda_{1,p,q}\) [2007.06303]. The key geometric input is Bishop’s volume comparison in polar coordinates,
\[
d\mathrm{Vol}_M=J_M(t,\theta)\,dt\,d\theta,\qquad J_M(t,\theta)\le J_k(t),
\]
together with monotonicity of the radial model eigenfunctions and an integration-by-parts argument that transfers the model ODE to the ambient manifold.

In the weighted Bakry–Émery setting, the same mechanism is recast in terms of the weighted measure \(d\mu_f=e^{-f}d\mathrm{vol}_g\). Under
\[
\mathrm{Ric}_f^m(\gamma'(t),\gamma'(t))\ge (m-1)K
\]
along minimizing radial geodesics, one has the generalized Bishop–Gromov estimate
\[
e^{-f(r,\theta)}\sqrt{\det g_{ij}(r,\theta)}\le s_K(r)^{m-1}
\]
and the radial Laplacian comparison
\[
\Delta_f r \le (m-1)\frac{s_K'(r)}{s_K(r)}.
\]
These inequalities make the transplanted radial function \(U(x)=\phi(r(x))\) an admissible trial function whose weighted Rayleigh quotient is bounded above by the model eigenvalue [2607.04588].

For global spectral bounds, Cheng’s method is often combined with min–max splitting. In the Kato-Ricci setting, Carron–Rose heat-kernel estimates, volume doubling, and a Poincaré inequality imply a Dirichlet estimate on each ball \(B(x,R)\),
\[
\lambda_1^{\mathrm{Dir}}(B(x,R))\le \mathrm{const}\cdot R^{-2},
\]
and since a manifold of diameter \(D\) contains two disjoint balls of radius \(D/2\), the Courant principle yields
\[
\lambda_1(M)\le \max\{\lambda_1^{\mathrm{Dir}}(B_1),\lambda_1^{\mathrm{Dir}}(B_2)\}\le \mathrm{const}\cdot D^{-2}
\]
[2003.07075]. In the non-smooth setting, this proof pattern is replaced by localization: the measure is disintegrated along one-dimensional rays, the comparison is proved fiberwise, and the global inequality is recovered by reintegration [2507.23671]. This suggests that the robust core of the Cheng method is not a specific curvature tensor inequality but the availability of a one-dimensional comparison structure together with a variational principle.

## 3. Nonlinear, weighted, and higher-rank extensions

The modern literature uses “Cheng-type” for a family of comparison principles that preserve the model-space logic while changing the operator, the boundary condition, or the geometric category.

| Setting | Operator | Comparison statement |
|---|---|---|
| Geodesic balls | \((p,q)\)-Laplacian system | \(\lambda_{1,p,q}(B(x_0,r_0))\le \lambda_{1,p,q}(V_n(k,r_0))\) under \(\mathrm{Ric}_M\ge (n-1)k\) |
| Geodesic balls and compact manifolds with boundary | Robin \(p\)-Laplacian | For \(\alpha>0\), model-ball or one-dimensional lower/upper bounds; inequalities reverse for \(\alpha<0\) |
| Weighted complete manifolds | Witten-Laplacian and weighted \(p\)-Laplacian | \(\lambda_{1,f}(B_o(R))\le \lambda_1^m(K,R)\) and \(\lambda_{1,p,f}(B_o(R))\le \lambda_{1,p}^m(K,R)\) |
| Closed manifolds | Hodge Laplacian on \(p\)-forms | Uniform upper bounds under Ricci, injectivity-radius, and diameter bounds |
| Compact Kähler manifolds | Complex Laplacian \(\Box=\bar\partial^*\bar\partial\) | \(\lambda_1\ge k\) if \(\mathrm{Ric}(g)\ge k\,g\) |

For the \((p,q)\)-Laplacian system, the Cheng comparison is sharp: equality holds if and only if the metric ball is isometric, via the exponential map, to the model space form ball [2007.06303]. In the weighted case, Chen–Ma–Mao prove the analogous sharp inequalities for both the Witten-Laplacian and the weighted \(p\)-Laplacian under a lower bound on the \(m\)-Bakry–Émery Ricci tensor; they also identify the classical Cheng theorem as the specialization \(f\equiv\mathrm{const}\), \(m=n\) [2607.04588].

Boundary conditions substantially modify the comparison model. For the first Robin eigenvalue \(\lambda_p(M,\alpha)\) of the \(p\)-Laplacian, Li–Wang prove that on a geodesic ball \(B_R(x_0)\) with \(\mathrm{Ric}_M\ge (n-1)\kappa\),
\[
\lambda_p(B_R(x_0),\alpha)\le \lambda_p(V(\kappa,R),\alpha)\quad\text{if }\alpha>0,
\]
while the inequality reverses for \(\alpha<0\). For compact manifolds with smooth boundary, Ricci lower bound, boundary mean-curvature lower bound, and inradius \(R\), the comparison is with a one-dimensional model problem involving
\[
T_{\kappa,\Lambda}(t)=\frac{C'_{\kappa,\Lambda}(t)}{C_{\kappa,\Lambda}(t)},
\]
and the Dirichlet problem is recovered as \(\alpha\to+\infty\) [2002.06472]. In a related mixed-boundary direction, the \(p\)-Laplacian on domains with an interior hole admits Faber–Krahn-type inequalities and Cheng-type comparison theorems on manifolds, together with comparison results for inner Dirichlet and outer Neumann data in minimal submanifolds in Euclidean space [1409.7145].

The label also extends beyond scalar second-order operators. Bhattacharya–Maity establish Cheng-type upper bounds for the Hodge Laplacian on differential forms under
\[
\mathrm{Ric}_g\ge (n-1)\xi,\qquad \mathrm{Inj}(M,g)\ge r_0,\qquad \operatorname{diam}M\le D,
\]
replacing the sectional-curvature hypotheses used by Dodziuk and Lott with Ricci lower bounds plus harmonic-coordinate control [2603.10633]. In Kähler geometry, Tam–Yu prove the lower bound
\[
\lambda_1\ge k
\]
for the first nonzero eigenvalue of the complex Laplacian under \(\mathrm{Ric}(g)\ge k\,g\), with rigidity to a Kähler–Einstein metric in the equality case and, under positive holomorphic bisectional curvature, rigidity to \(\mathbb{C}P^n\) with the Fubini–Study metric [1007.1498]. For higher-order elliptic operators, the poly-Laplacian literature includes sharp upper bounds for sums of the first \(k\) Dirichlet eigenvalues for arbitrary order, presented as an extension of Cheng–Wei and improved when \(l=2\) and \(k\) is large enough [1307.5313].

## 4. Weak curvature hypotheses and alternative geometric controls

A major line of development weakens the pointwise Ricci lower bound while retaining Cheng-type spectral control. Under a Kato condition on the negative part \(p_-\) of the Ricci curvature,
\[
K_{D^2}(p_-)\le \frac{1}{16n},\qquad D=\operatorname{diam}(M),
\]
Rose proves that on a closed manifold the first nonzero eigenvalue satisfies
\[
\lambda_1(M)\le C(n)\,D^{-2}.
\]
The proof uses Li–Yau type gradient estimates, Harnack inequalities, two-sided heat-kernel bounds, volume doubling, and a Poincaré inequality derived from the small Kato constant [2003.07075]. In the boundary case, the same paper introduces the Neumann Kato constant \(K_T^N(p_-)\) and obtains, under additional boundary geometry assumptions, a Cheng-type lower bound for the first nonzero Neumann eigenvalue \(\mu_1\).

A different weakening replaces Ricci control by scalar curvature. Munăteanu–Wang prove that if a complete noncompact \(3\)-manifold satisfies
\[
S_g\ge -6K,
\]
has finitely many ends, finite first Betti number, a lower Ricci bound, and a mild volume non-collapse condition, then the bottom of the spectrum obeys the sharp upper bound
\[
\lambda_0(M)\le K.
\]
This contrasts with the classical Ricci-based estimate \(\lambda_0(M)\le (n-1)^2K\) in dimension \(n\), and the rigidity question for the equality case remains open [2105.12103].

Integral curvature assumptions yield another variant. Kwong’s Theorem 17 assumes a one-sided integral bound on \(\mathrm{Ric}_-\) along all geodesics emanating from the center and obtains a model-ball comparison for the first Dirichlet eigenvalue, formulated there as
\[
\lambda_1(B(r,p))\ge \lambda_1(\widetilde B(r)),
\]
with equality if and only if the ball is isometric to the model space [1904.08595]. The breadth of these formulations suggests that “Cheng-type” now denotes a comparison strategy rather than a single fixed inequality: depending on the operator, the boundary condition, and the curvature input, the conclusion may be an upper bound, a lower bound, or a model-space rigidity statement.

## 5. Synthetic curvature and the non-smooth extension

In the non-smooth category, De Luca, De Ponti, Mondino, and Tomasiello extend Cheng’s comparison to essentially non-branching \(\mathsf{CD}^{\star}(K,N)\) spaces. Writing
\[
d\mathfrak m_{K,N}(\theta)=h_{K,N}(\theta)\,d\theta,\qquad
h_{K,N}(\theta)=\mathfrak s_{K/(N-1)}(\theta)^{N-1},
\]
they define the one-dimensional model eigenvalue \(\lambda_{K,N,r_0}\) by the weighted Dirichlet problem on \([0,r_0]\), and prove the sharp estimate
\[
\lambda_1^D(B(x_0,r_0))\le \lambda_{K,N,r_0}
\]
for every metric ball in an essentially non-branching \(\mathsf{CD}^{\star}(K,N)\) space [2507.23671]. When the space is a smooth \(N\)-manifold with \(\mathrm{Ric}\ge K\,g\), this recovers the classical Cheng upper bound.

The proof is driven by the localization technique. The ball is decomposed into one-dimensional rays associated with the distance function \(u(x)=d(x_0,x)\), the ambient measure is disintegrated into conditional measures \(h_q(t)\,dt\) satisfying one-dimensional curvature-dimension inequalities, and the model eigenfunction \(\phi_{K,N,r_0}(d(x_0,\cdot))\) is tested fiberwise before being reintegrated globally [2507.23671]. This replaces smooth polar-coordinate comparison by a synthetic one-dimensional convexity principle.

The corresponding rigidity theorem on \(\mathsf{RCD}^{\star}(K,N)\) spaces is highly structured. If equality holds in the Dirichlet comparison, then exactly one of three mutually exclusive configurations occurs: \(\partial B(x_0,r_0/2)\) is a single point, exactly two points, or at least three points, leading respectively to a one-dimensional weighted model, a local weighted interval model, or an \(\mathsf{RCD}^{\star}(N-2,N-1)\)-cone structure [2507.23671].

The same paper derives two further consequences. First, if the compact space has diameter \(D\), then for every \(j\ge 1\) the \(j\)-th Neumann eigenvalue satisfies
\[
\lambda_j^N(X)\le \lambda_{K,N,r_0},\qquad r_0=\frac{D}{2j}.
\]
Second, for a non-compact \(\mathsf{RCD}(K,N)\) space with \(K\le 0\) and \(N\ge 3\),
\[
\inf \sigma_{\mathrm{ess}}(-\Delta)\le -\frac{(N-1)K}{4}.
\]
A physical application identifies spin-2 Kaluza–Klein masses with Neumann eigenvalues of a weighted Laplacian and concludes
\[
m_j^2\le \lambda_{K,N,r_0},
\qquad r_0=\frac{\operatorname{diam}}{2j},
\]
for warped compactifications satisfying the reduced energy condition [2507.23671].

## 6. Rigidity, sharpness, and current directions

Rigidity is a persistent feature of Cheng-type comparison. In the \((p,q)\)-Laplacian system, equality forces the geodesic ball to be isometric to the model ball [2007.06303]. For the Witten-Laplacian and weighted \(p\)-Laplacian, equality holds if and only if \((B_o(R),g,f)\) is isometric to the model ball in \(M_K^m\), with \(f\) constant along each radius [2607.04588]. For the Robin \(p\)-Laplacian, equality in the inradius comparison characterizes the canonical \((\kappa,\Lambda)\)-model spaces [2002.06472]. In the Kähler setting, equality in \(\lambda_1\ge k\) implies Kähler–Einstein geometry, and under positive holomorphic bisectional curvature it characterizes \(\mathbb{C}P^n\) with the Fubini–Study metric [1007.1498]. In the synthetic setting, equality yields a one-dimensional or conical structure [2507.23671].

Sharpness is more nuanced. For Cheng’s closed-manifold diameter estimate, the quadratic growth in \(k\) is sharp for thin flat tori, whereas the volume form is asymptotically sharp in the sense of Weyl [1510.07281]. For the Hodge Laplacian, Bhattacharya–Maity emphasize that the Cheng-type upper bound has the same \(k\)-power as Weyl’s law, so it is sharp in order of growth [2603.10633]. In the scalar-curvature comparison of Munăteanu–Wang, \(\mathbb H^3(-K)\) attains \(\lambda_0=K\), but the rigidity question under only the scalar bound is still open, and higher-dimensional analogues fail in general [2105.12103].

A common misconception is that Cheng-type theorems concern only the original Dirichlet comparison for geodesic balls under pointwise Ricci lower bounds. Current usage is materially broader: it includes closed-manifold diameter and volume bounds, weighted and nonlinear analogues, Robin and mixed boundary problems, lower bounds such as the Kähler estimate \(\lambda_1\ge k\), bottom-spectrum estimates under scalar curvature, and synthetic \(\mathsf{CD}^{\star}(K,N)\) statements [1510.07281][1007.1498][2002.06472][2507.23671].

A related branch of the literature replaces direct model comparison by universal inequalities in the Cheng–Yang tradition. For divergence-form operators, including the Laplacian and the square Cheng–Yau operator, one obtains gap estimates of the form
\[
\lambda_{k+1}-\lambda_k\le C_{n,\Omega,T,\eta}(k+1)^{1/n},
\]
with the Laplacian case coinciding with the result of Chen–Zheng–Yang in the order of the eigenvalues [2408.05068]. For the drifting Laplacian on bounded domains, Chen–Gomes–Miranda extend Cheng–Yang average bounds and second-Yang type inequalities by incorporating Weyl asymptotics and extrinsic geometry [1607.00066]. On pinched Cartan–Hadamard manifolds, drifted Cheng–Yau operators satisfy universal inequalities derived from a Bochner-type formula and Rauch comparison [2107.09135]. In hyperbolic space, Cheng’s conjectured curvature-shifted inequality
\[
\sum_{i=1}^k(\lambda_{k+1}-\lambda_i)^2
\le \frac{4}{n}\sum_{i=1}^k(\lambda_{k+1}-\lambda_i)\Bigl(\lambda_i-\frac{(n-1)^2}{4}\Bigr)
\]
has been verified up to a factor \(1+\epsilon\) for two special classes of bounded domains, while the full conjecture for arbitrary domains remains open [2604.20343].

Taken together, these developments show that the Cheng comparison paradigm has evolved from a model-ball estimate for the scalar Laplacian into a general spectral methodology. Its modern forms may be driven by Bishop–Gromov comparison, Bakry–Émery geometry, heat-kernel and Kato techniques, Green’s-function identities under scalar curvature, localization on one-dimensional needles, or harmonic-coordinate control for differential forms. What remains invariant is the guiding principle: global or local geometric control is converted into a comparison with a tractable model spectrum.

Source: https://www.emergentmind.com/topics/cheng-type-eigenvalue-comparison-theorems