---
title: Bishop–Gromov Comparison Principles
url: https://www.emergentmind.com/topics/bishop-gromov-comparison
type: topic
---

# Bishop–Gromov Comparison Principles

Searching arXiv for relevant Bishop–Gromov comparison papers and related generalizations.
Bishop–Gromov comparison is a volume-comparison principle asserting that, under a lower curvature bound, the volume growth of geodesic balls is controlled by an explicit model space. In its classical Riemannian form, if a complete \(n\)-dimensional manifold satisfies \(\mathrm{Ric}_M\ge (n-1)\kappa\), then for every base point \(p\) the function
\[
r\longmapsto \frac{\Vol(B_p(r))}{\Vol_\kappa(r)}
\]
is nonincreasing, where \(\Vol_\kappa(r)=\omega_{n-1}\int_0^r[\sn_\kappa(t)]^{n-1}\,dt\) is the radius-\(r\) ball volume in the simply connected space form of constant sectional curvature \(\kappa\) [2404.09792]. The comparison has become a central template in comparison geometry, synthetic Ricci curvature, Alexandrov geometry, weighted and Finsler settings, Ricci flow, and spectral generalizations. In variable-curvature metric-measure spaces, the same structural principle persists, but the constant-curvature model \(\sin_{K/(N-1)}\) is replaced by Jacobi-model functions \(\mathfrak s_{k_x/(N-1)}\) derived from the ODE \(u''+\kappa(t)u=0\) [1506.03279].

## 1. Classical formulation and model-space normalization

For a complete Riemannian manifold \((M^n,g)\), the lower Ricci bound
\[
\mathrm{Ric}_M\ge (n-1)\kappa
\]
means that at every point \(p\in M\) and for every unit vector \(v\in T_pM\),
\[
\mathrm{Ric}_M(v,v)=\sum_{i=1}^{n-1}\langle R(e_i,v)v,e_i\rangle \ge (n-1)\kappa,
\]
where \(\{e_1,\dots,e_{n-1},v\}\) is any orthonormal basis of \(T_pM\) [2404.09792]. Writing \(B_p(r)=\{x\in M:d(p,x)<r\}\), the model-space comparison function is built from
\[
\sn_\kappa(t)=
\begin{cases}
\kappa^{-1/2}\sin(t\sqrt\kappa),&\kappa>0,\\
t,&\kappa=0,\\
|\kappa|^{-1/2}\sinh(t\sqrt{|\kappa|}),&\kappa<0,
\end{cases}
\]
and
\[
\Vol_\kappa(r)=\omega_{n-1}\int_0^r [\sn_\kappa(t)]^{n-1}\,dt
\]
[2404.09792].

The Bishop–Gromov theorem states that for all \(0<r<R\),
\[
\frac{\Vol(B_p(R))}{\Vol_\kappa(R)}
\;\le\;
\frac{\Vol(B_p(r))}{\Vol_\kappa(r)},
\]
so the normalized volume ratio is nonincreasing in \(r\) [2404.09792]. Equivalent formulations emphasize either the upper bound \(\Vol(B_p(r))\le \Vol_\kappa(r)\) or the monotonicity of the ratio \(V(r)/V_\kappa(r)\), where \(V(r)=\Vol(B_p(r))\) [2209.09288].

The comparison is sharp in the constant-curvature model. In the model space \(M_\kappa^n\) itself one has equality, and the ball volume is exactly the comparison volume [1003.4334]. This sharpness underlies the rigidity statements that accompany the theorem in many frameworks.

## 2. Mechanism of proof: Laplacian, area, and Riccati comparison

The classical derivation proceeds by passing from a radial differential inequality to an area comparison and then to a ball-volume comparison. If
\[
V(r)=\Vol(B_p(r)),\qquad A(r)=V'(r)=\Vol_{n-1}(\partial B_p(r)),
\]
then on the model space
\[
A_\kappa(r)=\omega_{n-1}\,\sn_\kappa^{\,n-1}(r),
\qquad
\frac d{dr}\bigl(\ln A_\kappa(r)\bigr)=(n-1)\ct_\kappa(r),
\]
where \(\ct_\kappa=\sn_\kappa'/\sn_\kappa\) [2404.09792].

On \((M,g)\), the radial distance \(r=d(p,\cdot)\) satisfies the Laplacian comparison
\[
\Delta r\le (n-1)\,\ct_\kappa(r)
\]
away from the cut locus, and equivalently the metric Jacobian \(J(r,\theta)\) in geodesic polar coordinates satisfies
\[
\frac{\partial}{\partial r}\ln J(r,\theta)\le (n-1)\,\ct_\kappa(r)
\]
[2404.09792]. Integrating in \(\theta\in S^{n-1}\) yields monotonicity of \(A(r)/A_\kappa(r)\), and a one-dimensional comparison lemma then implies monotonicity of \(V(r)/V_\kappa(r)\) [2404.09792].

A complementary formulation uses the Raychaudhuri equation. Along a unit-speed geodesic \(\gamma(t)\), the divergence \(\theta(t)\) of the geodesic spray obeys
\[
\dot\theta
=
-\frac1{d-1}\theta^2-\sigma^2+\omega^2-\Ric(\dot\gamma,\dot\gamma).
\]
For geodesic balls one has \(\omega=0\), and dropping the nonnegative shear term \(\sigma^2\) gives
\[
\dot\theta+\frac1{d-1}\theta^2\le -\Ric(\dot\gamma,\dot\gamma),
\]
which can be converted into scalar Jacobi inequalities for the radial Jacobian determinant [2209.09288]. This perspective is particularly important in later enhancements of Bishop–Gromov comparison that track shear rather than discarding it.

The equality case also follows the same chain of implications. If for some \(r_0>0\),
\[
\frac{\Vol(B_p(r_0))}{\Vol_\kappa(r_0)}
=
\frac{\Vol(B_p(r_1))}{\Vol_\kappa(r_1)}
\quad (0<r_1<r_0),
\]
then the Laplacian inequality, the shape-operator Riccati inequality, and the Jacobi-field comparisons all become equalities, and \(B_p(r_0)\) is isometric to the \(\kappa\)-model ball of radius \(r_0\) [2404.09792].

## 3. Variable curvature and synthetic metric-measure formulations

A major generalization replaces the constant lower curvature bound by a variable one. In Ketterer’s framework, \((X,d,m)\) is a complete, separable, geodesic, non-branching metric measure space with \(\mathrm{supp}\,m=X\), \(N>1\), and \(k:X\to\mathbb R\) lower-semicontinuous and locally bounded below on each ball [1506.03279]. For a base point \(x_0\in X\), define
\[
B_r=B_r(x_0),\qquad v(r)=m(B_r),
\]
and the Minkowski content
\[
s(r)=\limsup_{\varepsilon\to0^+}\frac{m(B_{r+\varepsilon}\setminus B_r)}{\varepsilon}.
\]
The radial infimum of the curvature function is
\[
k_x(r)=\inf\{k(y):y\in\partial B_r(x_0)\}.
\]
For \(\kappa\in C([0,L])\), the Jacobi-model function \(\mathfrak s_\kappa\) solves
\[
u''(t)+\kappa(t)u(t)=0,\qquad u(0)=0,\quad u'(0)=1,
\]
and
\[
\sigma_\kappa^{(t)}(L)=\mathfrak s_\kappa(tL)/\mathfrak s_\kappa(L)
\]
[1506.03279].

Under the curvature-dimension condition \(\mathrm{CD}(k,N)\), the variable-\(K\) Bishop–Gromov theorem states that if \(0<r<R\) and \(R\le \pi_{k_x/(N-1)}\), then the sphere comparison
\[
\frac{s(r)}{s(R)}
\ge
\frac{[\mathfrak s_{k_x/(N-1)}(r)]^{N-1}}
     {[\mathfrak s_{k_x/(N-1)}(R)]^{N-1}}
\]
and the ball comparison
\[
\frac{v(r)}{v(R)}
\ge
\frac{\int_0^r[\mathfrak s_{k_x/(N-1)}(t)]^{N-1}\,dt}
     {\int_0^R[\mathfrak s_{k_x/(N-1)}(t)]^{N-1}\,dt}
\]
hold [1506.03279]. In the constant-curvature case \(k\equiv K\), this recovers
\[
S_{K,N}(r)=\int_0^r[\sin_{K/(N-1)}(t)]^{N-1}\,dt,
\qquad
\frac{m(B_r)}{m(B_R)}\ge \frac{S_{K,N}(r)}{S_{K,N}(R)},
\]
which is the usual model-volume formula [1506.03279].

The proof uses a localized Brunn–Minkowski argument applied to thin annuli, a differential relation \(v'(r)=s(r)\), and an inequality of the form
\[
\frac{d}{dr}\bigl[\ln s(r)\bigr]
\ge
\frac{d}{dr}\bigl[(N-1)\ln \mathfrak s_{k_x/(N-1)}(r)\bigr],
\]
followed by two integrations [1506.03279]. This identifies Bishop–Gromov comparison as a direct consequence of the synthetic curvature-dimension condition rather than a specifically smooth Riccati computation.

Ketterer’s formulation is accompanied by structural stability: the \(\mathrm{CD}(k,N)\) condition is stable with respect to measured Gromov–Hausdorff convergence, and stable with respect to tensorization of finitely many metric measure spaces provided a non-branching condition is assumed [1506.03279]. A plausible implication is that Bishop–Gromov comparison in this variable-curvature setting is robust under the same limiting processes.

## 4. Alexandrov spaces and infinitesimal Bishop–Gromov comparison

For Alexandrov spaces, the comparison can be formulated infinitesimally. Let \((M,d)\) be a complete, locally compact, finite-dimensional length space of curvature \(\ge \kappa\), and let \(H^n\) be the \(n\)-dimensional Hausdorff measure [1003.4334]. Kuwae–Shioya define the radial expansion map \(\Phi_{p,t}\) based at \(p\in M\) on the domain
\[
W_{p,t}=\{x\neq p \mid \text{there is a minimal geodesic }[p,y]\ni x
\text{ with } r_p(x)/r_p(y)=t\}\cup\{p\},
\]
where \(r_p(x)=d(p,x)\) [1003.4334].

The infinitesimal Bishop–Gromov condition \(\mathrm{BG}(\kappa,n)\) for a metric-measure space \((M,d,\mu)\) requires that for every \(p\in M\), every \(t\in(0,1]\), and \(\mu\)-a.e. \(x\in M\) with \(r_p(x)<\pi/\sqrt\kappa\) when \(\kappa>0\),
\[
\frac{d(\Phi_{p,t\,*}\mu)}{d\mu}(x)
\ge
t\,\frac{s_\kappa(t\,r_p(x))^{n-1}}{s_\kappa(r_p(x))^{n-1}},
\]
where
\[
s_\kappa(r)=
\begin{cases}
\frac{\sin(\sqrt\kappa\,r)}{\sqrt\kappa},&\kappa>0,\\
r,&\kappa=0,\\
\frac{\sinh(\sqrt{|\kappa|}\,r)}{\sqrt{|\kappa|}},&\kappa<0
\end{cases}
\]
[1003.4334].

The main theorem is that if \(M\) is an \(n\)-dimensional Alexandrov space of curvature \(\ge\kappa\), then \(H^n\) satisfies \(\mathrm{BG}(\kappa,n)\) [1003.4334]. The proof passes through the \(C^\infty\)-regular part \(M^*=M\setminus S_{\delta_n}\), approximate differentiability of \(\Phi_{p,t}\), and a Jacobian estimate for the approximate differential:
\[
|\det(\ap d\Phi_{p,t}|_x)|
\le
t\,\frac{s_\kappa(t\,r_p(x))^{n-1}}{s_\kappa(r_p(x))^{n-1}}
\]
[1003.4334]. Applying the area formula then yields the measure contraction inequality.

From this infinitesimal form one recovers the standard global monotonicity. If
\[
V(r)=H^n(B_r(p)),\qquad
W(r)=v_{\kappa,n}(r)=\omega_{n-1}\int_0^r s_\kappa(t)^{n-1}\,dt,
\]
then
\[
\frac{V(R)}{W(R)}\le \frac{V(r)}{W(r)},\qquad 0<r<R<\pi/\sqrt\kappa
\]
for \(\kappa>0\), and similarly for \(\kappa\le0\) [1003.4334]. The same source also writes this equivalently as
\[
\frac{d}{dr}\ln\!\Bigl(\frac{V(r)}{W(r)}\Bigr)\ge 0,
\]
while its concluding summary states that
\[
r\mapsto \frac{H^n(B_r(p))}{v_{\kappa,n}(r)}
\quad\text{is nonincreasing}
\]
[1003.4334]. The monotonicity formulation is the substantive comparison statement.

The Alexandrov result also clarifies a common misconception: smoothness is not essential to Bishop–Gromov comparison. The proof uses approximate differentiability and measure-theoretic Jacobians rather than classical Jacobi fields everywhere.

## 5. Extensions beyond the classical smooth Riemannian setting

Bishop–Gromov comparison has been adapted to several geometric categories in which neither the classical Laplacian comparison nor the standard geodesic-ball picture is directly available.

### Weighted and Finsler frameworks

In weighted Finsler geometry with \(\epsilon\)-range, Lu–Minguzzi–Ohta formulate a volume comparison for a forward-complete weighted Finsler manifold \((M,F,m)\) under a radial lower bound on the weighted Ricci curvature \(\mathrm{Ric}_N\) and a two-sided control on the weight function along geodesics [2007.00219]. The reference volume is
\[
V_{K,N}^{\epsilon}(r)=\int_0^r e^{-c\,\psi(\dot\gamma(t))}\,[s_{K/(N-1)}(t)]^{N-1}\,dt,
\]
and the ratio
\[
R\longmapsto \frac{m(B_p(R))}{V_{K,N}^{\epsilon}(R)}
\]
is nonincreasing up to the first conjugate-point radius, or to \(\infty\) when \(K\le0\) [2007.00219]. In the special case \(\epsilon=1\), \(N=n\), one recovers the classical Bishop–Gromov inequality [2007.00219].

In Lorentz–Finsler geometry, ordinary geodesic balls are replaced by standard comparison sets for Lorentzian volumes (SCLVs), because unit-speed future-directed timelike vectors form a non-compact set [2111.10977]. Under the assumptions that \(\mathrm{Ric}_N(v)\ge K\) for all unit timelike radial \(v\in \bar U_x\) and that the cut-time \(t_x(v)\) is constant \(=T>0\) on \(\bar U_x\), Lu proves that
\[
\frac{\Vol_\mu(U_x(r))}{\int_0^r[S_{K/N}(t)]^N\,dt}
\]
is non-increasing in \(r\) for \(0<r<R<T\) [2111.10977]. The proof uses Jacobi tensor fields, the quantity
\[
h(t)=e^{-\psi(t)/N}[\det A(t)]^{1/N},
\]
and the differential inequality
\[
h''(t)+(K/N)h(t)\le 0
\]
followed by Sturm comparison [2111.10977].

Weighted Riemannian generalizations also occur for modified \(m\)-Bakry–Émery tensors with \(m\le1\). In that setting, under a pointwise lower bound on \(\Ric_{m,n}(\mathcal A_V)\) expressed in terms of a reparameterized radial variable \(s_p(r)\), one obtains annular comparisons and, when the weight is radially symmetric, monotonicity of
\[
r\longmapsto \frac{\mu_V(B(p,r))}{V_p(K,r)}
\]
[2111.15508].

### Subriemannian and extended-tensor settings

In three-dimensional Sasakian subriemannian geometry, Agrachev–Lee prove a Bishop comparison theorem with respect to the Popp-type volume \(n\). If the Tanaka–Webster scalar curvature satisfies \(k_{\min}\le k(x')\) on \(B(x,R)\), and \(k_{\min}\ge0\) or \(R<2\pi/\sqrt{-k_{\min}}\) when \(k_{\min}<0\), then
\[
\Vol_n B(x,R)\le \Vol_{n_{k_{\min}}}B_{k_{\min}}(R)
\]
[1105.2206]. The proof is based on Hamiltonian exponential maps, a canonical Darboux frame, and a matrix Riccati ODE for the linearized flow [1105.2206]. This is a direct subriemannian analogue of the classical Riccati-based Bishop argument.

For manifolds carrying a self-adjoint elliptic Codazzi tensor \(A\), the comparison can be rewritten for the \(A\)-weighted volume
\[
\Vol_A(B(x_0,R))
=
\int_{B(x_0,R)}\langle A\nabla r,\nabla r\rangle\,d\Vol_g.
\]
Under a lower bound on the extended Ricci tensor \(\Ric_{\mathrm{tr}}\) and auxiliary bounds on \(f_A\) and \(\mathrm{tr}\,A\), the ratio
\[
\Phi(R)=\frac{\Vol_A(B(x_0,R))}{V_{H,m}(R)}
\]
is nonincreasing in \(R\) [1811.11574]. Equality forces \(A\equiv \delta\,\mathrm{Id}\), \(f_A\equiv0\), and the model-space geometry [1811.11574].

These extensions indicate that the essential structure of Bishop–Gromov comparison is not tied to the usual Laplacian alone. The role of “Ricci lower bound” may be played by weighted Ricci, a subriemannian curvature invariant such as \(R_{11}\), or an extended Ricci tensor, provided one can derive an appropriate Riccati or Laplacian comparison.

## 6. Rigidity, sharpness, and enhanced or spectral variants

Rigidity is a persistent feature of Bishop–Gromov comparison. In the variable-curvature \(\mathrm{CD}(k,N)\) setting, if \(k\equiv K\) is constant and equality holds for all \(0<r<R\), then \(X\) is isometric, up to measure zero, to the classical model space of constant Ricci \(=K\) and dimension \(N\) [1506.03279]. In the variable-\(k\) case, the coefficient is sharp in the sense that one cannot replace \(k_x(r)\) by any larger lower-bound function, and equality forces radial splitting into a weighted one-dimensional model with curvature \(k_x(r)\) [1506.03279].

In the classical smooth setting, equality of normalized volume ratios at two radii forces the intermediate Laplacian and Jacobi inequalities to be equalities and identifies the corresponding ball with a model ball [2404.09792]. Similar rigidity statements appear in the subriemannian, weighted, and extended-tensor settings [1105.2206; 1811.11574].

Recent work has also produced comparison principles that are stronger than the classical pointwise Ricci formulation, or different in nature.

Brown–Freedman’s enhanced Bishop–Gromov theorem for homogeneous spaces keeps the full Ricci spectrum instead of only the minimal Ricci eigenvalue. If \((M^d,g)\) is homogeneous and \(\kappa(\omega)=\Ric_p(\omega,\omega)\) for \(\omega\in S^{d-1}\subset T_pM\), then
\[
\Vol(B_p(r))
\le
\int_0^r d\tau \int_{S^{d-1}}
[\sn(\kappa(\omega)/(d-1),\tau)]^{d-1}\,d\Omega(\omega)
\]
[2209.09288]. Classical Bishop–Gromov uses only \(\min_\omega\kappa(\omega)\); the enhanced form uses the full spectrum and coincides with the classical bound only in the Einstein case [2209.09288].

A distinct long-time refinement uses shear. Brown–Freedman identify the omission of the shear term \(\sigma^2\) in the Raychaudhuri equation as a major reason the classical upper bound is often not tight [2301.07812]. Under homogeneity and nonpositive sectional curvature, they derive a bound on the time-averaged expansion \(\langle\theta\rangle\) involving higher curvature invariants \(W^2\) and \((W')^2\), thereby obtaining a strictly smaller late-time exponent unless \(W\equiv0\) [2301.07812]. This does not replace the classical theorem; rather, it refines its asymptotic content in settings where shear can be controlled.

Another direction is spectral generalization. Antonelli–Xu show that if a closed Riemannian manifold \((M^n,g)\), \(n\ge3\), satisfies
\[
\lambda_1\!\Bigl(-\frac{n-1}{n-2}\Delta+\Ric\Bigr)\ge n-1,
\]
then
\[
\Vol(M)\le \Vol(\mathbb S^n),
\]
with rigidity to the round sphere in the equality case [2405.08918]. Their proof uses an unequally weighted isoperimetric profile satisfying the viscosity inequality
\[
I''(v)\,I(v)-\frac{(I'(v))^2}{n-1}-(n-1)\ge0
\]
and an ODE comparison against the sphere profile [2405.08918]. This suggests a “spectral Bishop–Gromov” principle in which the pointwise Ricci lower bound is replaced by a lower bound on the first eigenvalue of a Schrödinger operator.

Weighted spectral analogues also exist for the \(\infty\)-Bakry–Émery Ricci tensor. Under
\[
\lambda_1(L_f)\ge (n-1)K,\qquad L_f=-\Delta_f+\Ric_f,
\]
one has
\[
\frac{V_f(B(p,R))}{V_f(B(p,r))}
\le
\exp\!\Bigl(\frac{2}{n-1}\bigl(\sup_{B(p,R)}f-\inf_{B(p,r)}f\bigr)\Bigr)
\frac{V_{K,n}(R)}{V_{K,n}(r)}
\]
for the weighted volume \(V_f(B(p,r))=\int_{B(p,r)}e^{-f}\,d\vol_g\) [2509.23182]. The oscillation factor in \(f\) is intrinsic to the weighted setting.

## 7. Consequences, applications, and contemporary directions

Bishop–Gromov comparison has immediate consequences in global geometry. In the classical \(\kappa=0\) case it yields Bishop’s inequality \(\Vol(B_p(r))\le \omega_n r^n\), while for \(\kappa>0\) it implies Bonnet–Myers compactness and \(\diam M\le \pi/\sqrt\kappa\) [2404.09792]. Positive Ricci curvature then gives finiteness of the fundamental group [2404.09792]. These implications recur in synthetic, weighted, and spectral settings [1506.03279; 2405.08918; 2509.23182].

In nonsmooth spaces with integrable curvature deficit, recent localization methods extend the comparison beyond exact lower Ricci bounds. For essentially non-branching \({\sf CD}(k,N)\) spaces, with \(p>N/2\) and \(L^p\)-Ricci deficit
\[
\rho_p^k(E,K)=\int_E |(k-K)\wedge0|^p\,d\mathfrak m,
\]
one obtains a quantitative Bishop–Gromov estimate:
\[
\left(\frac{\mathfrak m(B_R(x)\cap T)}{v_{K,N}(R)}\right)^{1/(2p-1)}
-
\left(\frac{\mathfrak m(B_r(x)\cap T)}{v_{K,N}(r)}\right)^{1/(2p-1)}
\le
C_{K,N,p}(R)\,[\rho_p^k(T,K)]^{1/(2p-1)}
\]
for star-shaped sets \(T\) [2509.22514]. When the deficit vanishes, one recovers the classical monotonicity [2509.22514]. This indicates that Bishop–Gromov comparison is stable not only under exact lower bounds but also under sufficiently integrable deviations from them.

The comparison has also entered nonlocal analysis. In abstract metric measure spaces satisfying the generalized Bishop–Gromov inequality
\[
\frac{\mu(B_R(x))}{\mu(B_r(x))}
\le
\frac{V_{K,N}(R)}{V_{K,N}(r)},
\]
together with finite asymptotic volume ratio, Di Marino–Spector–Valdinoci derive Maz’ya–Shaposhnikova type asymptotic formulas for nonlocal \(p\)-energies:
\[
\lim_{s\to0^+} s\,\mathcal E_s(u)
=
2\,\mathrm{AVR}_{K,N}(X)\,\|u\|_{L^p(X,\mu)}^p
\]
[2402.11174]. The proof uses a near/far decomposition of the kernel integral, with Bishop–Gromov entering in the control of tail integrals [2402.11174]. This shows that volume comparison is not only a geometric conclusion from curvature bounds but also an analytical hypothesis with direct functional consequences.

A further direction replaces static comparison by evolving comparison under Ricci flow. Tian–Zhang prove a relative volume comparison estimate along Ricci flow,
\[
\frac{\Vol_{g_1}(B_{g_1}(x,r))}{r^m}
\ge
K(m,A)\,
\frac{\Vol_{g_0}(B_{g_0}(x_0,r_0))}{r_0^m},
\]
under a space-time Ricci lower bound, a final-time scalar curvature slice bound, and appropriate inclusion assumptions [1802.09506]. They explicitly describe this as an analogue of the Bishop–Gromov volume comparison for Ricci flow, recovering Bishop–Gromov in the static limit \(g(t)\equiv g\) with \(\Ric\ge0\) [1802.09506].

Across these settings, the central invariant is always a normalized radial volume density or volume ratio whose monotonicity encodes curvature information. The normalization may come from a space form, a variable-curvature Jacobi ODE, a weighted profile, an SCLV model, a synthetic one-dimensional localization density, or a spectral isoperimetric ODE. This suggests that “Bishop–Gromov comparison” is best understood not as a single theorem but as a general monotonicity paradigm linking lower curvature information to radial measure growth.

Source: https://www.emergentmind.com/topics/bishop-gromov-comparison