Lorenz Curves: Distribution & Inequality Metrics

• Lorenz curves are normalized cumulative-share functions that depict how a nonnegative resource is distributed across a population with inherent convexity, continuity, and normalization properties.
• They underpin measures such as the Gini coefficient and serve as the basis for multivariate, quantum, and dynamic extensions in both economics and other disciplines.
• The methodology includes statistical estimation, robust forecasting via dynamical systems, and practical applications ranging from welfare analysis to beam-quality assessment in optics.

Lorenz curves are normalized cumulative-share functions that represent how a nonnegative resource is distributed across a population ordered from lowest to highest values. For a distribution function $F$ with positive finite mean $\mu_F=\int_0^1 F^{-1}(u)\,du$, the classical Lorenz curve is

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$

and it is continuous, convex, increasing, and normalized by $L_F(0)=0$ and $L_F(1)=1$. In the discrete case, if incomes are ordered as $x_{(1)}\le \cdots \le x_{(n)}$, then $p_i=i/n$ and $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$ define the polygonal Lorenz curve through $(p_i,q_i)$. Within this geometry, Lorenz curves underlie the Gini coefficient, Lorenz dominance, generalized Lorenz comparisons, and a wide range of newer constructions in multivariate analysis, quantum information, transport theory, and dynamical systems (Ignatov et al., 2024, Schlemmer, 2021, Buscemi et al., 2016, Cohen, 24 Jul 2025).

1. Classical formulation and geometric structure

In the standard univariate setting, the Lorenz curve records the cumulative share of total income or wealth held by the bottom $p$-fraction of the population. The equality benchmark is the 45-degree line,

$\mu_F=\int_0^1 F^{-1}(u)\,du$0

which corresponds to perfect equality: the bottom $\mu_F=\int_0^1 F^{-1}(u)\,du$1 share of the population receives exactly the bottom $\mu_F=\int_0^1 F^{-1}(u)\,du$2 share of total income. The farther the Lorenz curve lies below this line, the more unequal the distribution (Schlemmer, 2021, Jacquemain, 2017).

Equivalent representations recur throughout the literature. For a random variable $\mu_F=\int_0^1 F^{-1}(u)\,du$3 with cdf $\mu_F=\int_0^1 F^{-1}(u)\,du$4, mean $\mu_F=\int_0^1 F^{-1}(u)\,du$5, and quantile function $\mu_F=\int_0^1 F^{-1}(u)\,du$6, one has

$\mu_F=\int_0^1 F^{-1}(u)\,du$7

and also

$\mu_F=\int_0^1 F^{-1}(u)\,du$8

In finite populations or grouped data, the empirical Lorenz curve is typically obtained by linear interpolation of cumulative population shares against cumulative resource shares (Jacquemain, 2017, Conti et al., 2018).

A closely related object is the generalized Lorenz curve,

$\mu_F=\int_0^1 F^{-1}(u)\,du$9

where $$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$0. The ordinary Lorenz curve is its mean-normalized version,

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$1

This distinction is important because the ordinary Lorenz curve encodes relative concentration, whereas the generalized Lorenz curve retains the effect of the mean and is therefore central in welfare comparisons (Ratnasingam et al., 2023, Ratnasingam et al., 2023).

2. Inequality ordering, scalar indices, and shape-sensitive refinements

The Gini coefficient is the most widely used scalar summary derived from the Lorenz curve. In continuous form,

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$2

and in the discrete ordered-sample formulation discussed in inequality measurement,

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$3

The Lorenz order is likewise defined pointwise: for Lorenz curves $$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$4,

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$5

Under this order, one distribution is more equitable than another when its Lorenz curve lies nowhere below the other (Schlemmer, 2021, Baíllo et al., 2021).

A central limitation of the ordinary Gini is that it aggregates all vertical deviations from equality with equal weight. Two Lorenz curves can have the same area from the equality line and thus the same $$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$6, while having very different shapes, with one more left-skewed and another more right-skewed. To address this, an asymmetry-sensitive extension defines

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$7

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$8

and then the skewness-adjusted Gini

$$L_F(x)=\frac{1}{\mu_F}\int_0^x F^{-1}(u)\,du,\qquad 0\le x\le 1,$$9

Because $L_F(0)=0$0, this can be rewritten as

$L_F(0)=0$1

The construction is calibrated so that $L_F(0)=0$2 when $L_F(0)=0$3, and the paper states that $L_F(0)=0$4 inherits scale invariance, population invariance, the Pigou–Dalton principle of transfers, the ability to accommodate zero and negative values, and a weaker form of decomposability from the Gini framework (Schlemmer, 2021).

The geometry of fixed-Gini Lorenz curves can also be analyzed directly. For a fixed Gini value $L_F(0)=0$5, the set of Lorenz curves with $L_F(0)=0$6 is compact and convex in $L_F(0)=0$7, and its extreme points are explicit piecewise affine curves. This permits an exact characterization of the maximal $L_F(0)=0$8-distance between Lorenz curves with prescribed Gini coefficients: $L_F(0)=0$9 When two Lorenz curves have equal Gini $L_F(1)=1$0, the maximal possible distance is

$L_F(1)=1$1

and the overall maximum occurs at $L_F(1)=1$2, giving

$L_F(1)=1$3

This shows that equality of Gini coefficients does not determine Lorenz-curve shape (Baíllo et al., 2021).

A different line of work constructs discrete empirical Lorenz curves $L_F(1)=1$4 from a Gini-stable recursion on ordered normalized vectors and proves that, as $L_F(1)=1$5, they converge to

$L_F(1)=1$6

The limiting family coincides exactly with the Lorenz curves of finite-mean Pickands generalized Pareto distributions under a direct Gini-based parametrization (Bertoli-Barsotti et al., 2023).

3. Estimation, inference, and forecasting

Several contributions address the statistical difficulty of estimating Lorenz-type objects when moments are unstable, data are sparse, or sampling is complex. One robust alternative replaces means by quantiles. For positive distributions $L_F(1)=1$7, three quantile-based Lorenz-type curves are defined by

$L_F(1)=1$8

$L_F(1)=1$9

with associated inequality coefficients

$x_{(1)}\le \cdots \le x_{(n)}$0

These quantities are defined for all positive income distributions, even when the mean does not exist, and their influence functions are bounded under smoothness assumptions, unlike the classical Lorenz curve and Gini, whose moment-based estimators are sensitive to outliers and heavy tails (Prendergast et al., 2015).

When only sparse summary information is available, a simple parametric reconstruction uses

$x_{(1)}\le \cdots \le x_{(n)}$1

for which

$x_{(1)}\le \cdots \le x_{(n)}$2

If bottom and top income shares at level $x_{(1)}\le \cdots \le x_{(n)}$3 are observed, with $x_{(1)}\le \cdots \le x_{(n)}$4, then $x_{(1)}\le \cdots \le x_{(n)}$5 is available in closed form through

$x_{(1)}\le \cdots \le x_{(n)}$6

where $x_{(1)}\le \cdots \le x_{(n)}$7, $x_{(1)}\le \cdots \le x_{(n)}$8, $x_{(1)}\le \cdots \le x_{(n)}$9, and $p_i=i/n$0. This avoids numerical error minimization when only the Gini index and a small number of grouped shares are available (Sitthiyot et al., 2021).

Under complex unequal-probability survey designs, a design-based Hájek estimator of the cdf is used as the plug-in basis for Lorenz inference: $p_i=i/n$1 From this one defines $p_i=i/n$2, $p_i=i/n$3, and

$p_i=i/n$4

The associated process $p_i=i/n$5 has a functional Gaussian limit, and a pseudo-population resampling scheme consistently approximates that law, enabling confidence bands for the Lorenz curve, confidence intervals for the Gini concentration ratio, and tests of Lorenz dominance (Conti et al., 2018).

For generalized Lorenz ordinates, modified empirical-likelihood methods address convex-hull failure and finite-sample undercoverage. The paper develops adjusted empirical likelihood, transformed empirical likelihood, and transformed adjusted empirical likelihood for

$p_i=i/n$6

and proves that the scaled log-likelihood ratios converge to $p_i=i/n$7. A distinct development treats the Lorenz curve itself as an errors-in-variables curve because both the cumulative population share and the cumulative income share are estimated with error. In that setting, simultaneous confidence bands are constructed as unions of confidence ellipses around the estimated planar curve, with calibration based on the supremum of a $p_i=i/n$8-process (Ratnasingam et al., 2023, Dong et al., 28 Jan 2025).

Lorenz curves have also been modeled directly as functional time series. For regional Italian income and wealth data, transformed Lorenz curves $p_i=i/n$9 are decomposed by a one-way functional ANOVA,

$q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$0

into a functional grand effect, a functional row effect, and residual functions. The residual functional dynamics are then forecast, bootstrap intervals are constructed, and isotonic regression is used to ensure forecast monotonicity (Shang, 6 Apr 2025).

4. Multivariate, relative, and quantum generalizations

A multivariate extension replaces scalar ranks by vector ranks from optimal transport. If $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$1 is the cyclically monotone vector quantile associated with $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$2, and $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$3, the vector Lorenz map is

$q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$4

Each component of $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$5 is the cumulative share of one resource. Pointwise comparison defines a multivariate Lorenz order, and the paper proves that this order is equivalent to preference by any social planner with inequality-averse multivariate rank-dependent social evaluation functional. It also defines a multivariate Gini index,

$q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$6

together with a family of multivariate $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$7-Gini indices and an Inverse Lorenz Function $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$8 whose level sets visualize two-dimensional inequality, including income-wealth inequality in the United States between 1989 and 2022 (Fan et al., 2022).

In quantum information theory, Lorenz curves are generalized via binary hypothesis testing. For density matrices $q_i=(n\mu)^{-1}\sum_{j=1}^i x_{(j)}$9, the testing region is

$(p_i,q_i)$0

and the quantum relative Lorenz curve is its upper boundary. The induced preorder,

$(p_i,q_i)$1

unifies classical majorization, relative majorization, thermomajorization, and noncommutative state comparison. The paper proves equivalence between this order, trace-norm inequalities

$(p_i,q_i)$2

families of Hilbert $(p_i,q_i)$3-divergences, and hypothesis-testing relative entropies (Buscemi et al., 2016).

5. Iteration, dynamics, and geometric structures on Lorenz-curve space

One recent direction treats the Lorenz transform itself as a dynamical operator. Starting from a nonnegative random variable with cdf $(p_i,q_i)$4, define $(p_i,q_i)$5 and iterate

$(p_i,q_i)$6

For any nonnegative $(p_i,q_i)$7 with finite positive mean, the iterates converge uniformly to the universal limit

$(p_i,q_i)$8

In the reflected setting based on the integrated tail transform, the limit is

$(p_i,q_i)$9

These results identify repeated Lorenzification as a nonlinear dynamical system with non-corner universal limits governed by the golden-ratio exponent (Ignatov et al., 2024).

A different dynamical formulation begins from a one-dimensional Fokker–Planck equation for a positive density $p$0. Writing

$p$1

and using $p$2, the Lorenz curve becomes $p$3. The identities

$p$4

lead to a transformed Lorenz dynamics on the compact interval $p$5: $p$6 For the heat equation this reduces to

$p$7

and for heat with Ornstein–Uhlenbeck drift it becomes

$p$8

The construction turns the Lorenz curve from a static summary into an evolving state variable for diffusion and kinetic-wealth models (Cohen et al., 2024).

An even more structural development endows the space of Lorenz curves with Wasserstein-inspired metric tensors. For positive probability densities $p$9, the Lorenz transform satisfies

$\mu_F=\int_0^1 F^{-1}(u)\,du$00

On the Lorenz side, one formal $\mu_F=\int_0^1 F^{-1}(u)\,du$01-type metric is

$\mu_F=\int_0^1 F^{-1}(u)\,du$02

while nonlinear-mobility and $\mu_F=\int_0^1 F^{-1}(u)\,du$03-type geometries induce weighted variants involving $\mu_F=\int_0^1 F^{-1}(u)\,du$04. The paper proves isometry results between the corresponding manifolds of probability measures and Lorenz curves and shows that transformed Lorenz PDEs remain gradient flows of the same underlying energies (Cohen, 24 Jul 2025).

The complementary Lorenz curve has also been used as a spectral object. If a pure power-law response yields

$\mu_F=\int_0^1 F^{-1}(u)\,du$05

then a heterogeneous system is represented by

$\mu_F=\int_0^1 F^{-1}(u)\,du$06

where $\mu_F=\int_0^1 F^{-1}(u)\,du$07 is an exponent spectrum. The local diagnostic

$\mu_F=\int_0^1 F^{-1}(u)\,du$08

is then used to quantify departures from a single exponent through measures such as $\mu_F=\int_0^1 F^{-1}(u)\,du$09 and the spectral entropy $\mu_F=\int_0^1 F^{-1}(u)\,du$10. This suggests that Lorenz-curve shape can encode microscopic heterogeneity when the underlying density departs from a pure power law (Das et al., 28 May 2026).

6. Applications, crossing phenomena, and domain-specific interpretations

Lorenz curves remain central in classical inequality analysis, but several applications use them as diagnostic devices for mechanisms that are not themselves economic. In resource dependent branching processes, the claim distribution $\mu_F=\int_0^1 F^{-1}(u)\,du$11 enters survival and extinction criteria through truncated first moments,

$\mu_F=\int_0^1 F^{-1}(u)\,du$12

which can be written in Lorenz form as

$\mu_F=\int_0^1 F^{-1}(u)\,du$13

This reformulation makes the Bruss–Duerinckx survival envelope visually interpretable: weaker inequality in claims enlarges both the certain-extinction and certain-survival regions, so equality increases predictability rather than uniformly favoring survival (Jacquemain, 2017).

In optics, the Lorenz curve of a light beam is constructed from the discretized joint near-field/far-field intensity distribution

$\mu_F=\int_0^1 F^{-1}(u)\,du$14

sorted in decreasing order and cumulatively summed. If one beam’s Lorenz curve lies everywhere above another’s, then the first majorizes the second and all Schur-concave measures of spreading are smaller. In particular, for Rényi entropies,

$\mu_F=\int_0^1 F^{-1}(u)\,du$15

and thus all entropic beam-width products are smaller. When Lorenz curves intersect, however, there is no universal ordering: different valid beam-quality criteria can disagree, and the Lorenz plot makes that criterion dependence explicit (Porras et al., 2017).

A recurring misconception is that a single scalar inequality index or a single width criterion fully determines comparative structure. Published work on inequality shows that distributions with the same ordinary Gini can differ materially in Lorenz-curve asymmetry and tail behavior, while published work on beam quality shows that intersecting Lorenz curves preclude universal ranking across all Schur-concave spread measures (Schlemmer, 2021, Porras et al., 2017). More generally, the literature repeatedly treats the Lorenz curve not merely as a picture but as a functional object: one that can be estimated under complex sampling, forecast over time, generalized to multivariate and quantum settings, iterated as an operator, and embedded in transport-inspired geometries.