“The test of our progress is not whether we add more to the abundance of those who have much; it is whether we provide enough for those who have too little.” – Franklin D. Roosevelt, Second Inaugural Address (1937)
RUNNING EXAMPLE – MARIA’S COFFEE SHOP
Maria and Richard Crane, the CEO of Meridian Capital Partners REIT, both buy a cup of coffee. The coffee costs $5. The transaction is identical for both parties on $d_1$: a five-dollar displacement along the monetary dimension. The barista does not charge Maria less because she earns less or charge Crane more because he earns more. The price is a scalar. It treats them as identical.
But $5 is not the same displacement on their respective manifolds.
For Maria, $5 is approximately 1.5% of her daily gross margin. The coffee is a meaningful cost that she tracks in a mental ledger of daily discretionary spending. The $5 activates $d_1$ (she feels the expenditure), $d_7$ (the coffee is a small ritual that affirms her identity as someone who appreciates quality), and $d_9$ (she is aware of the trade-off between this coffee and other uses of the money). The Mahalanobis distance from her current state to her post-coffee state is:
$$w_{\text{Maria}} = \frac{5^2}{\sigma_{1,\text{Maria}}^2} + \frac{\delta_7^2}{\sigma_{7,\text{Maria}}^2} + \frac{\delta_9^2}{\sigma_{9,\text{Maria}}^2} + \text{cross terms}$$
For Crane, $5 is approximately 0.0003% of his daily income. The coffee is below his perceptual threshold -- a purchase so small relative to his monetary scale that his System 1 does not register it as an economic event. The $5 activates $d_1$ negligibly (no felt expenditure), $d_7$ slightly (the coffee confirms his identity as a person of taste), and $d_9$ not at all (no perceived trade-off). The Mahalanobis distance is:
$$w_{\text{Crane}} = \frac{5^2}{\sigma_{1,\text{Crane}}^2} + \frac{\delta_7'^2}{\sigma_{7,\text{Crane}}^2} \approx \frac{5^2}{\sigma_{1,\text{Crane}}^2}$$
Since Crane’s monetary variance $\sigma_{1,\text{Crane}}^2$ is vastly larger than Maria’s $\sigma_{1,\text{Maria}}^2$ (his normal daily monetary fluctuations span thousands of dollars, while hers span hundreds), the Mahalanobis distance $w_{\text{Crane}} \ll w_{\text{Maria}}$. The same $5 coffee is a larger displacement on Maria’s manifold than on Crane’s.
This is not diminishing marginal utility – the claim that each additional dollar provides less satisfaction. It is a stronger and more precise claim: Maria and Crane experience different metrics on the same underlying manifold. A dollar is not “worth less” to Crane in some abstract utility sense; it covers less distance on his manifold. The geometry of their economic experience is different.
Economic inequality is among the oldest and most consequential facts of human society. It has been measured, debated, decried, and defended for centuries. The tools of measurement are well-established: the Gini coefficient, the Lorenz curve, income decile ratios, wealth percentile rankings. The theoretical frameworks are well-developed: welfare economics, optimal taxation, redistribution theory.
Yet the standard frameworks share a common limitation: they measure inequality on the scalar projection. The Gini coefficient measures the distribution of $d_1$ (income or wealth) across a population. The Lorenz curve plots the cumulative share of $d_1$ going to each percentile. Income ratios compare $d_1$ values at different points in the distribution. All of these are $d_1$-only measures applied to a nine-dimensional reality.
This chapter develops a geometric theory of inequality. The central claim is that inequality is not merely an uneven distribution of a scalar quantity. It is a metric distortion – different agents experience different metrics on the same economic manifold, and the difference in metrics is what makes inequality economically consequential. A dollar has a different geometric meaning for different agents, not because of diminishing marginal utility (a psychological hypothesis about the relationship between wealth and satisfaction) but because the metric tensor itself – the mathematical structure that assigns costs to displacements – varies with position on the manifold.
Chapter 4 established that the economic metric is position-dependent: the cost of a displacement depends on the agent’s current location on the manifold. This is not a special feature of the geometric framework; it is a standard property of Riemannian manifolds. Just as the metric of spacetime varies with position (stronger near massive objects, flatter far away), the economic metric varies with the agent’s endowment, social position, and history.
The key insight for inequality is that wealth affects the metric. An agent’s monetary variance $\sigma_1^2$ – the normal range of monetary fluctuation in their daily life – is a function of their wealth. A person with $1 million in liquid assets experiences daily monetary fluctuations of perhaps $\pm$5,000 from investment returns, spending, and income. For this person, $\sigma_1^2 \approx (5{,}000)^2$. A person living paycheck to paycheck experiences daily monetary fluctuations of perhaps $\pm$50 from variable expenses and income timing. For this person, $\sigma_1^2 \approx (50)^2$.
The metric component $g_{11} = 1/\sigma_1^2$ is therefore inversely proportional to the square of the monetary scale. For the wealthy agent, $g_{11} \approx 1/(5{,}000)^2 = 4 \times 10^{-8}$. For the poor agent, $g_{11} \approx 1/(50)^2 = 4 \times 10^{-4}$. A dollar displacement has metric cost:
$$\text{cost} = g_{11} \times (\$1)^2 = \frac{1}{\sigma_1^2}$$
For the poor agent, this cost is 10,000 times larger than for the wealthy agent. The metric assigns different costs to the same dollar displacement depending on the agent’s position on the manifold. This is not a subjective preference statement; it is a geometric fact about the structure of the manifold.
Proposition 40 (The Inequality Metric). The economic metric experienced by agent $i$ with monetary scale $\sigma_{1,i}$ assigns a cost to a $\$D$ displacement of:
$$w_i(D) = \frac{D^2}{\sigma_{1,i}^2}$$
For two agents with monetary scales $\sigma_{1,A}$ and $\sigma_{1,B}$, the ratio of metric costs for the same dollar displacement is:
$$\frac{w_A(D)}{w_B(D)} = \frac{\sigma_{1,B}^2}{\sigma_{1,A}^2}$$
The agent with the smaller monetary scale (lower wealth) experiences a proportionally larger metric cost for any given dollar displacement.
The standard economic account of “a dollar is worth more to a poor person” invokes diminishing marginal utility: the utility function $u(w)$ is concave, so $u'(w_{\text{poor}}) > u'(w_{\text{rich}})$. This is correct as far as it goes, but it is a statement about a scalar function on a one-dimensional space. It captures only the $d_1$ component of a nine-dimensional phenomenon.
The geometric account is richer in three ways.
First, the metric distortion operates on all nine dimensions, not just $d_1$. Wealth affects not only the monetary metric $g_{11}$ but also the fairness metric $g_{33}$, the autonomy metric $g_{44}$, and the epistemic metric $g_{99}$. A poor person is more sensitive to fairness violations ($g_{33}$ is larger) because unfairness at low wealth can be existentially threatening, while unfairness at high wealth is merely irritating. A poor person has less autonomy ($g_{44}$ is larger, meaning autonomy displacements are more costly) because economic constraints narrow the option set. A poor person faces greater epistemic challenges ($g_{99}$ is larger) because financial complexity imposes information costs that scale with the sophistication of the financial system, not with the agent’s resources to navigate it.
Second, the metric distortion includes cross-dimensional couplings that the scalar account cannot capture. For a poor person, the coupling between $d_1$ (money) and $d_4$ (autonomy) is tight: losing $100 may mean losing the ability to choose between transportation options, food options, or medical options. The off-diagonal component $g_{14}$ is large. For a wealthy person, this coupling is loose: losing $100 has no impact on autonomy. The metric is not just different in magnitude; it is different in structure.
Third, the metric distortion has boundary consequences. A poor person’s manifold has moral-economic boundaries that a wealthy person’s manifold does not. The boundary between “able to pay rent” and “evicted” is a discontinuity on the poor person’s manifold – a region where a small monetary displacement produces an infinite change in living circumstances. The wealthy person’s manifold has no such boundary at the same dollar amount. The boundary structure of the manifold – the phase transitions between economic regimes – is wealth-dependent.
RUNNING EXAMPLE – MARIA’S COFFEE SHOP
The metric distortion between Maria and Crane extends far beyond the $5 coffee.
Maria’s manifold has a boundary at approximately $-\$3{,}000$ from her current state: if her monthly margin drops by more than $3,000, she cannot make rent, and the business fails. This boundary creates a *cliff* on her manifold -- a region of discontinuously infinite metric cost. Every decision Maria makes is conditioned by the proximity of this boundary. She cannot take risks that would move her close to the cliff, even if the expected $d_1$ payoff is positive, because the boundary penalty is infinite.
Crane’s manifold has no corresponding boundary. A $3,000 loss is 0.002% of his annual compensation. His nearest equivalent boundary – the point at which the REIT faces a liquidity crisis – is tens of millions of dollars away from his personal financial position. His decisions are unconstrained by survival boundaries.
The consequence: Maria’s option set is geometrically smaller than Crane’s. She is confined to a narrow corridor on the manifold by the proximity of the boundary. Crane can explore a vast region. The effective dimension of Maria’s decision manifold is lower than Crane’s – not because she has fewer aspirations but because the boundaries on her manifold constrain her movement to a lower-dimensional subspace.
This is the geometry of poverty: not just having less, but having a manifold with less room to move.
If inequality is a metric distortion – a position-dependent variation in the metric tensor – then it is a tensorial quantity, not a scalar one. We define:
Definition 41 (Inequality Tensor). The inequality tensor $\mathcal{I}$ for a population of $N$ agents is the variance of the metric tensor across agents:
$$\mathcal{I}_{\mu\nu} = \text{Var}_{i=1}^N\left[g_{\mu\nu}^{(i)}\right] = \frac{1}{N}\sum_{i=1}^N \left(g_{\mu\nu}^{(i)} - \bar{g}_{\mu\nu}\right)^2$$
where $g_{\mu\nu}^{(i)}$ is the metric tensor experienced by agent $i$ and $\bar{g}_{\mu\nu}$ is the population-average metric.
The inequality tensor is a $9 \times 9$ symmetric matrix. Its diagonal entries $\mathcal{I}_{kk}$ measure the inequality along each dimension: $\mathcal{I}_{11}$ is the variance of the monetary metric (income/wealth inequality), $\mathcal{I}_{33}$ is the variance of the fairness metric (unequal exposure to unfairness), $\mathcal{I}_{55}$ is the variance of the trust metric (unequal access to trusted relationships and institutions). Its off-diagonal entries $\mathcal{I}_{jk}$ measure the cross-dimensional inequality: how the coupling between dimensions varies across the population.
The Gini coefficient is the most widely used scalar measure of income inequality. It ranges from 0 (perfect equality) to 1 (perfect inequality: one person has all the income). The geometric framework reveals it as a specific contraction of the inequality tensor.
Proposition 42 (Gini as Contraction). The Gini coefficient $G$ is a scalar contraction of the inequality tensor restricted to the monetary dimension:
$$G \propto \sqrt{\mathcal{I}_{11}}$$
Specifically, $G$ is the normalized mean absolute difference in $d_1$, which is a monotone function of the variance of $g_{11}^{(i)}$ across agents. The Gini discards all components of $\mathcal{I}$ except $\mathcal{I}_{11}$.
The Scalar Irrecoverability Theorem (Chapter 6) applies directly: the inequality information destroyed by the Gini contraction is mathematically irrecoverable from the Gini alone. Two societies with identical Gini coefficients can have radically different inequality tensors:
Society A: Gini = 0.40. High income inequality ($\mathcal{I}_{11}$ large), but strong social infrastructure: universal healthcare reduces $\mathcal{I}_{44}$ (autonomy inequality is low – both rich and poor have medical autonomy), robust public institutions reduce $\mathcal{I}_{88}$ (legitimacy inequality is low – the legal system treats all agents similarly), and a strong social safety net creates no survival boundaries near the lower end of the income distribution.
Society B: Gini = 0.40. Identical income inequality, but weak social infrastructure: no universal healthcare (high $\mathcal{I}_{44}$ – the wealthy have medical autonomy, the poor do not), corrupt institutions (high $\mathcal{I}_{88}$ – the wealthy can access the legal system, the poor cannot), and no safety net (survival boundaries close to the lower end of the distribution, creating cliff-like metric structure for low-income agents).
The Gini coefficient cannot distinguish Society A from Society B. The inequality tensor can. The policy implications are fundamentally different: Society A may need progressive taxation to address $\mathcal{I}_{11}$; Society B needs institutional reform to address $\mathcal{I}_{44}$, $\mathcal{I}_{88}$, and the boundary structure.
The argument extends to every scalar inequality measure – not just the Gini. The Theil index, the Atkinson index, the Palma ratio, the income decile ratios – all of these are projections of the inequality tensor onto $d_1$ (or at best onto a small subset of dimensions). By the Scalar Irrecoverability Theorem, they all destroy eight dimensions of inequality information.
This is not a criticism of the measures themselves; it is a structural limitation of scalar measurement applied to a tensorial quantity. The practical implication is that no single inequality number – however cleverly constructed – can serve as a sufficient statistic for the multi-dimensional phenomenon of inequality. Policy that targets a scalar inequality measure may reduce $\mathcal{I}_{11}$ while leaving $\mathcal{I}_{33}$, $\mathcal{I}_{44}$, $\mathcal{I}_{55}$, and $\mathcal{I}_{88}$ unchanged or even increased.
John Rawls’s A Theory of Justice (1971) introduced the most influential thought experiment in modern political philosophy: the veil of ignorance. Behind the veil, rational agents choose the principles of justice for their society without knowing their own position – their wealth, talents, social status, or conception of the good. Rawls argued that rational agents behind the veil would choose the maximin principle: the just society maximizes the welfare of the worst-off member.
The geometric framework provides a precise mathematical interpretation of the veil of ignorance.
Proposition 43 (Veil of Ignorance as Metric Averaging). Rawls’s veil of ignorance is the operation of replacing each agent’s position-dependent metric $g_{\mu\nu}^{(i)}$ with the population-average metric $\bar{g}_{\mu\nu}$:
$$g_{\mu\nu}^{\text{veil}} = \bar{g}_{\mu\nu} = \frac{1}{N}\sum_{i=1}^N g_{\mu\nu}^{(i)}$$
Behind the veil, all agents evaluate economic displacements using the same metric – the average metric. The veil eliminates metric distortion by construction: if all agents share the same metric, $\mathcal{I}_{\mu\nu} = 0$ for all $\mu, \nu$.
This interpretation explains several features of Rawls’s theory:
Why the veil produces egalitarian preferences. Behind the veil, a $100 tax on the wealthy and a $100 transfer to the poor have the *same* metric cost to the decision-maker (who does not know whether they are wealthy or poor). Without the veil, the wealthy person's metric assigns low cost to the tax ($g_{11}$ is small) and the poor person’s metric assigns high value to the transfer ($g_{11}$ is large). The veil eliminates this asymmetry by averaging the metrics.
Why Rawls arrives at maximin. The average metric $\bar{g}_{\mu\nu}$ is dominated by the contributions from agents with the largest metric components – the worst-off, who experience the highest costs for any given displacement. To see why: if agent $j$ has $g_{11}^{(j)} = 10^{-3}$ (wealthy) and agent $k$ has $g_{11}^{(k)} = 10^{1}$ (poor), the average $\bar{g}_{11} \approx 5 \times 10^{0}$ is much closer to the poor agent’s metric than to the wealthy agent’s. Decisions made using the average metric disproportionately weight the interests of those with the largest metric components – which, by the inequality metric (Proposition 40), are the poorest agents. This is the maximin principle, derived from metric averaging rather than from risk aversion.
Why the veil is coherent. The veil is often criticized as an impossible thought experiment – you cannot actually forget your position. The geometric interpretation sidesteps this criticism: the veil is not a psychological operation (forget who you are) but a mathematical one (replace your position-dependent metric with the population average). This operation is well-defined and computable regardless of whether any agent can actually perform it psychologically. The veil is a computational tool, not a phenomenological claim.
The metric-averaging interpretation also reveals limitations of the Rawlsian framework.
Averaging destroys information. The average metric $\bar{g}_{\mu\nu}$ is a contraction of the full set of agent metrics $\{g_{\mu\nu}^{(i)}\}_{i=1}^N$. By the Scalar Irrecoverability Theorem (applied to the contraction from $N$ metrics to one average metric), the averaging operation destroys information about the distribution of metrics. Two populations with different metric distributions can have the same average metric. The Rawlsian framework, by operating on the average metric, cannot distinguish between these populations.
The average may not represent any actual agent. The average metric $\bar{g}_{\mu\nu}$ may not correspond to any individual’s actual metric. It is a mathematical construct – the “metric experienced by the average person” – that no real person inhabits. Policies optimized for the average metric may be optimal for no one.
The veil operates on the wrong manifold for some questions. Rawls’s principles concern the basic structure of society, not individual economic decisions. The metric-averaging operation is appropriate for structural questions (what institutions should we build?) but may not be appropriate for individual questions (should Maria hire another barista?). The geometric framework clarifies the scope: the veil is a tool for institutional design, where the relevant metric is the population average, not a tool for individual decision-making, where the relevant metric is the agent’s own.
RUNNING EXAMPLE – MARIA’S COFFEE SHOP
Behind the veil, the rational agent does not know whether they will be Maria, Crane, or one of Maria’s minimum-wage employees. They evaluate the economic system using the average metric:
Maria’s metric: $\sigma_{1,\text{Maria}}^2 \approx (300)^2 = 90{,}000$. Daily monetary fluctuations of ~$300.
Crane’s metric: $\sigma_{1,\text{Crane}}^2 \approx (5{,}000)^2 = 25{,}000{,}000$. Daily monetary fluctuations of ~$5,000.
Sofia’s metric (Maria’s minimum-wage barista): $\sigma_{1,\text{Sofia}}^2 \approx (50)^2 = 2{,}500$. Daily monetary fluctuations of ~$50.
The average monetary variance: $\bar{\sigma}_1^2 \approx (90{,}000 + 25{,}000{,}000 + 2{,}500)/3 \approx 8{,}364{,}167$.
But the average metric $\bar{g}_{11} = 1/\bar{\sigma}_1^2$ is dominated by Crane’s metric, making the average metric insensitive to Sofia’s experience. This is a known limitation of the mean as an average – it is pulled by outliers.
The Rawlsian maximin corrects for this: instead of the average metric, use the metric of the worst-off agent (Sofia). Under Sofia’s metric, the cost of economic displacements is extreme: $g_{11}^{\text{Sofia}} = 1/2{,}500 = 4 \times 10^{-4}$. A $100 expense has metric cost $100^2 / 2{,}500 = 4.0$ for Sofia, compared to $100^2 / 90{,}000 \approx 0.11$ for Maria and $100^2 / 25{,}000{,}000 = 0.0004$ for Crane. The maximin principle says: design institutions that minimize the metric cost experienced by Sofia, because she experiences the largest costs for any given displacement.
The most pernicious feature of economic inequality is that it distorts the metric on all dimensions, not just $d_1$. Poverty does not merely mean having less money; it means experiencing a different geometry on the full manifold.
$d_2$ (Rights): In principle, property rights and legal protections apply equally to all. In practice, the metric for exercising rights varies enormously with wealth. Filing a lawsuit, defending a property claim, enforcing a contract – these actions require financial resources. The poor person’s metric on $d_2$ has higher cost: exercising a right is a larger displacement because the resources required to exercise it are a larger fraction of the poor person’s endowment. The result is that the poor have rights they cannot afford to exercise – rights that are formally equal but geometrically inaccessible.
$d_3$ (Fairness): The sensitivity to fairness violations increases with economic vulnerability. A worker who can walk away from an unfair offer ($g_{33}$ is low – the fairness displacement is cheap because the worker has alternatives) is less affected by unfairness than a worker who cannot afford to refuse ($g_{33}$ is high – the fairness displacement is costly because the boundary between “employed” and “destitute” is nearby). Inequality distorts the fairness metric by making unfairness more costly for those who can least afford to resist it.
$d_4$ (Autonomy): Autonomy requires options, and options require resources. The poor person’s option set is geometrically constrained – the accessible region of the manifold is smaller because boundaries (bankruptcy, eviction, starvation) are closer. The metric $g_{44}$ is larger for the poor because each autonomy displacement carries the risk of crossing a survival boundary. This is not a metaphorical claim about “feeling trapped”; it is a geometric claim about the structure of the manifold.
$d_5$ (Trust): The financial system is designed for wealthy participants. Banking services, investment products, insurance, legal protections – all of these trust-infrastructure components are calibrated to agents with significant resources. The poor person’s access to trust infrastructure is limited: unbanked or underbanked, unable to afford insurance, unable to retain a lawyer. The trust metric $g_{55}$ is higher for the poor because the cost of establishing and maintaining trusted relationships is proportionally larger.
$d_9$ (Epistemic): Financial literacy, access to professional advice, and the ability to process complex information are all wealth-dependent. The poor person faces a more complex information environment (the welfare system, predatory lending, low-quality health insurance) with fewer resources to navigate it. The epistemic metric $g_{99}$ is higher for the poor – the cost of resolving uncertainty is proportionally larger.
The aggregate effect is that inequality is not a single-dimensional phenomenon but a metric distortion that pervades all nine dimensions. The poor person experiences a manifold that is more costly to traverse in every direction, has boundaries closer to the current position, and has fewer paths with finite cost. The wealthy person experiences a spacious, low-curvature manifold with distant boundaries and abundant paths. The inequality between them is not captured by the ratio of their incomes; it is captured by the ratio of their metric tensors.
The multi-dimensional metric distortion creates a compounding effect that single-dimensional measures cannot capture. Consider the following cascade:
This is a positive feedback loop in the metric structure: metric distortion on one dimension causes metric distortion on other dimensions, which feeds back to the original dimension. The result is that inequality is self-reinforcing through the cross-dimensional structure of the metric. A person who starts with a disadvantage on $d_1$ accumulates disadvantages on $d_3$, $d_4$, $d_5$, and $d_9$ through the off-diagonal couplings of the metric tensor. The disadvantages compound not arithmetically but geometrically, because the metric distortion affects the cost structure of all future decisions.
This compounding explains why inequality is so persistent despite substantial transfer programs. A transfer that equalizes $d_1$ (monetary income) does not equalize the metric: the recipient’s metric on $d_3$, $d_4$, $d_5$, and $d_9$ may remain distorted due to the accumulated effects of prior inequality. The metric structure has a memory – it encodes the history of the agent’s position on the manifold, not just the current position. Equalizing the current position does not erase the metric distortion accumulated from the prior trajectory.
We now construct a detailed comparison of the metric experienced by Maria, her minimum-wage employee Sofia, and her REIT landlord Richard Crane for the same $100 expense.
Sofia is Maria’s barista. She earns $18/hour, works 30 hours/week, and takes home approximately $2,160/month after taxes. She rents a room in a shared apartment for $1,100/month. After rent, utilities, transportation, and food, she has approximately $200/month in discretionary income. Her nearest survival boundary (unable to pay rent) is approximately $200 away from her current state.
Maria is the shop owner. She takes home approximately $4,500/month in owner’s draw after business expenses. She rents an apartment for $2,200/month. After personal expenses, she has approximately $800/month in discretionary income. Her nearest survival boundary (business failure) is approximately $3,000 away from her current state.
Crane is the REIT CEO. He earns approximately $4.5 million annually, or $375,000/month. His monthly discretionary income, after mortgage, taxes, and living expenses, is approximately $150,000. His nearest financial boundary (material lifestyle change) is millions of dollars away.
Each agent must pay an unexpected $100 expense – a medical co-pay, a parking ticket, a broken appliance.
$d_1$ (monetary): $100 is 50% of Sofia's monthly discretionary income. The monetary scale $\sigma_{1,\text{Sofia}} \approx 50$ (her normal daily variation).
$$w_1^{\text{Sofia}} = \frac{100^2}{50^2} = 4.0$$
$d_3$ (fairness): Sofia perceives the expense as unfair – a medical co-pay for a condition caused by standing eight hours a day, a parking ticket in a neighborhood where free parking has been replaced by meters. The fairness metric activates: $\Delta d_3 \approx -0.6$, $\sigma_3 \approx 0.8$.
$$w_3^{\text{Sofia}} = \frac{0.6^2}{0.8^2} = 0.5625$$
$d_4$ (autonomy): The $100 eliminates Sofia's discretionary budget for two weeks. She cannot choose to go out with friends, buy new shoes, or visit her family. The autonomy metric activates: $\Delta d_4 \approx -0.7$, $\sigma_4 \approx 0.6$.
$$w_4^{\text{Sofia}} = \frac{0.7^2}{0.6^2} = 1.361$$
$d_9$ (epistemic): Sofia is anxious about what other unexpected expenses might follow. Her financial situation has no buffer, so uncertainty about future costs is acutely stressful. $\Delta d_9 \approx -0.5$, $\sigma_9 \approx 0.5$.
$$w_9^{\text{Sofia}} = \frac{0.5^2}{0.5^2} = 1.0$$
$d_1$-$d_4$ coupling (money-autonomy): The tight coupling between monetary loss and autonomy reduction produces a significant cross-term. With $g_{14}^{\text{Sofia}} \approx 0.3$:
$$w_{14}^{\text{Sofia}} = 2 \times 0.3 \times \frac{100}{50} \times \frac{0.7}{0.6} = 2 \times 0.3 \times 2.0 \times 1.167 \approx 1.4$$
Total metric cost for Sofia:
$$w_{\text{total}}^{\text{Sofia}} = 4.0 + 0.5625 + 1.361 + 1.0 + 1.4 + \text{other cross terms} \approx 9.1$$
Boundary proximity: The $100 expense moves Sofia from $200 discretionary to $100 discretionary -- cutting her distance from the survival boundary in half. This boundary proximity does not appear in the Mahalanobis distance but enters through increased anxiety ($d_9$) and reduced autonomy ($d_4$) in subsequent decisions. The effective metric cost, accounting for boundary proximity effects, exceeds 9.1.
$d_1$ (monetary): $100 is 12.5% of Maria's monthly discretionary income. The monetary scale $\sigma_{1,\text{Maria}} \approx 300$.
$$w_1^{\text{Maria}} = \frac{100^2}{300^2} = 0.111$$
$d_3$ (fairness): Maria perceives the expense as annoying but not unfair. The fairness metric activates mildly: $\Delta d_3 \approx -0.2$, $\sigma_3 \approx 0.8$.
$$w_3^{\text{Maria}} = \frac{0.2^2}{0.8^2} = 0.0625$$
$d_4$ (autonomy): The $100 reduces Maria's options slightly but does not eliminate any category of choice. $\Delta d_4 \approx -0.15$, $\sigma_4 \approx 0.6$.
$$w_4^{\text{Maria}} = \frac{0.15^2}{0.6^2} = 0.0625$$
$d_9$ (epistemic): Maria is mildly concerned but not anxious. Her buffer is sufficient. $\Delta d_9 \approx -0.1$, $\sigma_9 \approx 0.5$.
$$w_9^{\text{Maria}} = \frac{0.1^2}{0.5^2} = 0.04$$
Cross terms: With weaker coupling ($g_{14}^{\text{Maria}} \approx 0.1$, due to greater distance from boundaries):
$$w_{14}^{\text{Maria}} \approx 2 \times 0.1 \times \frac{100}{300} \times \frac{0.15}{0.6} \approx 0.017$$
Total metric cost for Maria:
$$w_{\text{total}}^{\text{Maria}} = 0.111 + 0.0625 + 0.0625 + 0.04 + 0.017 + \text{other small terms} \approx 0.33$$
Boundary proximity: The $100 moves Maria from $800 discretionary to $700 discretionary – a 12.5% reduction in her distance from the survival boundary. The effect is noticeable but not threatening.
$d_1$ (monetary): $100 is 0.00007% of Crane's monthly discretionary income. The monetary scale $\sigma_{1,\text{Crane}} \approx 5{,}000$.
$$w_1^{\text{Crane}} = \frac{100^2}{5{,}000^2} = 0.0004$$
$d_3$, $d_4$, $d_9$: These dimensions are not activated by a $100 expense. The expense is below Crane’s perceptual threshold – he may not even notice it on his credit card statement. All cross-dimensional activations are approximately zero.
Total metric cost for Crane:
$$w_{\text{total}}^{\text{Crane}} \approx 0.0004$$
Boundary proximity: Zero. No boundary of any kind is within millions of dollars of Crane’s position.
| Agent | $d_1$ cost | Full manifold cost | Ratio to Crane |
|---|---|---|---|
| Sofia | 4.0 | 9.1 | 22,750x |
| Maria | 0.111 | 0.33 | 825x |
| Crane | 0.0004 | 0.0004 | 1x |
The $d_1$-only inequality ratio between Sofia and Crane is $4.0 / 0.0004 = 10{,}000$. The full-manifold inequality ratio is $9.1 / 0.0004 = 22{,}750$. The multi-dimensional metric more than doubles the measured inequality, because the cross-dimensional activations (fairness, autonomy, anxiety) that the $100 expense triggers for Sofia are invisible on the $d_1$ projection.
This is the central empirical prediction of the geometric theory of inequality: scalar measures underestimate inequality by a factor of 2-3x, because they capture only the $d_1$ metric distortion while the full metric distortion includes cross-dimensional effects that amplify the inequality.
The Gini coefficient for this three-person economy, computed on $d_1$ income alone, would be approximately 0.67 (reflecting the extreme income disparity between Crane and the other two). But the Gini cannot distinguish between this economy and an alternative economy where:
The full-manifold inequality ratio in the alternative economy might be 12,000x instead of 22,750x – still enormous, but materially different. The Gini sees no difference. The inequality tensor sees the structural improvement.
The geometric theory of inequality implies that effective inequality reduction requires targeting not just $d_1$ (income redistribution) but the full metric tensor. Specifically:
Boundary removal ($d_4$, $d_2$ effects): Policies that move survival boundaries farther from agents’ current positions – social safety nets, universal basic income, guaranteed housing – reduce the effective curvature of the manifold for low-income agents. They do not increase income, but they reduce the metric cost of being poor by eliminating the cliff-like boundary structure that makes every decision high-stakes.
Trust infrastructure equalization ($d_5$ effects): Policies that extend trust infrastructure to low-income agents – banking access, insurance access, legal aid – reduce $g_{55}$ for those agents. The cost of establishing and maintaining trusted relationships should not depend on wealth.
Epistemic equalization ($d_9$ effects): Policies that reduce the information burden on low-income agents – simplified tax systems, transparent pricing, financial literacy programs, consumer protection – reduce $g_{99}$ for those agents. The cost of understanding the economic system should not scale with its complexity.
Autonomy expansion ($d_4$ effects): Policies that expand the option set for low-income agents – public transportation, affordable childcare, job training, portable benefits – reduce $g_{44}$ by increasing the accessible region of the manifold.
Each of these policies addresses a specific component of the inequality tensor. None of them shows up in the Gini coefficient (they do not change income distribution), but all of them reduce the full-manifold inequality ratio. The geometric framework provides the justification: these policies are metric equalizations that reduce the difference between the manifold experienced by the poor and the manifold experienced by the wealthy.
The geometric framework also explains why income redistribution alone is insufficient for reducing inequality. A cash transfer from Crane to Sofia changes $d_1$ – it moves Sofia’s position along the monetary dimension and reduces $\mathcal{I}_{11}$. But it does not change Sofia’s metric on $d_3$, $d_4$, $d_5$, or $d_9$, because those metric components are determined by the structure of the economic system (availability of healthcare, legal services, banking, information), not by Sofia’s income alone.
The result is that income redistribution compresses the income distribution (reducing the Gini) while leaving the full inequality tensor largely unchanged. The poor person with more money still faces a manifold with closer boundaries, higher cross-dimensional couplings, and larger metric costs on the evaluative dimensions. The transfer has moved their $d_1$ position but not changed the shape of their manifold.
Full inequality reduction requires both income redistribution (to reduce $\mathcal{I}_{11}$) and institutional reform (to reduce $\mathcal{I}_{33}$, $\mathcal{I}_{44}$, $\mathcal{I}_{55}$, and $\mathcal{I}_{99}$). The geometric framework clarifies why both are necessary: they target different components of the inequality tensor, and neither alone is sufficient to equalize the full manifold experience.
[Conditional Theorem.] If an agent’s monetary variance scales with wealth – $\sigma_1^2 \propto W^\alpha$ for some exponent $\alpha > 0$ – then the monetary metric component is:
$$g_{11}(W) = \frac{1}{\sigma_1^2} \propto W^{-\alpha}$$
For $\alpha = 2$ (variance proportional to the square of wealth, consistent with the observation that wealthier agents experience proportionally larger absolute fluctuations), $g_{11} \propto 1/W^2$, and the metric cost of a dollar displacement is:
$$w(D; W) = D^2 \cdot g_{11}(W) \propto \frac{D^2}{W^2} = \left(\frac{D}{W}\right)^2$$
This is the square of the relative displacement – the displacement as a fraction of wealth. The metric naturally converts absolute monetary displacements to relative ones, producing the empirically observed pattern that agents respond to proportional changes, not absolute changes (Weber’s law for economic stimuli).
For $\alpha = 1$ (variance proportional to wealth), $g_{11} \propto 1/W$, and the metric cost is $D^2/W$ – closer to a logarithmic utility function $u(W) \sim \ln(W)$ where $u'' = -1/W^2$ and the local curvature determines risk aversion. The geometric framework does not commit to a specific $\alpha$; it identifies $\alpha$ as an empirical parameter to be estimated from behavioral data.
[Established Mathematics.] The Gini coefficient for a population with incomes $y_1, \ldots, y_N$ is:
$$G = \frac{\sum_{i=1}^N \sum_{j=1}^N |y_i - y_j|}{2N\sum_{i=1}^N y_i}$$
The numerator is proportional to the mean absolute difference, which is a monotone function of the variance of $y_i$ for symmetric distributions (and approximately so for mildly skewed distributions). The denominator normalizes by the mean.
The connection to the inequality tensor: if $\sigma_{1,i}^2 = c \cdot y_i^2$ (monetary variance proportional to income squared), then $g_{11}^{(i)} = 1/(c \cdot y_i^2)$, and the variance of $g_{11}^{(i)}$ across agents is:
$$\mathcal{I}_{11} = \text{Var}\left[\frac{1}{c \cdot y_i^2}\right] = \frac{1}{c^2}\text{Var}\left[\frac{1}{y_i^2}\right]$$
This is related to but not identical to the Gini (which involves $|y_i - y_j|$, not $1/y_i^2$). The Gini is a specific functional of the income distribution; $\mathcal{I}_{11}$ is a different functional of the metric distribution. Both capture income inequality, but they weight different parts of the distribution differently. The inequality tensor is more sensitive to extreme poverty (because $1/y_i^2$ diverges as $y_i \to 0$), while the Gini is more sensitive to the overall shape of the income distribution.
[Conditional Theorem.] The off-diagonal component $\mathcal{I}_{14}$ (cross-inequality between money and autonomy) measures the population variance of the coupling $g_{14}^{(i)}$ between monetary and autonomy metrics. When $g_{14}^{(i)}$ is large for poor agents (money and autonomy are tightly coupled – losing money means losing autonomy) and small for wealthy agents (money and autonomy are decoupled – losing money does not affect autonomy), $\mathcal{I}_{14} > 0$, indicating cross-dimensional inequality.
This cross-dimensional inequality is invisible to any measure that examines $d_1$ and $d_4$ independently. It exists in the coupling between dimensions – the off-diagonal structure of the metric tensor – and can only be captured by the full inequality tensor.
[Conditional Theorem.] The Rawlsian maximin principle, in the geometric framework, is the solution to:
$$\min_{g_{\mu\nu}^{\text{policy}}} \max_i \left[\text{BF}_i(\gamma_i^* ; g_{\mu\nu}^{(i)} + g_{\mu\nu}^{\text{policy}})\right]$$
where $g_{\mu\nu}^{\text{policy}}$ represents the metric modification induced by a policy (e.g., a tax-transfer scheme that modifies $g_{11}$ for different agents, or a social insurance program that modifies $g_{44}$). The Rawlsian planner chooses the policy that minimizes the maximum behavioral friction experienced by any agent.
This formulation makes the Rawlsian program concrete: it specifies what the planner optimizes (minimum of the maximum metric cost) and what instruments are available (policy modifications to the metric tensor). The Rawlsian planner does not need to “imagine” being the worst-off agent; they need to compute the metric experienced by the worst-off agent and choose policies that reduce it.
[Modeling Axiom.] Amartya Sen’s capability approach (1999) argues that inequality should be measured not in terms of income or utility but in terms of capabilities – the substantive freedoms available to each agent. The geometric framework provides the mathematical formalization: an agent’s capabilities are the set of states reachable by geodesics from their current position with finite cost. The capability set is:
$$C_i = \{v \in \mathcal{E} : \exists \gamma \text{ from } v_0^{(i)} \text{ to } v \text{ with } \text{cost}(\gamma; g^{(i)}) < \infty\}$$
Inequality in capabilities is inequality in the size and shape of $C_i$ across agents – which is determined by the agent’s metric $g_{\mu\nu}^{(i)}$ and the boundary structure of the manifold near the agent’s position. The inequality tensor $\mathcal{I}_{\mu\nu}$ is therefore a measure of capability inequality in Sen’s sense, formalized through the metric structure.
The Gini coefficient is from Gini (1912), “Variabilità e Mutabilità,” Studi Economico-Giuridici della Regia Università di Cagliari. The Lorenz curve is from Lorenz (1905), “Methods of Measuring the Concentration of Wealth,” Publications of the American Statistical Association. The Atkinson index is from Atkinson (1970), “On the Measurement of Inequality,” Journal of Economic Theory. The Theil index is from Theil (1967), Economics and Information Theory.
Rawls’s theory of justice is from Rawls (1971), A Theory of Justice. The veil of ignorance, the maximin principle, and the difference principle are developed in Chapters 3 and 4. The metric-averaging interpretation of the veil is original to this framework.
Sen’s capability approach is from Sen (1999), Development as Freedom, and the earlier formulation in Sen (1985), Commodities and Capabilities. The formalization of capabilities as geodesically reachable states on the decision manifold is original to this framework.
The diminishing marginal utility account of “a dollar is worth more to a poor person” traces to Bernoulli (1738), “Specimen Theoriae Novae de Mensura Sortis,” Commentarii Academiae Scientiarum Imperialis Petropolitanae, and the modern formulation in Arrow (1971), Essays in the Theory of Risk-Bearing. The geometric reinterpretation as metric distortion rather than utility curvature is original to this framework.
The claim that scalar inequality measures underestimate full-manifold inequality by a factor of 2-3x is a prediction of the framework that awaits empirical testing. The specific numerical examples in the worked example are illustrative; precise estimates require calibrating the covariance matrix $\Sigma$ and the dimensional activation patterns $\delta_k$ from behavioral data.
The multi-dimensional compounding effect of inequality – where metric distortion on one dimension propagates to other dimensions through the off-diagonal coupling structure – is a structural prediction of the framework that connects to empirical work on the “poverty trap” (Banerjee and Duflo, 2011, Poor Economics) and the “bandwidth tax” of poverty (Mullainathan and Shafir, 2013, Scarcity). The geometric framework provides the mathematical mechanism for these empirically observed phenomena.