Lattice methods and supermodularity

Overview

Lattice methods and supermodularity are tools for analyzing optimization and equilibrium when choices are complements rather than substitutes. They provide clean, nonparametric rules—often without calculus—for predicting how optimal decisions and strategic equilibria move when parameters change. In plain terms, if “doing more of A makes doing more of B more attractive,” these methods show that optimal A and B rise together, and that in games with such complementarities, higher actions by one player induce higher actions by others. The approach is widely used in microeconomic theory to study firm decisions with complementary inputs, network effects, and industries where competitors’ actions are strategic complements (e.g., price setting with differentiated products).

Origins and Credit

The modern development is due to Donald Topkis (1978; monograph 1998), who formalized optimization on lattices and established core monotone comparative statics results. Paul Milgrom and John Roberts (1990) introduced supermodular games and showed existence and monotone properties of equilibria with strategic complementarities. Milgrom and Chris Shannon (1994) expanded the toolkit with general monotone comparative statics based on “single-crossing” (increasing differences) that require fewer smoothness assumptions. The foundational fixed-point result used in equilibrium analysis is Tarski’s fixed point theorem (1955).

Core Idea and Mechanics

Lattices. A lattice is a set of choices with a partial order where any two elements have a well-defined “minimum” and “maximum” under that order (called meet and join). Common examples are:

• Vectors ordered componentwise (e.g., investment levels across projects); the meet is the componentwise minimum, the join is the componentwise maximum.
• Sets ordered by inclusion (e.g., which markets to enter); the meet is intersection, the join is union.

Supermodularity and increasing differences. A function is supermodular if the value from increasing two components together is at least as large as the sum of increasing each one separately. Intuitively, the variables are complements. For real-valued choices with smooth payoffs, this often corresponds to nonnegative cross-partial derivatives (∂²f/∂x∂y ≥ 0). Increasing differences is a closely related property: the benefit of raising x is larger when y is higher. These properties underpin “single-crossing” comparisons that generate monotone predictions.

Topkis’s theorem (optimization). If an objective has increasing differences in (choice, parameter) and the feasible set is a lattice, then the set of optimal choices is nonempty, forms a lattice, and its extremal optimal choices move monotonically with the parameter. In practice: as market potential, subsidies, or input efficiency rise, optimal decisions that are complements move up together.

Supermodular games (strategic complementarities). In games where each player’s payoff is supermodular in their own action and has increasing differences in their action and others’ actions, best responses are increasing. Tarski’s theorem then guarantees existence of at least one equilibrium and identifies the smallest and largest equilibria. Comparative statics are monotone: pushing a parameter upward (e.g., demand, cost of effort falling) shifts all equilibria upward.

Key Assumptions and Conditions

• Ordered choice sets: Choices live on a lattice (e.g., vectors with lower/upper bounds, sets under inclusion).
• Complementarity: Objective functions exhibit supermodularity or increasing differences in relevant variables and parameters.
• Isotone best responses (games): Each player’s best response is nondecreasing in others’ actions.
• Closedness/compactness: Feasible sets are closed and bounded enough to ensure existence of extremal optimal choices (technical regularity).
• No need for differentiability: The theory works with discrete or kinked choices; smoothness helps but is not required.

Implications

• Monotone comparative statics: When parameters improve (e.g., higher demand, lower input prices), optimal complementary decisions rise weakly; policies generate predictable directional changes.
• Equilibrium existence and order: In strategic complementarities, equilibria exist and can be ordered from lowest to highest. Iterating best responses from the bottom or top converges to an extremal equilibrium.
• Robustness with discrete choices: Predictions hold even with nonconvexities (e.g., lumpy investments, entry/exit), where calculus-based methods struggle.
• Tipping and multiple equilibria: Complementarities can produce multiple stable outcomes (low-activity vs high-activity). Small parameter shifts or coordination can move the system between equilibria.
• Design and policy: Targeted interventions that raise one complementary margin can amplify effects through others, improving cost-effectiveness.

Example in Practice

Go-to-market complements: salesforce size and advertising. A firm chooses monthly ad spend (A) and number of account executives (S). Revenue R(A,S; θ) depends on market potential θ. These are technological and organizational complements: ads generate leads that sales can convert; more reps raise the marginal return to ads, and more ads raise the marginal return to reps. This is increasing differences: the gain from raising A is larger when S is higher, and vice versa.

Suppose the firm faces budget and staffing constraints, so the feasible (A,S) pairs form a lattice (all combinations within upper and lower bounds). By Topkis’s theorem, the set of optimal (A,S) choices is nonempty and nondecreasing in θ and in any subsidy that lowers the effective cost of A or S. Practically:

• When market potential improves, both optimal ad spend and headcount weakly increase.
• If a hiring pipeline opens (lower cost of S), optimal ad spend also rises—an amplification effect from complementarity.
• With discrete headcount and spending tiers, calculus is unnecessary; lattice methods still guarantee the direction of change.

In a competitive variant, if each firm’s best price is increasing in rivals’ prices (strategic complementarity, common with differentiated products), supermodular game theory ensures at least one price equilibrium exists. A rise in industry demand shifts all equilibrium prices up, and simple iterative algorithms that start from low or high price guesses converge to the lowest or highest equilibrium, aiding scenario planning.

Limitations and Common Misunderstandings

• Direction, not magnitude: The methods predict monotone movements, not elasticities or exact changes; they complement, not replace, estimation.
• Complementarities required: If variables are substitutes (benefit of one falls when the other rises), supermodularity fails and monotone results may not hold.
• Multiple equilibria: Existence is reassuring, but selection among equilibria is a separate question; coordination or history can determine outcomes.
• Cross-partials are sufficient, not necessary: Nonnegative cross-partials are a convenient check in smooth problems, but the theory works more generally via order properties.
• Not the same as convexity: Supermodularity is about complements and order; convexity is about averages and curvature. One can have supermodularity without convexity, and vice versa.
• Ordering matters: Results rely on a meaningful partial order. Poorly defined or non-lattice feasible sets can break the conclusions.

Related Concepts (names only)

Topkis’s Theorem; Strategic Complementarities; Increasing Differences; Supermodular Games; Tarski’s Fixed Point Theorem; Monotone Comparative Statics.