Skip to content

Best-Response Correspondence


Definition

The best-response correspondence of player ii assigns to every profile of the other players' strategies s−i∈S−is_{-i}\in S_{-i} the set of player ii's best responses:

BRi(s−i)={si∈Si ⁣:vi(si,s−i)≥vi(si′,s−i)for every si′∈Si}=arg⁡max⁡si∈Sivi(si,s−i).BR_i(s_{-i}) =\{s_i\in S_i\colon v_i(s_i,s_{-i})\ge v_i(s_i',s_{-i})\quad \text{for every } s_i'\in S_i\} =\arg\max_{s_i\in S_i}v_i(s_i,s_{-i}).
BRi(s−i)={si∈Si ⁣:vi(si,s−i)≥vi(si′,s−i)for every si′∈Si}=arg⁡max⁡si∈Sivi(si,s−i).\begin{aligned} &BR_i(s_{-i})\\ &=\{s_i\in S_i\colon v_i(s_i,s_{-i})\ge v_i(s_i',s_{-i})\\ &\qquad\text{for every } s_i'\in S_i\}\\ &=\arg\max_{s_i\in S_i}v_i(s_i,s_{-i}). \end{aligned}

A correspondence is a rule that assigns to each point a set, which may contain several elements or be the empty set. If BRi(s−i)BR_i(s_{-i}) contains exactly one element for every s−is_{-i}, the correspondence is an ordinary function, called a best-response function (or reaction function).


Interpretation

The correspondence summarizes all of player ii's optimal reactions in one object. Several strategies can be optimal against the same s−is_{-i} when payoffs tie, which is why a set is used rather than a single value.

Graph of the Bertrand best-response correspondence of firm 1 in the (p2, p1) plane: a shaded region for p2 below c, a vertical ray at p2 = c, nothing from c to p^m, and a flat line at p^m.

The figure draws Example 3 below: above each rival price p2p_2 it shows the whole set BR1(p2)BR_1(p_2). For p2<cp_2<c the set is the open region above the diagonal, at p2=cp_2=c it is the vertical ray [c,∞)[c,\infty), and for p2>pmp_2>p^m it is the single point pmp^m, where the correspondence behaves like a function. For c<p2≤pmc<p_2\le p^m nothing is drawn above the axis, because the set of best responses is empty.

Move the rival price p2p_2 and read off BR1(p2)BR_1(p_2) above it: a region for p2<cp_2<c, a ray at p2=cp_2=c, nothing between cc and pmp^m, and the single point pmp^m beyond.

A frequent mistake is to report only one maximizer when several tie; a Nash equilibrium can then be missed. A second is to assume BRi(s−i)BR_i(s_{-i}) is always nonempty: in games with infinite strategy sets the supremum of the payoff may not be attained.


Key Results

  • A profile s∗s^* is a Nash equilibrium if and only if si∗∈BRi(s−i∗)s_i^*\in BR_i(s_{-i}^*) for every i∈Ni\in N.
  • For two players with interval strategy sets, the Nash equilibria are the intersection points of the graphs of BR1BR_1 and BR2BR_2 drawn in the same (s1,s2)(s_1,s_2) plane.
  • If SiS_i is finite, BRi(s−i)BR_i(s_{-i}) is nonempty for every s−is_{-i}.

Examples

1. Ties. Suppose player 1 has strategies UU, MM, DD and, against player 2's strategy LL, receives 11, 33, 33. Then BR1(L)={M,D}BR_1(L)=\{M,D\}, a two-element set.

2. Cournot reaction functions. In a Cournot duopoly with price 100−q1−q2100-q_1-q_2 and unit cost 1010,

BR1(q2)=max⁡{90−q22, 0},BR2(q1)=max⁡{90−q12, 0}.BR_1(q_2)=\max\Big\{\frac{90-q_2}{2},\,0\Big\},\qquad BR_2(q_1)=\max\Big\{\frac{90-q_1}{2},\,0\Big\}.

Both are single-valued. Their intersection solves q1=(90−q2)/2q_1=(90-q_2)/2 and q2=(90−q1)/2q_2=(90-q_1)/2, so q1=q2=30q_1=q_2=30.

3. Bertrand: an empty set. In a Bertrand duopoly with identical products, unit cost cc, demand D(p)=max⁡{(a−p)/b,0}D(p)=\max\{(a-p)/b,0\} with a>ca>c and b>0b>0, and monopoly price pm=(a+c)/2p^m=(a+c)/2:

  • BR1(p2)=(p2,∞)BR_1(p_2)=(p_2,\infty) if p2<cp_2<c (any price that sells nothing);
  • BR1(p2)=[c,∞)BR_1(p_2)=[c,\infty) if p2=cp_2=c (profit is 00 at best);
  • BR1(p2)=∅BR_1(p_2)=\varnothing if c<p2≤pmc<p_2\le p^m: firm 1 wants to price just below p2p_2, but no largest price below p2p_2 exists;
  • BR1(p2)={pm}BR_1(p_2)=\{p^m\} if p2>pmp_2>p^m.

Prerequisites



Sources

Aliases: Best Response Correspondence, best-reply correspondence, BR correspondence, reaction curve, best-response curve Keywords: BR_i(s_{-i}), set-valued map, best-response function, reaction function, intersection of best-response curves Symbols: element of, infinity

Last updated: 27 October 2026 11:52