pith. sign in

arxiv: 2604.16944 · v1 · submitted 2026-04-18 · 💻 cs.GT

Selecting Normal-Form Nash Equilibria in Extensive-Form Games via a Sequence-Form Variant of Logit Quantal Response Equilibrium

Yuqing Hou This is my paper

Pith reviewed 2026-05-10 07:05 UTC · model grok-4.3

classification 💻 cs.GT
keywords logit quantal response equilibriumextensive-form gamessequence formNash equilibrium selectionpath-following methodperfect recallequilibrium computation
0
0 comments X

The pith

A sequence-form version of logit QRE lets researchers compute selected normal-form Nash equilibria in extensive-form games by following a differentiable path to the limit.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a sequence-form formulation of logit quantal response equilibrium for extensive-form games with perfect recall. This formulation avoids the exponential blowup of the normal-form strategy space and supports compact computation. Using it, the authors construct a differentiable path-following procedure that starts from any initial point and traces equilibria as the rationality parameter increases, with the endpoint being a normal-form Nash equilibrium. A sympathetic reader would care because many games of interest, such as those in economics and AI, are extensive-form yet their normal-form representations are too large to handle directly.

Core claim

The sequence-form logit QRE is equivalent to the normal-form version, and a path-following method based on it converges to a normal-form Nash equilibrium as the rationality parameter tends to infinity, starting from an arbitrary initial point.

What carries the argument

The sequence-form formulation of logit QRE, which expresses the equilibrium conditions in terms of sequences rather than complete strategies, enabling the differentiable path-following method.

If this is right

  • The method computes equilibria in extensive-form games without explicitly constructing the normal form.
  • Each point on the computed path corresponds to a logit QRE for a specific rationality level.
  • The limit point is a Nash equilibrium selected according to the logit response criterion.
  • It provides an efficient framework for exploiting the equilibrium selection property of logit QRE.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This could allow selection of particular equilibria in very large games like those arising in multi-agent reinforcement learning.
  • The differentiability may enable integration with gradient-based optimization techniques for game solving.
  • Similar reformulations might apply to other quantal response variants or equilibrium concepts.

Load-bearing premise

The sequence-form logit QRE is mathematically equivalent to the normal-form logit QRE, preserving the selection of the same normal-form Nash equilibrium in the limit, and the game has perfect recall.

What would settle it

Demonstrating a game with perfect recall where the limit point from the sequence-form path is not a Nash equilibrium of the normal-form game, or selects a different one than the standard normal-form logit QRE.

read the original abstract

Although logit quantal response equilibrium (logit QRE) offers a natural equilibrium selection mechanism and converges to Nash equilibrium as the rationality parameter tends to infinity, its computation in extensive-form games is generally intractable when based on the normal-form representation, due to the exponential growth of the strategy space. To address this difficulty, this paper develops a sequence-form formulation of logit QRE for finite n-player extensive-form games with perfect recall, which avoids explicit construction of the normal form and enables compact computation. Based on this formulation, we further develop a differentiable path-following method starting from an arbitrary initial point, such that each point on the path corresponds to a logit QRE associated with a particular value of the rationality parameter, and the limiting point yields a Nash equilibrium. In this way, the proposed method provides an efficient computational framework for exploiting the equilibrium selection property of logit QRE in extensive-form games. The effectiveness of the proposed method is validated by theoretical analysis and numerical experiments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript develops a sequence-form formulation of logit quantal response equilibrium (QRE) for finite n-player extensive-form games with perfect recall. This avoids explicit construction of the normal-form strategy space and enables compact computation. The authors further propose a differentiable path-following procedure that traces a continuum of logit QRE points parameterized by the rationality level, converging in the limit to a normal-form Nash equilibrium selected by the logit QRE mechanism. Theoretical analysis and numerical experiments are used to support the approach.

Significance. If the sequence-form variant preserves the equilibrium selection properties of standard normal-form logit QRE, the work would provide a practical computational framework for exploiting QRE-based selection in large extensive-form games without exponential blow-up. The differentiable path-following method and its ability to start from arbitrary initial points are notable technical strengths that could support reproducibility and further algorithmic extensions.

major comments (1)
  1. The central claim that the sequence-form logit QRE converges to the same normal-form Nash equilibrium that would be selected by applying logit QRE directly to the normal-form representation is load-bearing for the title and abstract. Because the logit response is applied locally to sequence probabilities at information sets rather than to full pure strategies (whose payoffs are defined over the exponential normal-form space), the selection dynamics may differ even under perfect recall and Kuhn equivalence. A formal argument or counterexample establishing that the limiting equilibria coincide (or explicitly characterizing when they do not) is required.
minor comments (1)
  1. Clarify in the experimental section how the initial points for the path-following are chosen and whether different starts can yield different limiting equilibria in the tested games.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the insightful comments on the relationship between sequence-form and normal-form logit QRE. We address the major comment below with a clarification of our claims and a commitment to strengthen the presentation.

read point-by-point responses

  1. Referee: The central claim that the sequence-form logit QRE converges to the same normal-form Nash equilibrium that would be selected by applying logit QRE directly to the normal-form representation is load-bearing for the title and abstract. Because the logit response is applied locally to sequence probabilities at information sets rather than to full pure strategies (whose payoffs are defined over the exponential normal-form space), the selection dynamics may differ even under perfect recall and Kuhn equivalence. A formal argument or counterexample establishing that the limiting equilibria coincide (or explicitly characterizing when they do not) is required.

    Authors: We agree that the distinction between local sequence-form responses and global normal-form responses is important and merits explicit treatment. The manuscript positions the contribution as a sequence-form variant of logit QRE whose limiting points are normal-form Nash equilibria; it does not assert that these limits are identical to those obtained by applying standard normal-form logit QRE to the exponentially expanded strategy space. Because the quantal-response function is defined on sequence probabilities at each information set (conditional on reaching that set), the selection dynamics are those of a behavioral/agent-form QRE rather than a normal-form QRE. Under perfect recall the two representations are strategically equivalent by Kuhn's theorem, yet the equilibrium-selection properties generally differ. In the revised manuscript we will (i) add a short clarifying paragraph to the abstract and introduction, (ii) insert a new subsection in the theoretical analysis that formally relates the two notions, and (iii) supply both a characterization of the games in which the selected equilibria coincide and a simple counter-example (a two-player game with a non-trivial information set) in which they diverge. These additions will make the precise scope of the selection claim transparent without altering the algorithmic contribution. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation of sequence-form logit QRE or path-following method

full rationale

The paper introduces a sequence-form formulation of logit QRE for perfect-recall extensive-form games and a differentiable path-following procedure whose limit is a normal-form Nash equilibrium. These rest on the standard property that logit QRE converges to Nash as the rationality parameter tends to infinity, together with the known equivalence of behavioral and mixed strategies under perfect recall. No equation or step is shown to reduce by construction to a fitted parameter, a self-defined quantity, or a load-bearing self-citation whose content is merely renamed. The computational framework and selection claim are therefore independent of the target result and self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption of perfect recall for the equivalence of sequence and normal forms, plus standard mathematical properties of QRE and path-following methods; no free parameters are fitted to data in the core method.

axioms (1)
  • domain assumption Players have perfect recall in the extensive-form game
    Invoked to ensure the sequence-form representation captures all information sets without loss, as stated in the abstract for finite n-player games.

pith-pipeline@v0.9.0 · 5471 in / 1272 out tokens · 84123 ms · 2026-05-10T07:05:56.773747+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

  1. [1]

    Jr.: Equilibrium points of bimatrix games

    Lemke, C.E., Howson, J.T. Jr.: Equilibrium points of bimatrix games. J. Soc. Ind. Appl. Math. 12(2), 413–423 (1964) https://doi.org/10.1137/0112033

    work page doi:10.1137/0112033 1964

  2. [2]

    Rosenm¨ uller, J.: On a generalization of the lemke–howson algorithm to noncooperative n-person games. SIAM J. Appl. Math.21(1), 73–79 (1971) https://doi.org/10.1137/0121010

    work page doi:10.1137/0121010 1971

  3. [3]

    Wilson, R.: Computing equilibria of n-person games. SIAM J. Appl. Math.21(1), 80–87 (1971) https://doi.org/10.1137/0121011

    work page doi:10.1137/0121011 1971

  4. [4]

    In: Hu, T.C., Robinson, S.M

    Garcia, C.B., Lemke, C.E., Luethi, H.: Simplicial approximation of an equilibrium point for non- cooperative n-person games. In: Hu, T.C., Robinson, S.M. (eds.) Mathematical Programming, pp. 227–260. Academic Press, New York (1973). https://doi.org/10.1016/B978-0-12-358350-5. 50011-7

    work page doi:10.1016/b978-0-12-358350-5 1973

  5. [5]

    van der Laan, G., Talman, A.J.J.: On the computation of fixed points in the product space of unit simplices and an application to noncooperative n person games. Math. Oper. Res.7(1), 1–13 (1982) https://doi.org/10.1287/moor.7.1.1

    work page doi:10.1287/moor.7.1.1 1982

  6. [6]

    Doup, T.M., Talman, A.J.J.: A new simplicial variable dimension algorithm to find equilibria on the product space of unit simplices. Math. Program.37(3), 319–355 (1987) https://doi.org/ 10.1007/BF02591741

    work page doi:10.1007/bf02591741 1987

  7. [7]

    Games Econ

    Herings, P.J.-J., van den Elzen, A.: Computation of the nash equilibrium selected by the tracing procedure in n-person games. Games Econ. Behav.38(1), 89–117 (2002) https://doi.org/10. 1006/game.2001.0856

    work page arXiv 2002

  8. [8]

    Herings, P.J.-J., Peeters, R.J.A.P.: A differentiable homotopy to compute nash equilibria of n-person games. Econ. Theory18(1), 159–185 (2001) https://doi.org/10.1007/PL00004129

    work page doi:10.1007/pl00004129 2001

  9. [9]

    Govindan, S., Wilson, R.: A global newton method to compute nash equilibria. J. Econ. Theory 110(1), 65–86 (2003) https://doi.org/10.1016/S0022-0531(03)00005-X

    work page doi:10.1016/s0022-0531(03)00005-x 2003

  10. [10]

    Chen, Y., Dang, C.: A reformulation-based smooth path-following method for computing nash equilibria. Econ. Theory Bull.4(2), 231–246 (2016) https://doi.org/10.1007/s40505-015-0083-7

    work page doi:10.1007/s40505-015-0083-7 2016

  11. [11]

    Cao, Y., Dang, C., Sun, Y.: Complementarity enhanced nash’s mappings and differentiable homotopy methods to select perfect equilibria. J. Optim. Theory Appl.192(2), 533–563 (2022) https://doi.org/10.1007/s10957-021-01977-x

    work page doi:10.1007/s10957-021-01977-x 2022

  12. [12]

    2011.05.014

    Cao, Y., Dang, C.: A variant of harsanyi’s tracing procedures to select a perfect equilibrium in normal form games. Games Econ. Behav.134, 127–150 (2022) https://doi.org/10.1016/j.geb. 2022.04.004

    work page doi:10.1016/j.geb 2022

  13. [13]

    Cao, Y., Chen, Y., Dang, C.: A variant of the logistic quantal response equilibrium to select a perfect equilibrium. J. Optim. Theory Appl.201(3), 1026–1062 (2024) https://doi.org/10.1007/ s10957-024-02433-2 15

    work page 2024

  14. [14]

    Chen, Y., Dang, C.: A differentiable homotopy method to compute perfect equilibria. Math. Program.185(1), 77–109 (2021) https://doi.org/10.1007/s10107-019-01422-y

    work page doi:10.1007/s10107-019-01422-y 2021

  15. [15]

    INFORMS J

    Cao, Y., Chen, Y., Dang, C.: A differentiable path-following method with a compact formulation to compute proper equilibria. INFORMS J. Comput.36(2), 377–396 (2023) https://doi.org/10. 1287/ijoc.2022.0148

    work page arXiv 2023

  16. [16]

    Dokl.akad.nauk Sssr3(3), 62–64 (1962)

    Romanovskii, I.V.: Reduction of a game with complete memory to a matrix game. Dokl.akad.nauk Sssr3(3), 62–64 (1962)

    work page 1962

  17. [17]

    Games Econ

    Koller, D., Megiddo, N.: The complexity of two-person zero-sum games in extensive form. Games Econ. Behav.4(4), 528–552 (1992) https://doi.org/10.1016/0899-8256(92)90035-Q

    work page doi:10.1016/0899-8256(92)90035-q 1992

  18. [18]

    Games Econ

    von Stengel, B.: Efficient computation of behavior strategies. Games Econ. Behav.14(2), 220– 246 (1996) https://doi.org/10.1006/game.1996.0050

    work page doi:10.1006/game.1996.0050 1996

  19. [19]

    Games Econ

    Koller, D., Megiddo, N., von Stengel, B.: Efficient computation of equilibria for extensive two- person games. Games Econ. Behav.14(2), 247–259 (1996) https://doi.org/10.1006/game.1996. 0051

    work page doi:10.1006/game.1996 1996

  20. [20]

    Koller, D., Pfeffer, A.: Representations and solutions for game-theoretic problems. Artif. Intell. 94(1), 167–215 (1997) https://doi.org/10.1016/S0004-3702(97)00023-4

    work page doi:10.1016/s0004-3702(97)00023-4 1997

  21. [21]

    Econometrica70(2), 693–715 (2002) https://doi.org/10.1111/ 1468-0262.00300

    von Stengel, B., van den Elzen, A., Talman, D.: Computing normal form perfect equilibria for extensive two-person games. Econometrica70(2), 693–715 (2002) https://doi.org/10.1111/ 1468-0262.00300

    work page arXiv 2002

  22. [22]

    Miltersen, P.B., Sørensen, T.B.: Computing a quasi-perfect equilibrium of a two-player game. Econ. Theory42(1), 175–192 (2010) https://doi.org/10.1007/s00199-009-0440-6

    work page doi:10.1007/s00199-009-0440-6 2010

  23. [23]

    Govindan, S., Wilson, R.: Structure theorems for game trees. Proc. Natl. Acad. Sci.99(13), 9077–9080 (2002) https://doi.org/10.1073/pnas.082249599

    work page doi:10.1073/pnas.082249599 2002

  24. [24]

    Hou, Y., Cao, Y., Dang, C., Wang, Y.: A sequence-form differentiable path-following method to compute nash equilibria. Comput. Optim. Appl. (2025) https://doi.org/10.1007/ s10589-025-00702-y

    work page 2025

  25. [25]

    arXiv (2025)

    Hou, Y., Cao, Y., Dang, C., Wang, Y.: A sequence-form characterization and differentiable path- following method for computing normal-form perfect equilibria in extensive-form games. arXiv (2025). https://doi.org/10.48550/arXiv.2505.13827

    work page doi:10.48550/arxiv.2505.13827 2025

  26. [26]

    Osborne, M.J., Rubinstein, A.: A Course in Game Theory vol. 1. The MIT Press, Cambridge (1994)

    work page 1994

  27. [27]

    Selten, R.: Reexamination of the perfectness concept for equilibrium points in extensive games. Int. J. Game Theory4(1), 25–55 (1975) https://doi.org/10.1007/BF01766400

    work page doi:10.1007/bf01766400 1975

  28. [28]

    Harsanyi, J.C., Selten, R.: A General Theory of Equilibrium Selection in Games vol. 1. The MIT 16 Press, Cambridge (1988) 17

    work page 1988