A Note on Learnable Nash Equilibrium

Songzi Du

arxiv: 2606.22701 · v1 · pith:M5EGRQKSnew · submitted 2026-06-21 · 💻 cs.GT · econ.TH

A Note on Learnable Nash Equilibrium

Songzi Du This is my paper

Pith reviewed 2026-06-26 09:16 UTC · model grok-4.3

classification 💻 cs.GT econ.TH

keywords Nash equilibriumlearnabilitymyopic adjustmentindex theorysymmetric gamesasymptotic stabilitygame dynamics

0 comments

The pith

In generic symmetric two-player games a Nash equilibrium is learnable if and only if it has index +1.

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

The paper establishes that learnability of a Nash equilibrium is completely characterized by its index in this class of games. Learnability is defined as the existence of at least one myopic adjustment dynamic under which the equilibrium is asymptotically stable. The result therefore partitions equilibria into those that can arise as stable outcomes of some learning process and those that cannot, regardless of the specific dynamic chosen from the standard families. Because the games are required to be generic and symmetric, the index is well-defined and the best-response structure behaves regularly enough for the equivalence to hold.

Core claim

In generic symmetric two-player games, a Nash equilibrium is learnable if and only if it has index +1. A Nash equilibrium is learnable when there exists a myopic adjustment dynamic (continuous-time or discrete-time) for which the equilibrium is asymptotically stable.

What carries the argument

The index of a Nash equilibrium, a topological invariant that governs whether any myopic adjustment dynamic can render the equilibrium asymptotically stable.

If this is right

Any Nash equilibrium with index -1 is unstable under every myopic adjustment dynamic.
Every Nash equilibrium with index +1 admits at least one myopic adjustment dynamic that makes it asymptotically stable.
The characterization applies only inside the class of generic symmetric two-player games.
Learnability is independent of the particular choice of continuous- or discrete-time myopic process once the index condition is satisfied.

Where Pith is reading between the lines

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

The same index condition may supply a necessary filter for which equilibria can appear in laboratory experiments that implement myopic learning.
The result suggests that topological invariants could classify learnable outcomes in nearby classes such as asymmetric or multi-player games.
If the index can be computed from payoff data alone, experimenters could predict in advance which equilibria are reachable by learning subjects.

Load-bearing premise

The game is generic, so the index of each equilibrium is well-defined and the best-response correspondence satisfies the regularity needed for the equivalence.

What would settle it

Exhibit a generic symmetric two-player game containing either an index-+1 equilibrium that remains unstable under every standard myopic adjustment dynamic or an index--1 equilibrium that is asymptotically stable under at least one such dynamic.

read the original abstract

A Nash equilibrium is learnable if there exists a myopic adjustment dynamic for which it is asymptotically stable. In generic symmetric two-player games, a Nash equilibrium is learnable if and only if it has index +1.

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.

Desk Editor's Note private letter to a colleague

Clean if-and-only-if between index +1 and learnability under some myopic dynamic, but sufficiency looks like the part that needs checking.

read the letter

The central claim is that in generic symmetric two-player games a Nash equilibrium is learnable exactly when it has index +1, where learnable means there exists some myopic adjustment dynamic making it asymptotically stable.

Necessity follows from standard Poincaré index arguments for continuous flows on the simplex, so that part is on solid ground. The paper states the equivalence cleanly and directly ties learnability to a topological invariant that is already computed in many games.

The soft spot is sufficiency. Myopic dynamics restrict the admissible vector fields: the velocity at a point must be zero only on best replies and otherwise point toward payoff improvement. It is not automatic that this restricted class can produce a stable Jacobian at every index +1 equilibrium. The abstract supplies no construction or proof steps, so it is impossible to see whether the authors have an explicit dynamic for each such point or whether the restriction blocks some of them.

This is a short technical note aimed at people who already work on learning dynamics and index theory in games. Specialists who care about stability criteria under adjustment processes could use the result if the sufficiency direction holds, but the paper does not yet give enough detail to rely on it.

It deserves a serious referee to examine the construction for sufficiency and confirm that the myopic constraint does not create counterexamples. I would send it out rather than desk reject.

Referee Report

1 major / 2 minor

Summary. The manuscript claims that in generic symmetric two-player games, a Nash equilibrium is learnable (there exists a myopic adjustment dynamic making it asymptotically stable) if and only if it has index +1.

Significance. If established, the result would give a clean topological criterion for which equilibria in symmetric games can be asymptotically stable under some myopic learning process. Necessity follows from standard index arguments for flows on the simplex; sufficiency would require showing the myopic class is rich enough to stabilize every index-+1 point.

major comments (1)

[Proof of sufficiency] The necessity direction (index +1 required for asymptotic stability) is standard from Poincaré-Hopf theory for continuous flows. The sufficiency direction, however, requires an explicit construction (or existence argument) of a myopic vector field for which every generic index-+1 equilibrium is asymptotically stable. The myopic restriction—that admissible velocities satisfy abla·x = 0 only on best-reply supports—may constrain the possible Jacobians. The manuscript supplies no such construction or verification that the myopic class suffices, which is load-bearing for the claimed equivalence.

minor comments (2)

The abstract states the theorem cleanly but omits any indication of the proof strategy; a single sentence outlining the necessity and sufficiency arguments would improve readability.
Clarify whether 'myopic adjustment dynamic' is restricted to continuous-time processes or also includes discrete-time maps, and state the precise regularity conditions on the vector field.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed report and for highlighting the need for an explicit construction in the sufficiency direction. We agree this is essential to establish the claimed equivalence and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Proof of sufficiency] The necessity direction (index +1 required for asymptotic stability) is standard from Poincaré-Hopf theory for continuous flows. The sufficiency direction, however, requires an explicit construction (or existence argument) of a myopic vector field for which every generic index-+1 equilibrium is asymptotically stable. The myopic restriction—that admissible velocities satisfy abla·x = 0 only on best-reply supports—may constrain the possible Jacobians. The manuscript supplies no such construction or verification that the myopic class suffices, which is load-bearing for the claimed equivalence.

Authors: We acknowledge that the original manuscript provides only a sketch for sufficiency and does not contain an explicit construction. In the revision we will add a dedicated section constructing, for any generic symmetric two-player game, a myopic adjustment dynamic under which every index-+1 equilibrium is asymptotically stable. The construction proceeds by taking a small perturbation of the best-reply vector field that is supported only on best-reply faces (hence myopic) and whose linearization at each index-+1 point has all eigenvalues with negative real part; the perturbation is chosen via a local potential that respects the symmetry and the support condition. This shows the myopic class is rich enough to realize the required stability. We will also verify that the resulting flow remains Lipschitz and satisfies the myopic support restriction globally. revision: yes

Circularity Check

0 steps flagged

No circularity: direct mathematical characterization via standard index theory

full rationale

The paper states a biconditional theorem linking learnability (existence of a myopic adjustment dynamic making the equilibrium asymptotically stable) to Poincaré index +1 in generic symmetric games. Necessity follows from standard degree theory for flows on the simplex (an asymptotically stable rest point must have index +1), which is an external topological fact independent of the paper's definitions. Sufficiency is asserted via existence of suitable dynamics, but the provided abstract and context give no equations or self-citations that reduce the claimed equivalence to a fitted parameter, renamed input, or self-referential definition. The derivation is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The claim rests on the standard definition of Nash equilibrium, the notion of myopic adjustment dynamics, and the topological index; no free parameters or new entities are introduced in the abstract.

axioms (2)

domain assumption A Nash equilibrium is learnable if there exists a myopic adjustment dynamic for which it is asymptotically stable.
This is the explicit definition supplied in the abstract on which the entire claim depends.
domain assumption The game is generic and symmetric two-player so that the index is well-defined.
The theorem is stated only under this restriction.

pith-pipeline@v0.9.1-grok · 5534 in / 1184 out tokens · 22412 ms · 2026-06-26T09:16:07.606201+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

11 extracted references · 4 canonical work pages

[1]

Journal of Economic Theory , author=

From evolutionary to strategic stability , year=. Journal of Economic Theory , author=. doi:None , url=
[2]

Evolutionary Games and Population Dynamics , publisher=

Hofbauer, Josef and Sigmund, Karl , year=. Evolutionary Games and Population Dynamics , publisher=
[3]

Strategic Characterization of the Index of an Equilibrium

von Schemde, Arndt and von Stengel, Bernhard. Strategic Characterization of the Index of an Equilibrium. Algorithmic Game Theory. 2008

2008
[4]

2009 , type =

Anne Balthasar , title =. 2009 , type =

2009
[5]

A note on the Lemke-Howson algorithm

Shapley, Lloyd S. A note on the Lemke-Howson algorithm. Pivoting and Extension: In honor of A.W. Tucker. 1974. doi:10.1007/BFb0121248

work page doi:10.1007/bfb0121248 1974
[6]

Chapter 45 Computing equilibria for two-person games , series =

Bernhard. Chapter 45 Computing equilibria for two-person games , series =. 2002 , issn =. doi:https://doi.org/10.1016/S1574-0005(02)03008-4 , url =

work page doi:10.1016/s1574-0005(02)03008-4 2002
[7]

and Johnson, Charles R

Horn, Roger A. and Johnson, Charles R. , year=. Matrix Analysis , publisher=
[8]

Sustainable Equilibria in Culturally Familiar Games

Myerson, Roger B. Sustainable Equilibria in Culturally Familiar Games. Understanding Strategic Interaction: Essays in Honor of Reinhard Selten. 1997. doi:10.1007/978-3-642-60495-9_10

work page doi:10.1007/978-3-642-60495-9_10 1997
[9]

Evolution and the Theory of Games , publisher=

Maynard Smith, John , year=. Evolution and the Theory of Games , publisher=
[10]

Conference Book of the XV Italian Meeting on Game Theory and Applications , address =

Hofbauer, Josef , title =. Conference Book of the XV Italian Meeting on Game Theory and Applications , address =. 2003 , note =

2003
[11]

Swinkels , abstract =

Jeroen M. Swinkels , abstract =. Adjustment Dynamics and Rational Play in Games , journal =. 1993 , issn =. doi:https://doi.org/10.1006/game.1993.1025 , url =

work page doi:10.1006/game.1993.1025 1993

[1] [1]

Journal of Economic Theory , author=

From evolutionary to strategic stability , year=. Journal of Economic Theory , author=. doi:None , url=

[2] [2]

Evolutionary Games and Population Dynamics , publisher=

Hofbauer, Josef and Sigmund, Karl , year=. Evolutionary Games and Population Dynamics , publisher=

[3] [3]

Strategic Characterization of the Index of an Equilibrium

von Schemde, Arndt and von Stengel, Bernhard. Strategic Characterization of the Index of an Equilibrium. Algorithmic Game Theory. 2008

2008

[4] [4]

2009 , type =

Anne Balthasar , title =. 2009 , type =

2009

[5] [5]

A note on the Lemke-Howson algorithm

Shapley, Lloyd S. A note on the Lemke-Howson algorithm. Pivoting and Extension: In honor of A.W. Tucker. 1974. doi:10.1007/BFb0121248

work page doi:10.1007/bfb0121248 1974

[6] [6]

Chapter 45 Computing equilibria for two-person games , series =

Bernhard. Chapter 45 Computing equilibria for two-person games , series =. 2002 , issn =. doi:https://doi.org/10.1016/S1574-0005(02)03008-4 , url =

work page doi:10.1016/s1574-0005(02)03008-4 2002

[7] [7]

and Johnson, Charles R

Horn, Roger A. and Johnson, Charles R. , year=. Matrix Analysis , publisher=

[8] [8]

Sustainable Equilibria in Culturally Familiar Games

Myerson, Roger B. Sustainable Equilibria in Culturally Familiar Games. Understanding Strategic Interaction: Essays in Honor of Reinhard Selten. 1997. doi:10.1007/978-3-642-60495-9_10

work page doi:10.1007/978-3-642-60495-9_10 1997

[9] [9]

Evolution and the Theory of Games , publisher=

Maynard Smith, John , year=. Evolution and the Theory of Games , publisher=

[10] [10]

Conference Book of the XV Italian Meeting on Game Theory and Applications , address =

Hofbauer, Josef , title =. Conference Book of the XV Italian Meeting on Game Theory and Applications , address =. 2003 , note =

2003

[11] [11]

Swinkels , abstract =

Jeroen M. Swinkels , abstract =. Adjustment Dynamics and Rational Play in Games , journal =. 1993 , issn =. doi:https://doi.org/10.1006/game.1993.1025 , url =

work page doi:10.1006/game.1993.1025 1993