{"id":"16f242a8-5936-4e4c-a840-80e5f92d7eef","arxiv_id":"2411.19377","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Feedback Nash equilibria of scalar discrete-time LQ games correspond to intersections of 2^N auxiliary curves with a horizontal line, giving explicit conditions on the number of equilibria.","lead":"This paper develops a graphical way to count feedback Nash equilibria in scalar discrete-time linear quadratic games with any number of players. It shows how many equilibria exist depending on the system parameters, extending results previously known only for the continuous-time case.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2 items v and vi are not checkable parameter conditions: they quantify over the unknown internal variable xi, so the claimed uniqueness statements are not established as stated.","rationale":"I agree with the reader's conditional recommendation, but I weight the two fragilities differently. The Assumption 2 issue is real yet explicitly scoped: Lemma 2 and Theorem 2 are stated under sigma_N > -1, and Remark 2 tells the reader what to do when it fails. That is a limitation, not an unstated gap. The more load-bearing problem is that Theorem 2(v)-(vi) purport to be conditions on system parameters but contain xi, the internal variable of the solution. This undermines the paper's central promise of deriving parameter-based counting conditions. The concrete test above would settle whether the theorem is merely ambiguously phrased or genuinely false under one of the natural readings. The core Lemma 2 construction and items i-iv appear defensible; the garbled equation (11) is a typographical issue that should be fixed but does not change the verdict. Therefore the reader's CONDITIONAL verdict stands unchanged.","tokens_in":17,"tokens_out":36162,"duration_ms":749176,"concrete_test":"Take N=3 with sigma1=0.01, sigma2=sigma3=-0.4 and a=10^-6. Enumerate all solutions of equation (10) by scanning the eight auxiliary functions f_l, recording every intersection with the horizontal line at level a and the corresponding xi. Check whether there is exactly one solution and whether the inequality in Theorem 2(v) holds at that solution and, separately, for all xi in the relevant domain. Repeat with sigma1=0.01, sigma2=sigma3=-0.9, for which the inequality can fail at some xi. If the first case is unique while the inequality fails at the solution, or if the second case has multiple equilibria despite the inequality holding at a solution, the quantifier ambiguity in Theorem 2(v) is substantive rather than cosmetic.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central reduction in Lemma 2 is plausible and, under Assumption 2, Lemma 1 does supply condition (6c), so I do not regard the sigma_N > -1 assumption as a hidden flaw; it is explicit and its scope is acknowledged in Remark 2. The real gap is in the counting theorem. Items v and vi of Theorem 2 require verifying an inequality that contains xi, the internal variable used to parameterize the prospective fixed point (xi = (1/acl - acl)/2), not a system parameter. For instance, item v requires \"(N-1)+|xi|/sqrt(xi^2+1) < sum_i |xi|/sqrt(xi^2-sigma_i)\" with no quantifier. If the inequality is intended at the equilibrium value of xi, it cannot be checked before solving the game, and the theorem reduces to a restatement of the graphical construction rather than a parameter condition. If it is intended to hold for all xi in the relevant domain, the paper does not say so, and the proof does not show that this global condition follows from the parameter-only inequality or that it yields exactly one crossing. Moreover, Lemma 3(vi), which is invoked, states only a non-equality condition for monotonicity, whereas Theorem 2 uses a strict inequality; the transition is not justified. As printed, items v and vi do not deliver the promised a priori conditions in terms of system and cost parameters. The garbled equation (11) in Lemma 2's proof should also be corrected (the intended identity is b_i^2 k_i^2 + 2 xi b_i k_i + sigma_i = 0), but I do not build the verdict on that typo because the intended algebra is recoverable.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies infinite-horizon, discrete-time, scalar-state, N-player linear quadratic dynamic games with quadratic costs. It recalls the standard characterization of linear feedback Nash equilibria through coupled algebraic equations, then introduces the change of variable xi = (1/acl - acl)/2 to eliminate the value functions. This leads to a single scalar equation, equation (10), and a graphical representation via 2^N auxiliary functions f_ell(xi). The paper uses this representation to state conditions, in Theorem 2, on the number of feedback Nash equilibria and on the closed-loop gain acl, and illustrates the results with a three-player numerical example.","tokens_in":11751,"tokens_out":9432,"duration_ms":76227,"significance":"The central reduction in Lemma 2 is appealing and, if correct, provides a clean algebraic picture of feedback Nash equilibria for a class of games where the coupled equations are otherwise difficult to analyse. The paper is self-contained from the stated algebraic equations, and the graphical reformulation is a genuine mathematical reduction rather than an assumption of the desired result. The comparison with the continuous-time scalar case in Remark 5 and the extension of the earlier two-player result to N players are useful contributions. The main value of the paper lies in the counting statements of Theorem 2; those statements are what must be made precise. The manuscript does not provide machine-checked proofs or code, but the algebraic derivations are sufficiently explicit to be verifiable by hand.","major_comments":[{"comment":"Items v and vi are not checkable parameter conditions as stated because they contain the internal variable xi = (1/acl - acl)/2, which is itself defined through the unknown equilibrium closed-loop gain. For example, item v contains the condition (N-1)+|xi|/sqrt(xi^2+1) < sum_i |xi|/sqrt(xi^2-sigma_i) with no quantifier. If this is meant to hold at the equilibrium value of xi, then the condition cannot be verified before solving the game and the theorem reduces to a restatement of the graphical construction. If it is meant to hold for all xi in the relevant domain, the paper neither says so nor proves that this global condition follows from the parameter-only part of the statement. The same issue applies to the inequality condition in item vi.","section":"Theorem 2, items v and vi"},{"comment":"The proof of items v and vi invokes Lemma 3(vi), but Lemma 3(vi) is a non-equality condition: it states that if (N-1)+|xi|/sqrt(xi^2+1) != sum_i |xi|/sqrt(xi^2-sigma_i) for all xi != 0, then f_L is strictly monotone for xi<0 and f_1 is strictly monotone for xi>0. Item v, however, uses the strict inequality '<', and the proof asserts strict decrease of these functions. The transition from 'monotone' to 'strictly decreasing', and the use of the strict inequality form, is not justified in the manuscript. As printed, the uniqueness claims in items v and vi are therefore not established.","section":"Theorem 2 proof, items v and vi; Lemma 3(vi)"}],"minor_comments":[{"comment":"Equation (11) is garbled in the typeset text: as printed, '0 = bi 2 k2 i + xi ki + sigma i 2bi' is dimensionally inconsistent and does not lead to the solutions in equation (9). The intended identity appears to be b_i^2 k_i^2 + 2 xi b_i k_i + sigma_i = 0, which should be corrected.","section":"Lemma 2, equation (11)"},{"comment":"The derivative expression in the proof of Lemma 3(vi) is written as (N-1) +/- xi/sqrt(xi^2+1) + sum_i tau_{ell,i} xi/sqrt(xi^2-sigma_i). The sign of the xi/sqrt(xi^2+1) term depends on the branch of hat f, and the displayed formula should state explicitly which sign applies for f_1 and f_L respectively.","section":"Lemma 3(vi), proof"},{"comment":"The text refers to 'f1(ell), f2(ell), f7(ell) and f8(ell)' in the discussion of the second example; the notation should be f_ell(xi), or the functions should be identified consistently with the labels used in Figure 2.","section":"Section 5, numerical example"},{"comment":"The sentence beginning 'or for any i != l, i != j, if the game is such that any sigma_l = sigma_j' is difficult to parse; the intended statement about repeated sigma values should be reformulated for clarity.","section":"Remark 4"},{"comment":"The phrase 'Assumption 1 can be introduced without loss of generality' is slightly imprecise: reordering the players is indeed without loss of generality, but the assumption itself is a convention, not a restriction.","section":"Assumption 1"}],"recommendation":"major_revision","confidential_remarks":"The central algebraic reduction is sound, and the two-player predecessor is properly cited. The main weakness is that Theorem 2 items v and vi are not a priori parameter conditions as they stand; this is fixable by either quantifying the xi conditions or deriving genuine parameter-only sufficient conditions, but as written the main counting claims in those two items are not established. The paper is appropriate for the journal's scope, though the scalar-state restriction limits its breadth."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core of the paper is the reduction in Lemma 2: in the scalar case, with sigma_N > -1, every nontrivial FNE corresponds to an intersection of one of 2^N auxiliary functions (13) with the horizontal line at level a. That is a clean, honest algebraic elimination, and it converts a coupled cubic system into a one-dimensional graphical problem. The paper does good work here: it extends the two-player construction to N players, treats a wider range of cost parameters, and derives genuinely new counting statements (2^N - 1 equilibria for |a| large; at least one equilibrium when sigma_1 > 0; uniqueness under sigma_N >= 0 with |a| < 1). The debt to Engwerda's continuous-time study and to the authors' own two-player paper is acknowledged openly, and Remark 5 correctly points out where the discrete-time behavior differs from continuous time. The numerical example in Section 5 does what it claims: it shows 7 intersections for |a| large and illustrates the extra equilibria that appear when sigma_N < 0. On the technical side, Lemma 1's use of Assumption 2 to guarantee condition (6c) is explicit and, as far as I can tell, sound; I do not consider that a hidden flaw. The real soft spot is Theorem 2 items v and vi. Both contain inequalities involving the internal variable xi, which is defined as xi = (1/acl - acl)/2 and is not known before solving the game. As printed, those conditions are not checkable a priori, and the theorem's promise of parameter-based conditions is not fulfilled there. The stress-test note is right that the quantifier is missing: if the inequality is meant to hold for all xi in the relevant domain, the paper needs to say so and prove it; if it is meant only at the equilibrium xi, then items v and vi are restatements of the graphical construction, not conditions in terms of system parameters. The jump from Lemma 3(vi)'s non-equality condition to the strict inequality used in Theorem 2(v) also needs justification. Item 11 in Lemma 2's proof is garbled, though the intended identity is recoverable (the quadratic in b_i k_i with linear term 2 xi b_i k_i + sigma_i), so I would count that as a typo, not a substantive error. The asymptotic statements ('|a| >> 1', '|a| << 1') are informal; they should either be made quantitative or explicitly asymptotic. These issues do not sink the core method, and items i-iv and vii look solid to me, but as printed items v and vi do not deliver what they claim. Net: this is worth a serious referee and probably worth citing for the graphical reduction and the sigma_N >= 0 uniqueness result, but the author's revision should fix the quantifier problem in the two fragile items, correct the garbled equation, and sharpen the asymptotics before I would trust the full counting theorem.","headline":"A genuinely useful graphical reduction for scalar discrete-time LQ games, with a couple of theorem-statement gaps that need fixing before the counting results can be taken at face value.","tokens_in":12382,"tokens_out":728,"would_cite":true,"duration_ms":8182,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A25","91A10","93C55","49N10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Feedback Nash equilibria of scalar N-player discrete-time linear quadratic games are exactly the intersections of 2^N auxiliary curves with a horizontal line, and the paper derives parameter conditions for how many equilibria exist.","keywords":["feedback Nash equilibrium","linear quadratic games","discrete-time dynamic games","scalar dynamics","graphical representation","equilibrium multiplicity","coupled algebraic equations","infinite-horizon games"],"falsifier":"For a concrete game with $\\sigma_N<-1$, solve equation (10) graphically, recover the candidate gains from (9), and test each candidate against condition (6c). A candidate that fails (6c) while its intersection root satisfies (10) would demonstrate that the graphical count overcounts equilibria outside Assumption 2, and Theorem 2's counts cannot be applied there without the extra check.","tokens_in":11247,"feed_emoji":"📈","tokens_out":8640,"duration_ms":65576,"temperature":0.7,"pith_summary":"This paper proves a geometric way to count feedback Nash equilibria in infinite-horizon, discrete-time, linear quadratic games where one scalar state is driven by N players with scalar inputs. For such games, the algebraic conditions for a linear feedback Nash equilibrium are cubic in the strategy gains, and multiple equilibria can coexist. The paper shows that under mild parameter ordering every equilibrium with nonzero closed-loop gain corresponds to an intersection of one of $2^N$ auxiliary curves with a horizontal line at height $a$, the open-loop dynamics parameter. From the shape of those curves it derives parameter conditions for the number of equilibria, for instance $2^N-1$ equilibria for large $|a|$, at least one equilibrium when $\\sigma_1>0$, and a unique equilibrium when $\\sigma_N\\ge 0$ and $|a|<1$. A reader cares because these conditions come from system and cost parameters alone, replacing the need to solve coupled cubic equations case by case.","feed_headline":"Graph crossings count Nash equilibria in scalar LQ games","feed_subtitle":"Auxiliary curves intersecting a horizontal line decide how many feedback equilibria a discrete-time game admits.","key_machinery":"The load-bearing object is the family of $2^N$ auxiliary functions $f_\\ell(\\xi)=\\hat f(\\xi)+N\\xi+\\sum_{i=1}^N \\tau_{\\ell,i}\\sqrt{\\xi^2-\\sigma_i}$ for sign tuples $\\tau_\\ell\\in\\{-1,1\\}^N$. The function $\\hat f(\\xi)$ encodes the stabilizing closed-loop gain $a_{cl}=\\hat f(\\xi)$, and the square-root terms encode each player's possible equilibrium gain in (9). Intersections of $f_\\ell$ with the horizontal line at level $a$ are exactly the feedback Nash equilibria of the game, and the asymptotic slopes, monotonicity intervals and definition gaps in Lemma 3 determine how many such intersections can occur for a given $a$.","core_discovery":"The central claim, stated in Lemma 2, is that for the scalar $N$-player game with $\\sigma_i=b_i^2 q_i/r_i$ ordered decreasingly and $\\sigma_N>-1$, the linear feedback strategies $u_i^\\star(k)=k_i x(k)$ with gains $k_i=-\\xi-t_i\\sqrt{\\xi^2-\\sigma_i}/b_i$ form a feedback Nash equilibrium with $a_{cl}\\neq 0$ exactly when there exist signs $t_i\\in\\{-1,1\\}$ and a real $\\xi\\neq 0$ solving $a=\\hat f(\\xi)+N\\xi+\\sum_{i=1}^N t_i\\sqrt{\\xi^2-\\sigma_i}$, where $\\hat f(\\xi)$ is the stabilizing branch of $-\\xi+\\sqrt{\\xi^2+1}$ selected by the sign of $\\xi$. Introducing $\\xi=\\frac12(1/a_{cl}-a_{cl})$ turns the coupled equations (6) into this single scalar equation, so each equilibrium is a graphical intersection. Theorem 2 then reads off the number and properties of equilibria from the auxiliary functions $f_\\ell(\\xi)$, including a closed-loop bound $|a_{cl}^\\star|\\le|\\sqrt{\\sigma_1}-\\sqrt{\\sigma_1+1}|$ when $\\sigma_1>0$ and the positive $\\sigma_i$ are not all equal.","pith_inferences":["The same intersection picture could be turned into a homotopy or continuation method that counts equilibria as $a$ varies, which the paper does not explicitly pursue.","Items v and vi of Theorem 2 involve the internal variable $\\xi$ in their inequalities, so they are not directly checkable from system parameters; deriving parameter-only reformulations would make those conditions fully a priori.","Remark 4 shows that coincident intersections can correspond to a single equilibrium rather than several, so any automatic enumeration must track degeneracies when parameters are tied; counting multiplicities is a separate question the paper leaves implicit.","Although the paper treats scalar state and inputs, the reduction through $\\xi = \\frac12(1/a_{cl}-a_{cl})$ suggests the geometric viewpoint might extend to systems whose closed-loop dynamics are scalarizable, at least as a heuristic."],"forward_implications":["For sufficiently large $|a|$, exactly $2^N-1$ feedback Nash equilibria exist; the missing intersection corresponds to the one sign combination whose asymptotic slope does not meet the level line.","If $\\sigma_1>0$, every open-loop parameter $a$ admits at least one equilibrium; for $a=0$ the trivial zero-gain strategy is counted separately as in Remark 3.","If all $\\sigma_i\\ge 0$ and the open-loop system is stable ($|a|<1$), the equilibrium is unique.","When $\\sigma_1>0$ and the positive $\\sigma_i$ are not all equal, every equilibrium closed loop satisfies $|a_{cl}^\\star|\\le|\\sqrt{\\sigma_1}-\\sqrt{\\sigma_1+1}|$.","In discrete time, a rapidly stabilizing open loop does not by itself guarantee uniqueness when $\\sigma_N<0$, unlike the continuous-time scalar case; uniqueness then requires the additional hypotheses of items v and vi."],"supporting_citations":[{"why":"Supplies the definition of feedback Nash equilibrium and the standard dynamic-game formulation that Theorem 1 specializes to the discrete-time scalar case.","marker":"Ba¸ sar & Olsder (1998)"},{"why":"Its Theorem 3.2 proves the two-player version of the FNE conditions that Theorem 1 adapts to N players.","marker":"Monti et al. (2024)"},{"why":"Establishes the two-player scalar results that Lemma 2 and Theorem 2 generalize.","marker":"Nortmann et al. (2023)"},{"why":"The continuous-time scalar LQ differential-game analysis whose geometric approach and uniqueness results are the discrete-time counterpart and comparison.","marker":"Engwerda (2016)"},{"why":"Dynamic programming principle underpinning the sufficiency part of the FNE characterization in Theorem 1.","marker":"Bellman (1957)"}],"fun_headline_variants":["Curve crossings dictate how many equilibria scalar LQ games have","Counting feedback Nash equilibria via auxiliary curve intersections","Scalar LQ games: equilibrium count from curve crossings","One equation counts all feedback Nash equilibria in LQ games"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The counting results rely on Assumption 2 ($\\sigma_N > -1$), which guarantees that every algebraic solution of the coupled equations automatically satisfies the well-posedness condition (6c); if that fails, intersections of the auxiliary curves may not all be genuine feedback Nash equilibria.","fun_headline_variants_meta":{"raw":{"variants":["Curve crossings dictate how many equilibria scalar LQ games have","Counting feedback Nash equilibria via auxiliary curve intersections","Scalar LQ games: equilibrium count from curve crossings","One equation counts all feedback Nash equilibria in LQ games"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001244,"raw_usage":{"total_tokens":5064,"prompt_tokens":864,"completion_tokens":4200,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":480,"completion_tokens_details":{"reasoning_tokens":4132}},"tokens_in":480,"tokens_out":4200,"duration_ms":24561,"temperature":1.0,"reasoning_tokens":4132,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:14:44.680622+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a concrete game with $\\sigma_N<-1$, solve equation (10) graphically, recover the candidate gains from (9), and test each candidate against condition (6c). A candidate that fails (6c) while its intersection root satisfies (10) would demonstrate that the graphical count overcounts equilibria outside Assumption 2, and Theorem 2's counts cannot be applied there without the extra check.","supporting_citations":[{"cited_title":"& Sas- sano, M","cited_arxiv_id":null,"evidence_quote":"Its Theorem 3.2 proves the two-player version of the FNE conditions that Theorem 1 adapts to N players."},{"cited_title":"& Mylvaganam, T","cited_arxiv_id":null,"evidence_quote":"Establishes the two-player scalar results that Lemma 2 and Theorem 2 generalize."},{"cited_title":"(2016), ‘Properties of feedback Nash equi- libria in scalar LQ differential games’, Automatica 69, 364–374","cited_arxiv_id":null,"evidence_quote":"The continuous-time scalar LQ differential-game analysis whose geometric approach and uniqueness results are the discrete-time counterpart and comparison."},{"cited_title":"(1957), Dynamic Programming, Princeton University Press","cited_arxiv_id":null,"evidence_quote":"Dynamic programming principle underpinning the sufficiency part of the FNE characterization in Theorem 1."}],"review_version":1}