--- title: Row Impartial Terminus (RIT) Game Analysis url: https://www.emergentmind.com/topics/row-impartial-terminus-rit type: topic --- # Row Impartial Terminus (RIT) Game Analysis Searching arXiv for the primary RIT paper and closely related partition-game literature. I’m going to look up the cited arXiv records for RIT and nearby partition-game work. Attempting arXiv lookup: 2509.03390, 2207.04990, and related impartial-game papers. Row Impartial Terminus (RIT) is an impartial combinatorial game on integer partitions, represented as Young diagrams, in which legal moves shorten a uniquely determined row while preserving the partition condition. Its defining structural result is that every position admits a unique decomposition into a **core** and a **remnant**, and that the Conway pair of the RIT position coincides with the Conway pair of the Nim position defined by the remnant. Consequently, both normal-play and misère-play analysis reduce completely to Nim, yielding an explicit winning strategy and placing RIT in the Conway–Gurvich–Ho hierarchy as **forced** and **miserable**, but **not pet** [2509.03390]. ## 1. Formal definition and state space RIT is played on partitions \[ \lambda=\{\lambda_1,\lambda_2,\dots,\lambda_r\},\qquad \lambda_1\ge \lambda_2\ge \cdots \ge \lambda_r>0, \] with \(\lambda\vdash n\) meaning \(\lambda_1+\cdots+\lambda_r=n\). The empty partition \(\{\ \}\) is the unique partition of \(0\). As usual, a partition is identified with its Young diagram: a left-justified array of boxes with \(\lambda_j\) boxes in row \(j\), counted from the top. A legal move is defined as follows. For a partition \(\lambda=\{\lambda_1,\dots,\lambda_r\}\) and an integer \(k\in[\lambda_1]=\{1,2,\dots,\lambda_1\}\), let \(i\) be the largest index such that \(\lambda_i\ge k\). The move replaces \(\lambda_i\) by \(k-1\) and leaves all other row lengths unchanged: \[ \bar\lambda_i=k-1,\qquad \bar\lambda_j=\lambda_j\quad (j\neq i). \] This is called a move on the \(i\)-th row. Equivalently, one chooses a column \(k\), finds the lowest row that reaches that column, and shortens exactly that row to length \(k-1\). The only terminal position is the empty partition \(\{\ \}\). RIT is studied under both standard conventions. Under **normal play**, the player making the last move wins, so the terminal position is a \(P\)-position. Under **misère play**, the player making the last move loses, so the terminal position is an \(N\)-position. The paper uses \(G_G(p)\) for the Sprague–Grundy value of a position \(p\) in a game \(G\), \(G^-_G(p)\) for the misère Grundy value, and the **Conway pair** \[ \bigl(G_G(p),\,G^-_G(p)\bigr). \] ## 2. Core–remnant decomposition The central structural device in RIT is the decomposition of a partition into paired rows. Given \[ \lambda=\{\lambda_1,\ldots,\lambda_r\}, \] the rows are grouped as \[ (\lambda_1,\lambda_2),\ (\lambda_3,\lambda_4),\ (\lambda_5,\lambda_6),\ \dots \] with the convention \(\lambda_j=0\) for \(j>r\) when needed. The **core** is defined by duplicating the even-indexed rows: \[ core(\lambda)=\{\lambda_2,\lambda_2,\lambda_4,\lambda_4,\ldots,\lambda_{2\lfloor r/2\rfloor},\lambda_{2\lfloor r/2\rfloor}\}. \] Thus the core retains the lower row in each adjacent pair and repeats it twice. For example, if \[ \lambda=\{8,5,5,3,2\}, \] then \[ core(\lambda)=\{5,5,3,3\}. \] The **remnant** is the tuple of odd-even row differences: \[ rem(\lambda)=\bigl(\lambda_1-\lambda_2,\ \lambda_3-\lambda_4,\ \ldots,\ \lambda_{2\lceil r/2\rceil-1}-\lambda_{2\lceil r/2\rceil}\bigr), \] again with \(\lambda_j=0\) for \(j>r\). In the same example, \[ rem(\lambda)=(8-5,\ 5-3,\ 2-0)=(3,2,2). \] This decomposition is unique. If \[ rem(\lambda)=(d_1,\dots,d_m), \] then the core determines the even rows, and each odd row is recovered by \[ \lambda_{2i-1}=\lambda_{2i}+d_i. \] If \(r\) is odd, the final even row is interpreted as \(0\). The uniqueness of this reconstruction is the structural basis for the reduction of RIT to Nim [2509.03390]. ## 3. Odd and even rows as the mechanism of reduction The decisive technical distinction is between moves on odd-numbered rows and moves on even-numbered rows. If \(\bar\lambda\) is obtained from \(\lambda\) by a move on an **odd** row, then the core is unchanged: \[ core(\bar\lambda)=core(\lambda), \] and the remnant changes by exactly one Nim move: \[ rem(\lambda)\to_{Nim} rem(\bar\lambda). \] Conversely, every Nim move from \(rem(\lambda)\) arises from an odd-row move in RIT that preserves the core. In effect, an odd-row move decreases exactly one coordinate \[ \lambda_{2i-1}-\lambda_{2i}, \] which is formally identical to reducing a single heap in Nim. If \(\bar\lambda\) is obtained from \(\lambda\) by a move on an **even** row, then there exists an odd-row move from \(\bar\lambda\) to some \(\tilde\lambda\) such that \[ rem(\tilde\lambda)=rem(\lambda). \] An even-row move therefore changes the remnant only provisionally; the next player can restore the original remnant by reducing the odd row immediately above by the same amount. This is the copycat mechanism behind the strategic irrelevance of the core. These two lemmas imply that odd-row moves are the genuine Nim moves of the game, while even-row moves do not alter the strategically relevant state in a stable way. A plausible implication is that the game’s entire nontrivial behavior is encoded in the pairwise row differences rather than in absolute row lengths. That implication is made exact by the Grundy theorems below [2509.03390]. ## 4. Equality with Nim under normal and misère play The remnant \[ rem(\lambda)= (\lambda_1-\lambda_2,\ \lambda_3-\lambda_4,\ \ldots) \] is interpreted as a Nim position whose heap sizes are precisely these coordinates. The central normal-play theorem states: \[ G_{RIT}(\lambda)=G_{Nim}(rem(\lambda)). \] Since the Nim Grundy value is the nim-sum of the heap sizes, this yields \[ G_{RIT}(\lambda) = (\lambda_1-\lambda_2)\oplus(\lambda_3-\lambda_4)\oplus\cdots\oplus(\lambda_{2\lceil r/2\rceil-1}-\lambda_{2\lceil r/2\rceil}). \] Hence \(\lambda\) is a normal-play \(P\)-position exactly when this xor is \(0\). The paper proves the parallel misère statement: \[ G^-_{RIT}(\lambda)=G^-_{Nim}(rem(\lambda)). \] Combining the two equalities gives the Conway-pair identity \[ \bigl(G_{RIT}(\lambda),\,G^-_{RIT}(\lambda)\bigr) = \bigl(G_{Nim}(rem(\lambda)),\,G^-_{Nim}(rem(\lambda))\bigr). \] Thus the normal and misère theories of RIT coincide exactly with those of the corresponding Nim remnant [2509.03390]. The proof is by induction on the size of the partition. Odd-row options are exactly the Nim options on the remnant, while even-row options can be answered by an odd-row move restoring the original remnant. Therefore the mex defining the RIT Grundy value matches the mex over Nim followers of the remnant. The misère proof uses the same induction, with terminal value \(1\) instead of \(0\). ## 5. Outcome theory, winning strategy, and representative examples Under normal play, let \[ rem(\lambda)=(h_1,\dots,h_m),\qquad h_i=\lambda_{2i-1}-\lambda_{2i}. \] Then \(\lambda\) is a \(P\)-position if and only if \[ h_1\oplus h_2\oplus\cdots\oplus h_m=0. \] If the xor is nonzero, a winning move is obtained exactly as in Nim: choose a coordinate \(h_i\) and reduce it to \(h_i'