Not all Chess960 positions are equally complex

TL;DR

The paper investigates whether all Chess960 starting positions are equally complex by introducing an information-cost measure S(n) that quantifies the cumulative decision difficulty to identify optimal moves. Using Stockfish evaluations across 960 positions, it shows a near-universal first-move advantage for White and substantial heterogeneity in opening complexity, decomposed into White- and Black-side costs S_W and S_B. The study identifies extreme and balanced configurations (e.g., #226, #198) and reveals that standard chess (#518) sits in the middle for overall complexity but carries above-average asymmetry, challenging the notion that the classical start is uniquely privileged. The findings establish an information-theoretic framework for assessing opening depth and fairness, with implications for fair starting-position selection and potential extensions to other strategic games.

Abstract

We analyze strategic complexity across all 960 Chess960 (Fischer Random Chess) starting positions. Stockfish evaluations show a near-universal first-move advantage for White ($\langle E \rangle = +0.30 \pm 0.14$ pawns), indicating that the advantage conferred by moving first is a robust structural feature of the game. To quantify decision difficulty, we introduce an information-based measure $S(n)$ describing the cumulative information required to identify optimal moves over the first $n$ plies. This measure decomposes into contributions from White and Black, $S_W$ and $S_B$, yielding a total opening complexity $S_{\mathrm{tot}} = S_W + S_B$ and a decision asymmetry $A=S_B-S_W$. Across the ensemble, $S_{\mathrm{tot}}$ varies by a factor of three, while $A$ spans from $-2.5$ to $+1.8$ bits, showing that some openings burden White and others Black. The mean $\langle A \rangle = -0.25$ bits indicates a slight tendency for White to face harder opening decisions. Standard chess (position \#518, \texttt{RNBQKBNR}) exhibits above-average asymmetry (91st percentile) but typical overall complexity (47th percentile). The most complex opening is \#226 (\texttt{BNRQKBNR}), whereas \#198 (\texttt{QNBRKBNR})is the most balanced, with both evaluation and asymmetry near zero. These results reveal a highly heterogeneous Chess960 landscape in which small rearrangements of the back-rank pieces can significantly alter strategic depth and competitive fairness. Remarkably, the classical starting position-despite centuries of cultural selection-lies far from the most balanced configuration.

Figures (7)

• Figure 1: Probability distribution of pieces for the 960 initial positions across back-rank squares (from a to h). The bishops, knights, and queen each occupy any of the eight back-rank squares with equal probability $1/8$. By contrast, the king is more likely to appear near the central files, while the rooks are correspondingly more likely to be placed toward the sides.
• Figure 2: Distribution of Stockfish initial position evaluations across all 960 Chess960 starting positions (Stockfish 17.1, depth 30). The distribution is centered at $\langle E \rangle = +0.30 \pm 0.14$ pawns, with 99.6% of positions favoring White. Vertical lines indicate position #518 (standard chess, solid green), position #279 (maximum White advantage, red dashed), position #535 (almost perfect balance, blue dashed), and the ensemble mean (black dashed).
• Figure 3: (a) Average branching factor $\langle b \rangle$ versus ply depth $n$ across all 960 starting configurations (blue), compared with standard chess (position #518, green). Error bars indicate one standard deviation. The dashed line marks the commonly cited middlegame value $b \approx 35$. (b) Number of unique positions $N(n)$ versus ply depth on a semi-logarithmic scale, showing exponential growth $N(n) \sim b_{\mathrm{eff}}^n$ with effective branching factor $b_{\mathrm{eff}} \approx 14.41$, significantly smaller than $\langle b \rangle$ due to transpositions.
• Figure 4: Information cost analysis across the 960 Chess960 starting positions based on $n = 10$ plies per player of optimal play (Stockfish 17.1, depth 14, discrimination threshold $\Delta_0 = 10$ cp). Values represent averages over 5 games per position. (a) Distribution of White's cumulative cost $S_W(10)$ showing the decision complexity faced by White across different starting configurations. The green vertical line marks position #518 (standard chess) at $\langle S_W \rangle = 4.29$ bits, near the ensemble mean. (b) Distribution of Black's cumulative cost $S_B(10)$. Position #518 exhibits $\langle S_B \rangle = 4.97$ bits, slightly above the ensemble mean. (c) Asymmetry $A = S_B - S_W$ versus total information cost $S_{\mathrm{tot}} = S_W + S_B$. Negative asymmetry indicates White faces more complex decisions; positive values indicate Black faces greater complexity. Standard chess (position #518, indicated by a gold star) shows positive asymmetry ($\langle A \rangle = +0.69$ bits) and moderate total complexity (46.5th percentile). (d) Distribution of asymmetry across all positions. The mean asymmetry is $\langle A \rangle = -0.25$ bits (and standard deviation $1.42$ bits), suggesting a weak structural advantage for Black across Chess960 configurations. Standard chess lies at the 91.2th percentile for asymmetry.
• Figure 5: Evolution of the complexity asymmetry $A(n) = S_B(n) - S_W(n)$ as a function of ply $n$ for all 960 Chess960 starting positions. Gray lines show the trajectories of individual positions (each averaged over multiple games). The blue shaded region represents the ensemble mean (blue line) $\pm 1\sigma$. The green line with markers highlights position #518 (standard chess), with its corresponding shaded band indicating $\pm 1\sigma$. Parameters: $\Delta_0 = 10$ cp, depth 20, and $n = 10$ moves per player.