Engine in chess

Engine

Definition

In chess, an engine is a specialized computer program that evaluates positions and calculates moves. It uses search algorithms and evaluation functions to determine the best continuation from any position, outputting recommendations (principal variations), numerical scores (in centipawns), and mate announcements (e.g., M5 for “mate in 5”).

How It’s Used in Chess

Engines are ubiquitous tools for analysis, training, and opening preparation. Players use them to:

Check tactics and blunders after a game.
Explore opening ideas, novelties, and transpositions.
Study endgames—especially with tablebases that provide perfect play for few pieces.
Annotate games with engine lines (principal variations) and evaluations.
Play practice games at adjustable strength levels.

Important ethics note: Using an engine during rated human vs. human play (over-the-board or most online time controls) is cheating and strictly prohibited. Engines are typically allowed for post-game analysis and in certain correspondence formats if explicitly permitted by the event rules.

Outputs and Terminology

Evaluation (Eval): Numeric score in centipawns (cp). +0.80 means White is slightly better; -1.50 means Black is clearly better.
Mate score: Shown as Mx (e.g., M3), indicating a forced checkmate in x moves for the side to move.
Principal Variation (PV): The engine’s best predicted line of play; MultiPV shows multiple top choices.
Depth: How far the engine has searched (in plies); higher depth usually means more reliable results.
Nodes per second (NPS) and Hash: Performance metrics; hash refers to the size of the transposition table in memory.
Tablebase hit: Endgame resolution using perfect knowledge (e.g., Syzygy for 5–7 pieces).

Strategic and Historical Significance

Engines have reshaped opening theory, tactical training, and our understanding of dynamic play. The landmark moment was Kasparov vs. Deep Blue (1997), where a dedicated IBM machine defeated a reigning World Champion in a match—an early signal that engines would soon surpass human strength. Later, open-source engines (notably Stockfish) and neural-network projects (Leela Chess Zero) pushed playing strength beyond 3500 Elo-equivalent, influencing modern preparation at every level.

Opening evolution: Engine-approved gambits and pawn sacrifices have revived many lines. “Engine novelties” frequently appear at elite level.
Endgame clarity: Tablebases solved many theoretical positions, refining classical endgame knowledge (e.g., fortress evaluations, complex rook endings).
Style shifts: Neural networks (AlphaZero, Leela) popularized long-term piece activity, pawn storms, and king safety concepts even in seemingly quiet positions.

How Engines Work (Briefly)

Search: Most engines use alpha–beta pruning with iterative deepening, transposition tables, quiescence search, and heuristics like move ordering, late move reductions, and null-move pruning.
Evaluation: Classic engines score features (material, king safety, pawn structure, mobility). Modern Stockfish uses NNUE—a compact neural net efficiently evaluated on CPUs—to enrich evaluation.
Neural engines: Leela (Lc0) evaluates positions with deep neural networks and searches via Monte Carlo tree search (MCTS), typically accelerated by GPUs.
Protocols: Engines connect to GUIs via UCI or XBoard/CECP protocols, exposing options like Threads, Hash, Skill Level, and MultiPV.

Examples

1) Tactics and mate detection: Even simple tactics are instantly confirmed by an engine. For example, after the dubious 3...Nf6?? in the Scholar’s Mate pattern, the engine shows an immediate mate:

The engine’s evaluation would display something like “M1” for White at the final move.

2) Tablebase certainty: In K+P vs K endgames, engines with tablebases provide perfect answers. The following position is a theoretical draw with best play because the defending king is directly in front of the pawn and it’s White to move:

A tablebase-equipped engine reports 0.00 and shows precise drawing moves for Black.

3) Famous landmark: Kasparov vs. Deep Blue, 1997. Deep Blue’s strategic opening choices and calculation accuracy in critical moments marked a transformative point in human–computer chess. Modern engines far exceed Deep Blue’s strength and run on consumer hardware.

4) Brilliancies and verification: Topalov vs. Shirov, Linares 1998, featured the astonishing ...Bh3!! in a bishop and pawns endgame. Early engines struggled to see it; today’s engines confirm the concept rapidly and provide exact winning sequences.

Practical Tips for Effective Engine Use

Analyze first, engine later: Write down your candidate moves and plans before turning on the engine. Compare and learn from the differences.
Use MultiPV: Seeing 2–4 top lines reveals alternative plans, not just one “best” move.
Let it think: For complex positions, allow deeper search or increase Hash/Threads for more reliable evaluations.
Prefer plans over moves: Ask “why” a move works. Identify themes (weak squares, king safety, pawn breaks) instead of memorizing engine lines.
Check endgames with tablebases: In low-material positions, rely on tablebase results for absolute truth.
Respect fair play: Never consult an engine during games unless the rules explicitly allow it (e.g., specific correspondence formats).

Interesting Facts and Anecdotes

Early pioneers: Alan Turing’s “Turochamp” concept and Claude Shannon’s 1950 paper laid theoretical foundations for computer chess before practical engines existed.
Open-source dominance: Stockfish’s community-driven development and NNUE integration made it a perennial top engine, often used by major chess platforms.
Neural revolution: AlphaZero’s 2017 matches against Stockfish popularized sacrificial, space-gaining play; Leela Chess Zero adopted similar ideas in an open setting.
Championships: Engine competitions like TCEC and CCC track state-of-the-art progress with standardized hardware and conditions.
Everyday impact: The “eval bar” seen in broadcasts is driven by an engine’s score, translating raw centipawns or mate distances into a visual advantage indicator.