Why Engines Do Not Need an Opening Book
Engines once relied on stored opening moves because they could not calculate their way out of move one. That stopped being true, and it changed opening theory.
· 4 min read
An opening book is a stored list of positions and the moves to play in them. For most of computer chess history every serious engine had one, for a simple reason: the opening is the hardest part of the game to calculate, because nothing has happened yet and everything is still possible.
An engine of the 1990s given the starting position and thirty seconds would produce something playable but idiosyncratic. Given a book, it played twelve moves of theory instantly and started thinking from a position it could handle.
What changed
Two things, over about a decade.
Search improved to the point where the opening stopped being uniquely hard. A current engine analysing the starting position settles on the standard first moves within seconds, and its evaluations of the main openings agree with received theory far more than they used to.
Evaluation improved in exactly the area openings live in: long-term compensation, structure, activity without immediate material. The old failure mode, where an engine happily grabbed a pawn and reported a comfortable advantage in a position every strong player knew was miserable, was a scoring problem and it largely went away.
Where books still exist
| Use | Why |
|---|---|
| Engine tournaments | Saves clock time, and forces varied openings so games are not identical |
| Preparation | Steers toward positions the operator has analysed deeply |
| Playing programs | Gives a human opponent recognisable openings rather than novelties |
| Analysis | Rarely, since the engine reaches good moves unaided |
The tournament case is the interesting one. Engine matches usually impose an opening from a book precisely so that both engines are forced out of their comfort and the games differ from each other. Left to themselves, two copies of the same engine would play the same game repeatedly.
The effect on human opening theory
This is where it matters for a player. Engines that evaluate compensation correctly have revised assessments that stood for decades.
Several gambits written off as unsound in the 1980s are respectable now, because the old verdict came from engines that could not see the compensation. Some sharp lines went the other way, refuted by concrete analysis nobody could do by hand. The overall effect has been to push top-level opening theory much deeper, since the engine can verify a line to move thirty and the players are expected to have done so.
What to do with an engine in the opening
Two cautions are worth keeping.
Do not trust a shallow verdict on a mainline. If your engine dislikes a move that has been played for a hundred years, the likely explanations in order are: your analysis is too shallow, the position is one where practical considerations outrank the evaluation, or the move is genuinely a modern refutation. The first is much more common than the third.
Do not memorise a line whose positions you cannot play. An engine will happily give you twenty moves of a sharp variation rated a tenth of a pawn better than the alternative. If move twenty-one leaves you in a position type you have never played, that tenth of a pawn is the most expensive thing you have ever bought.
The useful application is checking your own repertoire for lines where you are worse than you thought, and finding the point where your knowledge runs out. That point is usually earlier than you expect, and it is where preparation actually pays.
A test worth running on your own repertoire
Take the opening you play most and set the board to the last position you know by heart. Analyse it properly, at depth 20 or more.
Three outcomes, each telling you something different:
- The evaluation is fine and the engine's move is one you know. Your knowledge runs at least this far. Push two moves deeper and repeat.
- The evaluation is fine but the engine's move surprises you. This is the productive case. Find out why, because the alternative you were playing is probably worse for a reason you can learn.
- The evaluation is meaningfully worse for you. Your line has a problem, and it is better to find it now than across a board.
Most players discover their repertoire ends three or four moves earlier than they believed, and that the position where it ends is one they have been improvising in for years.
The database is not the book
Worth separating two things that get conflated. An opening book tells an engine what to play. An opening database tells you what has been played, by whom, and with what results.
The second is far more useful to a human. Knowing that a move scores 38% for White across two thousand games is information no evaluation captures, because it includes how hard the resulting positions are for people to play. That is the number that predicts your result, and no engine produces it.
Common questions
- Do chess engines use opening books?
- In competition, usually yes, to save time and to steer toward positions the operator has prepared. For analysis, mostly not. A modern engine analysing from the starting position reaches sensible opening moves on its own within seconds.
- Why does the engine dislike a mainline opening move?
- Usually because the analysis is shallow, or because the move is played for practical reasons the engine cannot weigh. A move that has been played for a century by strong players is rarely refuted by twenty seconds of analysis at depth 18.
Try it on your own position
The analysis board runs the same engine in your browser, free. The app adds what a web page cannot: reading a position straight from a photograph, saved boards, and analysis with no connection.
