Arasan

Arasan is a chess engine, that is, a console-based program that plays the game of chess.

By itself, it has no graphical interface, but can be used together with interface programs such as Winboard and xboard. There is a separate Windows-only GUI for Arasan distributed with the Windows releases: it is not hosted on Github.

For communicating with a chess interface, Arasan supports either the standard CECP protocol (version 2) or the UCI protocol. CECP is the native protocol used by Winboard and xboard. UCI-compatible chess interfaces include ChessBase and Fritz. Arena, a free interface, supports both protocols.

Arasan is multi-platform (Windows, Linux, Mac OS, Unix) and supports multi-threading for higher performance.

Copyright, license

Copyright 1994-2024 by Jon Dart. All Rights Reserved.

Arasan is licensed under the MIT License. See the LICENSE file.

Program variants

Several different binaries for Arasan can be built. The default executable (arasanx-32 for 32-bit operating systems, or arasanx-64 for 64-bit ones) is designed to be runnable on most systems. Default x86 or x86_64 builds assume SSE2 instructions are available (these were introduced in 2000 with the Pentium IV).

Other program variants include:

1. arasanx-64-old - assumes no SIMD instructions, not even SSE2. Will run on very old x86 hardware.

2. arasanx-64-modern - requires a x86 processor with POPCNT, SSSE3 and SSE4.1 instructions (Intel Nehalem or later, i.e. 2008 era or more recent).

3. arasanx-64-avx2 - requires a more modern x86 processor with AVX2 instructions, as well as those required by the "modern" build. Recommended for AMD Excavator through Zen2 processors.

4. arasanx-64-avx2-bmi2 - requires a more modern x86 processor with AVX2 and BMI2 instructions, as well as those required by the "modern" build. Recommended for Intel Haswell or later processors and AMD processors on the Zen3 architecture (Nov. 2020) or later.

5. arasanx-64-avx512 - requires a modern x86 processor with AVX512 instructions (Intel Skylake X and Cannon Lake, or later), as well as those required by the "avx2" build.

6. arasanx-64-neon - ARM executable with support for NEON instruction set.

In addition it is possible to build a version with support for NUMA (Non- Uniform Memory Access) systems - generally these are large multi-CPU systems. NUMA-enabled versions of Arasan require the HWLOC library version 2.0 or higher: see https://www.open-mpi.org/software/hwloc/v2.0/.

Command-line switches

Arasanx recognizes the following command-line options:

• -H: specifies the hash table size in bytes ('K', 'M', or 'G' can be used to indicate kilobytes, megabyates or gigabytes).

• -c: specifies the number of threads to use (default is 1).

• -f: specifies a position file (FEN) to load on startup

• -t: shows some debugging output in the UI window or log file. To enable debug output, you also need to use the "-debugMode true" switch if using xboard/Winboard, or the "debug=true" option if using Arasan as a UCI engine. You normally shouldn't need to turn this on.

• -ics: outputs additional info when playing on a chess server.

Note: when using Arasan with a GUI or other interface program, usually hash size and thread count are set in the interface. But the -H and -c switches, if specified on the arasanx command line, take precedence over those passed to Arasan via a GUI.

If the word "bench" is specified on the command line, then Arasan will run the bench command (for performance reporting) and then exit.

Option file (arasan.rc)

Many aspects of Arasan's behavior can be modified by changing the arasan.rc file. This file is installed in the program directory. If present, Arasan will also read an arasan.rc file (.arasan.rc on MacOS/Linux) that is in the user's home directory, the contents of which will overwrite the one that is in the program directory.

You don't have to use this file to configure Arasan. Most of the options in it can also be set from a GUI that understands the UCI or CECP protocol, although there are some settings that are only in the arasan.rc file presently.

UCI Options

When used as a UCI engine, Arasan supports the following options:

• Hash - hash size in kilobytes
• Threads - number of threads to use
• Ponder - true to enable pondering (thinking on opponent's time)
• Contempt - followed by a number from -200 to 200. This is the inverse of the value of a draw in centipawns (1/100 pawn value). Negative values mean the opponent is higher rated, so a draw is desirable. Positive values mean the opponent is lower rated, so a draw should be avoided.
• Use tablebases - true to use tablebases, false to disable
• SyzygyUses50MoveRule - true to observe the 50 move draw rule when probing tablebases; false to have tablebase probes ignore the rule. Default is true.
• SyzygyProbeDepth - maximum depth to probe tablebases. Default is 4.
• SyzygyTbPath - path to the Syzygy tablebases. Multiple directories can be specified, separated by ':'.
• MultiPV - followed by a number. Number of distinct best lines to display. Default is 1.
• OwnBook - true or false. True to enable use of the Arasan book (book.bin) if installed. Default is true.
• BookPath - path to the Arasan book. Default is book.bin in the same directory as the Arasan engine executable.
• Favor frequent book moves - value from 0 to 100, default 50. Higher values favor selecting more frequent moves in the Arasan book.
• Favor best book moves - value from 0 to 100, default 50. Higher values favor selecting better scoring moves in the Arasan book.
• Favor high-weighted book moves - value from 0 to 100, default 100. Higher values favor selecting moves with positive manually set weights in the Arasan book.
• Randomize book moves - value from 0 to 100, default 50. Higher values favor selecting moves with a greater degree of randomness.
• Position learning - true to enable position learning (storing key position results in a file and later retrieving them).
• Learn file name - name & location of the file to store position learning data
• UCI_LimitStrength - true or false, default false. True to limit the engine playing strength.
• UCI_Elo - desired playing strength, settable from Elo 1000 to Elo 2600. This is only effective if UCI_LimitStrength is set true.
• Set processor affinity - for NUMA builds only. If set true, binds threads to cores.
• Move overhead - value settable from 0 to 1000. This is a value in milliseconds that will be subtraced from the time available to make a move. It helps Arasan account for network or interface delays in calculating its time usage.
• Use NNUE - use neural network evaluation. True or false.
• NNUE File - the name of the neural network file to load. Currently assumed to be in the same directory as the Arasan engine executable.

The defaults for many of these options are the values set in the arasan.rc file.

CECP options

Besides the standard option-setting commands such as "memory," the following additional option settings are available in CECP mode:

• Can resign - sets whether or not the engine can resign a game. Note: under UCI, the interface program is always in charge of resignation.
• Resign threshold - sets how bad the score must be before resignation. Only effective if "Can resign" is set true. Value is in centipawns (1/100 pawn) and is negative. So for example -500 means resign when down 5 pawns in score value.
• OwnBook - true to enable Arasan's native opening book, false to disable.
• BookPath - path to the Arasan book. Default is book.bin in the same directory as the Arasan engine executable.
• Favor frequent book moves - value from 0 to 100, default 50. Higher values favor selecting more frequent moves in the Arasan book.
• Favor best book moves - value from 0 to 100, default 50. Higher values favor selecting better scoring moves in the Arasan book.
• Favor high-weighted book moves - value from 0 to 100, default 100. Higher values favor selecting moves with positive manually set weights in the Arasan book.
• Randomize book moves - value from 0 to 100, default 50. Higher values favor selecting moves with a greater degree of randomness.
• Position learning - true to enable position learning (storing key position results in a file and later retrieving them).
• Learn file name - name & location of the file to store position learning data
• Strength - search strength, settable on a scale of 0 (weakest) to 100 (strongest)
• Set processor affinity - for NUMA builds only. If set true, binds threads to cores.
• Move overhead - value settable from 0 to 1000. This is a value in milliseconds that will be subtraced from the time available to make a move. It helps Arasan account for network or interface delays in calculating its time usage.
• Use NNUE - use neural network evaluation. True or false.
• NNUE File - the name of the neural network file to load. Currently assumed to be in the same directory as the Arasan engine executable.

Note: the defaults for all these options are taken from the values in arasan.rc, if that file is present.

Neural Network

Arasan now supports chess evaluation utilizing a neural network. The neural network file is external to the program and is loaded at runtime. The file name for the network can be specified using UCI or CECP options, or by setting the search.nnueFile option in the arasan.rc file. There is a default network filename set there. In the options, if no directory path is part of the option or a relative path is used, the file is assumed to be in or relative to the same directory as the Arasan executable.

Tablebases

Arasan supports Syzygy format compressed endgame tablebases. You can configure Arasan to use tablebases by editing the arasan.rc file, or by using a GUI that supports UCI or CECP option commands. Note that many chess GUIs will override the arasan.rc settings, and set their own defaults. So if using a GUI you should if possible set the tablebase path and related options in the GUI, not in the file.

The Arasan distribution does not come with any tablebase files. Syzygy tablebases can be downloaded from https://syzygy-tables.info/.

Programming and build information

See BUILD.md for compilation instructions. See the Programmer's Guide for additional technical information.

Related projects

Arasan relies on the following other projects, also authored by or modified by Jon Dart:

• Fathom, Syzygy tablebase probing code (fork of basil00/Fathom)
• nnue, Chess neural network reading code
• stats, Python module for SPRT computation (derived from glinscott/fishtest)

Additional information

Arasan's website hosts binaries and additional information related to Arasan.