Life and death in Go is the study of whether a group of stones can remain on the board against the opponent’s best play. A living group can avoid capture; a dead group cannot ultimately be saved if the opponent carries out the necessary sequence. Many positions are not yet settled: their outcome depends on who plays first, the availability of outside connections, or a fight elsewhere on the board. The Japanese term tsumego usually refers to composed or extracted problems about making groups live or killing them, although problem collections also include related capturing and tactical tasks.
Life-and-death judgment connects the game’s elementary capture rule to its largest strategic decisions. The status of a small corner group can determine territory, provide threats in a ko fight, or change the strength of nearby stones. Accurate analysis therefore requires more than recognizing familiar eye shapes. It combines the mechanics of liberties and capture, the timing of forcing moves, mutual survival, repetition rules, and the relationship between a local position and the rest of the board.
Capture, eyes, and the meaning of life
Stones of the same color joined along horizontal or vertical grid lines form a connected chain. Its liberties are the empty intersections immediately adjacent to any stone in that chain. Diagonal contact alone does not connect stones. When a legal move removes the last liberty of an opposing chain, that entire chain is captured. A chain with one remaining liberty is in atari. These definitions apply to physically connected stones; in ordinary Go language, a “group” may also include several chains that can connect or cooperate defensively.
An eye is an enclosed empty point or region that contributes to a group’s ability to resist capture. The elementary model of secure life is a connected group with two separate, genuine one-point eyes. An opponent cannot ordinarily occupy either eye while the other remains empty, because the surrounding group would still have a liberty and the invading stone would have none. Filling an eye does not become legal merely because it is enclosed: under rules that prohibit suicide, the move must capture something or otherwise leave the newly played chain with a liberty.
The distinction between one eye and two eyes is consequently structural rather than numerical. A single enclosed region containing two empty intersections is not automatically two eyes. An attacker may be able to play inside it, force a capture, and leave only one usable eye space. Conversely, a large connected interior can be secure if its owner can always divide it into two eyes. Life depends on the available moves and connections, not simply on the amount of empty space surrounded.
Two secure eyes are a sufficient condition for ordinary independent survival, but they are not a complete definition of life. Groups can survive through mutual dependence without each possessing two eyes. A group can also live by connecting to another living group rather than making eyes locally. Statements such as “this group needs two eyes” normally assume that it cannot escape, connect, or survive through a mutual-capture arrangement.
The distinction between alive, dead, and unsettled includes the side to move. A position is commonly called unsettled when the defender can live by moving first but the attacker can kill by moving first. A settled dead group need not be captured immediately: its stones can remain on the board until the end of the game. Their physical presence does not establish survival, just as a group lacking two visible eyes is not necessarily dead if it can still construct them.
Eye shape and the geometry of enclosed space
The study of eye shape classifies enclosed spaces by their size, geometry, boundary conditions, and vital points. Small shapes are especially important because their outcomes recur inside much larger fights. Under the usual assumptions of a fully enclosed group, secure surrounding stones, and no relevant outside complications, a straight three-point eye space can be killed by an attacking move at its center. If the defender occupies that center first, the remaining endpoints can function as separate eyes. A bent three-point space has a corresponding vital point at the bend.
Four-point spaces demonstrate why area alone is insufficient. A straight four-point space normally allows the defender to secure two eyes even after an initial invasion, because its two middle points provide alternative means of division. A compact two-by-two square does not provide the same security. Occupying or capturing stones within it can leave a three-point space whose vital point remains accessible to the attacker. The exact continuation still depends on the surrounding position, but the contrast between straight four and square four is a standard introduction to shape analysis.
A false eye resembles an eye but lacks the supporting connections needed for independent survival. At an interior point, the condition of the surrounding diagonal intersections is often decisive. Two diagonally adjacent enemy stones can undermine an apparent one-point eye by separating its boundary into chains that may be captured individually. At the edge, one hostile diagonal stone can be enough to create the corresponding weakness. These are recognition rules for common local arrangements, not substitutes for checking whether the boundary chains are already connected elsewhere.
The corner and edge alter eye-making because the board boundary completes part of the enclosure. A corner group usually needs fewer stones to surround useful space than a central group. The same boundary also limits escape and can create special capturing sequences. Thus a corner shape cannot always be evaluated by translating it into the center of the board: intersections that would be outside liberties in the center simply do not exist at the edge.
Larger familiar shapes include bulky five and rabbity six, whose names describe characteristic configurations of five and six empty intersections. Their instructional importance lies in concentrated vital points that can prevent division into two eyes. They belong to nakade, the technique of destroying the eye-making capacity of an enclosed space, often by allowing an attacking stone or chain to be captured. A named shape identifies a tactical possibility; it does not prove death without examining move order, external liberties, boundary weaknesses, and possible ko.
Killing techniques and defensive resources
Many life-and-death sequences work by reducing the defender’s choices rather than by filling every empty point. An attacker may first force a reply at the edge, then occupy a vital point that prevents the remaining space from dividing. The defender’s central problem is often to preserve flexibility: a move that seems to enlarge an eye can be inferior to a connection that guarantees two distinct eye regions. Both attack and defense therefore depend on the relationship between eye space and the strength of its boundary.
A throw-in deliberately places a stone where it can be captured. Its purpose is to change the position after capture. The defender may be forced to fill an important liberty, join stones in an unfavorable arrangement, or reduce a spacious interior to a killable shape. The sacrificed stone is not evaluated by whether it survives. Its value lies in the forced exchange and the structure left behind. Some throw-ins must be played before an outside liberty is filled; in other positions that timing would allow the defender to connect or escape.
A snapback is a related capture mechanism. One side allows a stone or chain to be taken, then immediately captures the newly enlarged opposing chain. It is not a ko when the recapture removes a different, larger configuration rather than recreating the immediately preceding position. Snapbacks frequently overturn a superficial eye count: an apparent defending capture may actually remove the last obstacle to the attacker’s decisive recapture.
Other techniques manipulate the boundary. A placement deep inside an eye space can attack several prospective eye divisions at once. A shortage-of-liberties sequence can prevent a defender from connecting two chains even when the connecting point is empty. A tesuji, or locally effective tactical move, may exploit an apparently irrelevant weakness in a neighboring stone to force an exchange that changes the life-and-death result. Such moves explain why secure-looking walls require inspection rather than assumption.
Defensive resources include connecting, enlarging the eye space, making a second eye directly, and capturing invading stones at the correct moment. An important defensive pattern is miai: two alternatives such that, if the opponent takes one, the defender can take the other. A group may be alive because two different eye-making moves remain available, even though neither has yet been played. The opponent cannot deny both in a single turn.
The order of otherwise plausible moves is often decisive. A direct vital-point invasion may fail until the attacker has reduced liberties; filling those liberties first may fail if it gives the defender time to strengthen the interior. Accurate analysis identifies not merely the final shape but the sequence of forcing moves that makes that shape unavoidable.
Capturing races and mutual life
A capturing race, or semeai, occurs when opposing groups are each trying to capture the other. Their lives may depend less on making eyes than on which side can remove the opponent’s liberties first. The simplest race involves two eyeless groups with separate outside liberties and no tactical complications. Counting the effective moves needed to capture each side can then establish the outcome, with the first move often deciding an otherwise equal race.
Practical races rarely reduce to counting visible empty points. Some liberties cannot be filled immediately because the approach would be self-atari. An attacker may first need to play an approach move, connect an exposed stone, or prevent a snapback. A liberty requiring such preparation effectively costs more than one move to remove. Conversely, a move can sometimes reduce several defensive possibilities at once. The relevant quantity is the length of the best legal capturing sequence, not the raw number of adjacent empty intersections.
Shared liberties require special attention. These are empty points adjacent to both opposing groups. Filling one may reduce the liberty count of the player’s own group as well as that of the opponent. In many standard eyeless races, players first remove outside liberties and only then approach the shared region. Internal eyes change this relationship because an attacker may have to enter an eye only at a stage when doing so captures the surrounding group. Familiar semeai formulas are useful only when their assumptions about eyes, shared liberties, and approach moves are satisfied.
Seki, commonly translated as mutual life, arises when neither side can capture the other without enabling its own capture. In a basic example, two opposing eyeless groups share two liberties and have no outside liberties. If either player fills one shared liberty, that player’s own group is reduced to one liberty; the opponent can then occupy the remaining shared point and capture it. Neither side benefits from initiating the sequence, so both remain.
Seki is broader than this elementary two-liberty pattern. Some examples contain eyes, several interacting chains, or complex internal capturing possibilities. What makes the position seki is its stable mutual dependence under the applicable rules, not a particular diagram. A position that merely appears balanced may cease to be seki if an outside connection is cut or an approach move becomes available.
Seki separates two ideas often conflated in elementary instruction: being uncapturable and possessing independently secured territory. Stones can survive while adjacent empty points are not counted as territory under a particular scoring system. Identifying mutual life is therefore necessary both for tactical play and for determining the final score.
Ko and conditional survival
A ko is a repeating capture situation in which immediate recapture is prohibited. The elementary case consists of a single-stone capture that could otherwise be answered by restoring the preceding board position. The ko rule interrupts this loop by requiring play elsewhere before recapture can become possible. In life-and-death analysis, this means that a group may be neither unconditionally alive nor simply dead: its survival can depend on the outcome of a ko fight.
A ko threat is a move elsewhere that creates a consequence the opponent may prefer to answer rather than finish the ko. The threatening player can then recapture if the resulting position and the repetition rule permit it. A group’s local status consequently depends on the number and value of threats available across the board. A local solution described as “ko for life” does not guarantee survival. It establishes that the defender has obtained a conditional route to life rather than a direct, unconditional one.
Different ko results can have very different practical values. A direct ko may immediately decide the group’s fate when one side wins it. An approach ko requires one side to make an additional preparatory move before obtaining the decisive capture. Other positions generate repeated preliminary fights or multiple ko dependencies. Counting only the final capture overlooks the extra turns, threats, and opportunities required to reach it.
Threat value must be compared with what is at stake. If ignoring a threat loses a small boundary point while winning the ko captures a large group, the opponent may rationally ignore it. A threat can also be ineffective because its threatened continuation does not actually work, because it damages the threatening player’s own position, or because answering it creates a replacement threat for the opponent. Life-and-death problems that involve ko therefore need a convention about whether outside threats are assumed, excluded, or explicitly supplied.
The details of repetition rules matter. Simple ko prohibits the immediate reversal of the preceding capture. Superko rules prohibit broader classes of repeated board positions, sometimes also taking account of the player to move. Cycles involving several separate captures can consequently have different legal continuations under different rules.
For ordinary instruction, a problem may stop once a ko is established, treating it as the required local result. Full-game analysis must continue further: it must identify who takes the ko first, what threats are available, whether additional moves are needed to complete the capture, and whether the game’s repetition rules allow the intended continuation.
Reading and the structure of a proof
Reading in Go means mentally examining sequences of legal moves and replies. Life-and-death reading is adversarial: a move succeeds only if it withstands the opponent’s strongest response. A line in which the defender cooperates proves little. The attacker must account for every materially different defensive resource, while the defender must find a continuation that survives each serious attacking attempt.
This can be expressed through a game tree. Each position is a node, and each legal move leads to a successor. At an attacking choice, one successful continuation may be enough to establish that the attack works; at a defensive reply, every relevant alternative must be shown to fail before death is proved. The roles reverse when proving life. This alternating structure explains why a sequence that looks forcing can be difficult to verify: the apparent main line may conceal a decisive branch several moves earlier.
Human reading narrows the search through candidate moves. Vital points, ataris, cuts, connections, throws-in, and moves that create two simultaneous threats are considered before arbitrary placements. Recognition of standard shapes reduces the number of continuations requiring detailed examination. Nevertheless, a remembered pattern is a hypothesis rather than a proof. One extra outside liberty or one missing boundary stone can reverse a familiar result.
A complete local analysis needs explicit boundary conditions. It must establish whether the surrounding stones are secure, whether either group can connect outside, whose turn it is, and which repetition rules apply. In published problems, surrounding stones are often conventionally treated as strong, and play outside the diagram is excluded unless indicated. Without those conventions, a supposedly local position may be underdetermined because an omitted escape route or ko threat changes the answer.
Different tasks also require different success criteria. “Black to live” may mean unconditional life, life in seki, or the best available result, depending on the stated problem. “Black to kill” can mean preventing independent life rather than capturing every stone immediately. Problems that require ko should distinguish a ko result from an outright win. A technically correct continuation can fail to solve the intended task if it reaches a weaker result than another available line.
The educational value of reading lies partly in error diagnosis. A failed solution may result from overlooking a legal reply, miscounting liberties, confusing a false eye with a true eye, or assuming that a forcing move must be answered. These are distinct errors. Strong analysis identifies the exact move at which the claimed result ceases to be forced, rather than merely replacing one memorized variation with another.
Strategic consequences beyond the local position
Life and death determines how freely stones can act. A securely living group generally does not need immediate reinforcement and can support nearby fighting. An unsettled group may force its owner to answer moves that would otherwise be small. A dead group no longer requires rescue, but attempts to capture it can still expose the attacker to weaknesses or supply opportunities for forcing exchanges.
The value of an attack is therefore not measured solely by whether it kills. Pursuing a weak group may build an outside wall, secure territory, split the opponent’s forces, or gain the initiative. A defender may survive locally but concede enough elsewhere that the attack remains successful. Conversely, an ambitious killing attempt can fail strategically if the attacker spends too many moves, strengthens the opponent, or leaves surrounding stones vulnerable.
The terms sente and gote describe an important aspect of this timing. A move played in sente induces a reply and allows the player to retain the initiative; a sequence ending in gote leaves the opponent free to play elsewhere. A life-and-death threat can be extremely valuable when it genuinely demands an answer. Its value disappears if the threatened group is already alive or if sacrificing it is preferable to responding.
An unresolved weakness can retain latent tactical value, often called aji. Dead stones may obstruct connections, provide a future cutting point, create a shortage of liberties, or serve as ko threats. Their owner may intentionally leave them in place rather than force immediate clarification. Similarly, a group that can live locally may remain strategically vulnerable if securing that life requires abandoning surrounding stones.
Sacrifice is an essential counterpart to saving groups. Stones are means of controlling intersections and shaping future play, not units that must survive at all costs. A player may allow a corner group to die in exchange for outside strength or a larger capture elsewhere. Sound evaluation distinguishes the local result—death—from the overall exchange, which can still be favorable.
Whole-board evaluation also changes the preferred degree of risk. A player with a sufficient lead may choose an uncomplicated living sequence over a larger but ko-dependent result. A player who needs a major gain may preserve complications. These choices do not alter the objective tactical status of a shape, but they affect which available result best serves the position. Life-and-death analysis supplies the possible outcomes; strategy compares their consequences with the rest of the game.
End-of-game status and differences between rules
Life-and-death judgment is necessary because games normally end without every doomed chain being physically captured. After play stops, players identify dead stones and account for them according to the scoring rules. This convention avoids filling territories with moves whose results are already understood. Disagreement about status, however, requires a procedure for establishing whether a claimed capture or defense actually works.
Under territory-scoring rules, the score generally includes surrounded empty territory and captured opposing stones. Dead stones remaining on the board are removed and treated according to the rules governing prisoners and territory. Under area scoring, the score generally counts occupied intersections together with surrounded empty intersections. The systems often produce the same winner, but their treatment of extra moves, pass procedures, and unsettled groups is not identical.
The distinction affects how disputed status is resolved. Under area scoring, continuing play inside one’s own territory usually replaces an empty area point with a stone of the same color rather than directly reducing the occupied-plus-empty area total. Under territory scoring, an unnecessary play inside one’s territory can reduce the score unless a rule or accounting procedure compensates for it. Formal rules therefore specify when play resumes, whose turn it is, and whether moves used solely to confirm status affect scoring.
Japanese rules include a specialized post-game framework for determining life and death. The 1989 Japanese rules do not simply equate a stone’s status with whether an ordinary continuation from the stopped position would capture it. Their provisions for hypothetical play and ko are part of the status determination. This matters in exceptional shapes, where practical play before the end of the game and confirmation after the game can have different procedural assumptions.
The standard corner configuration known as bent four in the corner is the best-known example. It is treated as dead under Japanese rules, while under other rules its resolution may require actual play and consideration of the surrounding position, including available ko threats. The expression refers to a particular corner configuration and its resulting sequence, not to every four-point eye space that happens to bend.
Seki also exposes scoring differences. Under Japanese territory rules, empty points belonging to seki are not counted as territory. Under area-scoring systems, occupied stones still contribute to the score, and ownership of empty intersections follows the relevant area rules. Consequently, a correct statement that two groups survive does not by itself specify the numerical score. Status, legal continuation, and counting must all be interpreted within a named ruleset.
Problem literature and historical transmission
Life-and-death study has a long history within the written and practical traditions of Go in China, Japan, and Korea. Problems preserve tactical knowledge in a compact form: a diagram, a side to move, and an objective can encode a sequence far more elaborate than the initial position suggests. Collections range from examples drawn from ordinary games to highly artificial compositions constructed around a surprising move or an exceptional rules interaction.
A major Chinese source is the Xuanxuan Qijing, compiled in 1349 by Yan Defu and Yan Tianzhang. It includes tactical problems and became influential in later East Asian Go literature. Its continued importance illustrates the durability of local patterns: changes in opening fashion and professional competition do not alter the elementary geometry of liberties, eyes, and capture.
In Japan, problem composition developed within the institutional world of the hereditary Go houses during the Edo period. The Igo Hatsuyōron, associated with Inoue Dōsetsu Inseki and completed in 1713, is renowned for exceptionally demanding compositions. Its problems often extend well beyond simple eye-making into long tactical sequences involving sacrifices, capturing races, and interactions across large parts of the board. Their difficulty arises not merely from length but from the precise coordination of apparently unrelated resources.
Hayashi Genbi’s Gokyō Shumyō, published in 1812, became another important problem collection. Compared with spectacular large-scale compositions, many of its positions are more directly connected to practical shape knowledge. The coexistence of such works shows two enduring purposes of tsumego: systematic training in recurring local structures and the creation of intricate tactical compositions valued for their construction.
Modern printed books, magazines, teaching materials, and digital collections organize problems by strength, theme, or intended result. Elementary sets emphasize true and false eyes, basic vital points, and short forced sequences. Advanced collections introduce deceptive first moves, quiet defensive resources, ko, and unusual capturing races. The same diagram can be presented with different objectives, so editorial accuracy requires specifying the side to move and the desired result.
Problem solving also differs from tournament play. A composed problem signals that something tactically significant is present, whereas a player in a real game must first notice the possibility and decide whether it matters. Recognition, calculation, and whole-board judgment are therefore separable skills. Historically, problem literature has concentrated tactical material so that complex mechanisms can be studied without reproducing the full game from which a position might have arisen.
Computer analysis and formal life detection
Life and death has been a major problem in computer Go because it combines precise local tactics with potentially remote dependencies. An apparently isolated group may depend on a ladder reaching across the board, an outside connection, or a ko threat in another corner. A program that incorrectly declares such a group dead can misvalue a large region; one that treats every uncertain group as alive can overlook decisive captures.
One approach is to identify a conservative subset of groups whose survival can be proved without tactical speculation. David B. Benson’s 1976 work on unconditional life provided an influential formal method for recognizing pass-alive structures. In this setting, a group is secure even if its owner passes while the opponent plays. The method reasons about chains and enclosed regions, retaining mutually supporting structures and eliminating those that fail its conditions.
Such an algorithm supplies a strong guarantee within its model, but it is not a complete solver for every living group. Tactical life, seki, and ko-dependent survival can fall outside the recognized set. This illustrates an important distinction in automated analysis: a sufficient test can certify some positive cases without deciding all others. Failure to certify life does not establish death.
Dedicated tactical solvers instead explore move trees, using search techniques to prove or disprove a specified result. Methods can prioritize forcing moves, merge repeated positions, and store previously analyzed states. Proof-oriented search distinguishes a demonstrated result from a promising evaluation. Its reliability still depends on the correctness of move generation, repetition handling, boundary assumptions, and the definition of success.
The development of artificial intelligence transformed practical Go analysis through systems combining search with learned evaluation. AlphaGo’s 2016 match victory over Lee Sedol demonstrated the strength of this approach at the highest professional level. Later systems further improved play and made powerful analysis widely available. Their assessments often illuminate life-and-death positions, but a high estimated win rate or a confident suggested move is not itself a mathematical proof of a local result.
Modern machine-learning systems can be especially effective at recognizing plausible shapes and directing search toward strong moves. Difficulties remain when a correct answer depends on a rare configuration, a very long forcing sequence, or a rules interpretation different from that used during training and analysis. A program’s stated score may also assume that both sides will eventually settle an unresolved group correctly.
For formal verification, the task must therefore specify the exact board, player to move, legal-move rules, repetition history where relevant, and terminal objective. “Find a strong move” and “prove this group unconditionally alive” are different computational problems. The former can be answered by a powerful playing system; the latter requires a result whose logical scope and assumptions are explicit.