WorksComputer Science
A faster algorithm for commutator length in free groups
Abstract
We compute the commutator length \cl(w) of a reduced word of length in a free group. Given a target , a deterministic algorithm decides \cl(w)\le g and, when it holds, returns a decomposition into at most commutators, in steps; trying gives \cl(w) in O(M^{3\cl(w)}) steps. For the same test takes expected steps on a uniformly random reduced input in . All bounds are in the unit-cost model.
Introduction
Let be the free group on a finite set with , the case of one generator having ; throughout, is a reduced word over of length . For put
and when , the set above being empty. Throughout, is the target, the bound on the number of commutators given as part of the input: the task is to decide and, when that holds, to exhibit a decomposition. The exponent below is linear in , so for each fixed target the bounds are polynomial in , as in [2]. At a node of the recursion rem denotes the target still available there.
Culler [3] identifies with the least genus of a one-face map with boundary . The best published bound we know is that of Bartholdi, Fialkovski and Ivanov [2] (henceforth BFI): their Theorem A gives, for non-trivial , a cancellation-free factorisation
with . Repeating those four cut points at each of the levels costs on an -RAM (their Corollary C), and their §2.4 improves the case to an average by a matrix trace test.
2020 Mathematics Subject Classification. Primary 68Q25, 68W40; Secondary 20E05, 20F10.
We keep this recursion and replace two of its ingredients. Base case: one Manacher sweep builds an arc graph in which a word of class one shows up as a walk of at most three arcs from a vertex to its antipode ( worst case, in expectation, Theorem 2.8). Recursion: a peel is an exchange of two blocks, so three cut points replace four and the worst case is . That bound alone does not make the search complete: Theorem 3.6 supplies, at every node, a peel lowering the class by exactly one.
All bounds are in the unit-cost model, one step per letter read, lookup or comparison. BFI count group operations, so a residual of length is steps here and one there, and their is steps: comparing exponents alone is the conservative choice.
Theorem 1.1 (main). For a reduced of length and a target , a deterministic algorithm returns a decomposition of into at most commutators, or reports , in steps and space. The algorithm is deterministic; for its expected running time is when the input is drawn uniformly from reduced words of length conditioned on lying in — the expectation is over inputs only.
For the answer is the empty decomposition; assume below, so . Only targets need be searched. By Culler [3] the class of is the genus of a one-face map whose edges are the inverse pairs of letters of , so that map has edges and Euler’s formula gives $2\kappa = E - V + 1 \le E = M/2$; hence , and being an integer, . A target is replaced by and searched at that value: the search still returns a decomposition into at most commutators. Throughout the search .
The class-one case: an expected-linear base test
Notation and the cost model
Write and read the indices mod . A -word is a word of class one. By Wicks [5] a cyclically reduced word is a -word iff some cyclic shift of it equals without cancellation, , , being contiguous blocks, possibly empty; in general is a -word iff it is conjugate to such a word. The three cut points of that shift are what the base test looks for, on the cyclically reduced residual. A word of has even length and trivial abelianisation. Since is conjugacy-invariant we first strip the longest mirror-inverse prefix , that is the largest with for all . Then with and ; for reduced and non-trivial one has , where , so keeps at least four letters and is cyclically reduced (Lemma 2.6).
Cost model. Letters are integers. An average is taken over uniform among the reduced words of length over , where ; these are the words the algorithm has on hand, and the average of Theorem 2.8 is over those of them that lie in , that is, conditioned on . Free reduction and the test are stack passes, dominated by every bound below.
Pruning invariant. Let be a reduced residual and write , where is its cyclically reduced core, unique up to cyclic rotation. For an ordered pair of letters, let be the number of cyclic positions of with , indices read mod . No such block is its own inverse, since is cyclically reduced. The defect is
A cyclic rotation of permutes the counted blocks, so does not depend on the chosen rotation of the core or on where was anchored. We use the following self-contained bound.
Lemma 2.1 (defect bound). For every reduced non-trivial ,
Proof. Write as above, , and let . Choose a pairing attaining the maximum in Proposition 3.2 below and set , with and ; then . Cyclically reducedness gives for all , since would mean , that is, adjacent inverse letters across the cyclic seam of . A block has an inverse block at precisely when , equivalently ; as has no fixed points, this happens exactly when lies in a 2-orbit of . Each 2-orbit yields exactly the two matched block positions , one block and its inverse. Let be the number of 2-orbits, so positions are matched. Fix an inverse-pair class with original counts and matched blocks of each kind. Removing the matched pairs subtracts the same from both counts, so the imbalance stays ; the unmatched positions of the class number . Summing the imbalance over classes gives at most the total number of unmatched positions: . Let the remaining orbits have lengths . Then and . Since for ,
which is (4).
In general, a residual whose own target is is rejected at : then (4) gives , hence above its target. Thus at the root , while a child of a node with target has .
The arc graph and a three-hop walk
Definition 2.2. Let be even, . There is an arc of length from , written (), if
indices mod . The arc graph has vertex set , with an arc for every pair satisfying the condition.
On three unit arcs give , while needs the length-2 arc . For fixed the conditions are not nested in , but for a fixed centre they are: the arc has centre
and its conditions are one palindromic sweep of an interleaved word about .
Lemma 2.3 (the arcs of one centre). Read , on , so that has length . The arc of centre exists iff the positions of form a palindrome of even radius about ; hence, being the radius of there, the arcs of centre are the lengths with .
Proof. With the condition is , that is
by the letter sits at , and . The two indices add up to : each condition is one mirror pair about , and the of them walk through the block from its ends inwards. Conversely such a palindrome meets all conditions of the arc at , so is an integer exactly when .
The centre of (5) is a class modulo ; in , whose indices are modulo , it has the two lifts and . That is what identifies the two twin sources: the arc at has its palindrome centre at in , the arc at at the antipodal , and the two radii agree on lengths , because the shift carries arcs to arcs and the mirror pairs about read as well — the same conditions, read at the antipodal positions. The full radii of at the two centres need not agree, but only lengths up to are arcs, so one radius per centre suffices and the sweep of sees every arc. □
So a length determines its two sources, and . One Manacher sweep [4] on three copies of , a word of length , so that no radius is cut off at an end, computes all circular radii in and lists the arcs of a centre in time proportional to their number: the two sources are twins sharing the centre, and an arc’s centre is determined by its source and its length, so nothing is emitted twice. Building costs with ; let be the maximum degree, in or out, and the number of two-arc paths, so .
The search. The shift carries the arc to , so it is an automorphism of and it suffices to test the starts .
Arcs are read modulo , so the intermediate vertices of a longer walk may wrap; in this range the only arc that could be missed is the single arc of length , which cyclic reducedness forbids as in the proof of Proposition 2.4. So only two and three hops need testing. Mark the in-neighbours of in a Boolean array that carries a version stamp, so that a slot counts as marked only during the start that wrote it and no array is cleared between starts; then run through the arcs out of . A first hop is already a hit if is marked; otherwise the arcs out of are looked up in the array one by one, queries of . Every walk of at most three arcs from is inspected this way, and the first hit returns its intermediate vertices as cut points.
Proposition 2.4 (Wicks’ criterion as a walk). Let be a cyclically reduced word of even length. Then iff, for some , has a walk of at most three arcs from to ; the intermediate vertices are the cut points.
Proof. If without cancellation, with , , , the three blocks give arcs , , of lengths , which sum to ; an empty block simply drops an arc.
Conversely, let a walk of at most three arcs lead from to , and pad it by repeating vertices until it reads , with lengths ; a length is the degenerate arc at a repeated vertex and contributes an empty block. Their sum lies in and is , hence equals or . No arc of length can occur: such an arc out of any requires , and those two letters are cyclically adjacent in a cyclically reduced . Hence the sum cannot be , and when it is every : the single-arc case and its permutations are excluded too, and the walk is monotone.
Now cut at and . We get , , , and the arc conditions say that the antipodal half reads . Every junction but the base one joins adjacent letters of the reduced word ; the base one does not cancel, again by cyclic reducedness. That is Wicks’ form, so . The decomposition is explicit: , so the search returns and . These decompose the cyclically reduced residual; a node holding returns . □
Proposition 2.5 (correctness and worst case). Let be as in Proposition 2.4. If is a -word, the build and search above return its cut points. One pass costs , hence in the worst case.
Proof. The build is and correctness is Proposition 2.4. For the search, a start pays for the marking, and for each arc out of it; summing over the starts gives at most per pass, and the build adds . With (one arc of each length per vertex) and , that is .
The bound is attained on , , which is not a -word, so every one of the starts is examined. Each of its three runs faces a run at the antipodal position, and for each length this yields arcs of length , plus arcs of length one; hence . Convolving the in/out degrees of this arc family gives . □
The expected cost
Lemma 2.6 (the longest strip suffices). Let be reduced, non-trivial and of even length. Then , the residual is cyclically reduced, and ; so the base test decides $\operatorname{cl}(w)=1 correctly.
Proof. Put with and , so and with ; conjugation gives . The mirror condition holds at level exactly when . At the next level it reads , and since and , this is precisely : the two ends of the residual are inverse. Thus if were not cyclically reduced the prefix would extend to , contradicting maximality. This covers , where , and , where a reduced two-letter word is automatically cyclically reduced.
If , then and the condition at that level reads , which reducedness forbids; hence . Equality would give and , which is reduced, hence not , and has non-trivial abelianisation, contradicting . Therefore , is cyclically reduced, and Proposition 2.4 decides , that is . □
Lemma 2.7 (the abelianisation of a random reduced word). Let be uniform among the reduced words of length over , and let be its abelianisation, so that reads . Then
(i) the letters form a Markov chain, uniform on and, given , uniform among the letters not inverse to ; (ii) there are , depending only on , with
for every whose parity is compatible with , and the upper bound holds for all ; (iii) if is a union of at most intervals of positions with , and is a pattern on , then
where depends only on and ; moreover as .
Proof. (i) is the measure itself: is uniform on , and each subsequent letter is uniform on the letters other than the inverse of its predecessor. Proof of (ii). Observe first that is an additive functional of the aperiodic irreducible chain of (i). Its increment has stationary mean zero and stationary increment covariance ; the transition operator has eigenvalue on the mean-zero symmetric subspace and on the antisymmetric one, so the chain is uniformly ergodic. The observable is antisymmetric, hence lies in the latter eigenspace, with , and .
Fourier inversion now gives uniformly for . Write . Since every letter contributes , lies in the index-two coset . Let denote the vector of a letter , so that and . Since is irreducible and aperiodic, Wielandt’s equality case forces , with and diagonal unitary, whenever . Entrywise this reads for every ; the left side is independent of and each has predecessors, so all coincide and for every . Taking and gives : all share one common value , so the only characters of modulus one are and . Near , , and carries the same peak since ; the two add on and cancel off it. Away from both, (). The main term is the same Gaussian integral in the upper and lower bounds, whence ; the same bounds give the upper estimate for every , with no restriction on its size.
Proof of (iii). Write the complement of as gaps, some empty, with free letters and . An interior gap has both flanking letters fixed by ; only the two end gaps have an endpoint outside . Condition further on a complete assignment of those at most two outer endpoint letters. Given and , the letters in distinct gaps are independent: a gap with free letters is a Markov bridge of transitions between its fixed endpoints, its increment weight being the corresponding product of the refined transition matrix , (the fixed right endpoint shifts by a constant); an empty gap contributes the point mass at .
The spectral gap gives uniformly in , so fixed endpoints cost a factor bounded independently of them; the Fourier estimate of (ii) applies to the refined product and gives uniformly in and in the endpoints when . Since and the gap increments are independent given , the conditional law of is the convolution of the gap laws, shifted by . A convolution’s point mass is at most the largest point mass of any factor, and some gap has , giving uniformly in .
Averaging over the at most values of preserves this bound, so . Dividing by the lower bound of (ii) at (here is even) proves (iii).
Theorem 2.8 (expected linear time). The cl = 1 test runs in expected steps on a uniformly random reduced word of length , conditioned on , in the unit-cost model.
Proof. There is a constant with
for every : the sweep is , emitting the arcs is , and the walk costs at most by Proposition 2.5.
The unconditional measure. We first estimate in the uniform measure on reduced residuals of length , so here; at the end . Write . For a fixed arc candidate the conditions pair the positions and for . These pairs are disjoint, and a pair whose positions are adjacent has probability by reducedness. Otherwise reveal the word from left to right: when the later endpoint of a pair is reached, the earlier letter is already known, and the Markov rule prescribes its inverse with probability either (if forbidden by the preceding letter) or exactly . Multiplying over the pairs gives .
A vertex carries at most one arc of each length, so a union bound over the candidate arcs gives . A two-arc path with lengths presupposes the longer of its two arcs, hence has probability at most , so
Adding the three terms of (6) applied to gives for : a bound linear in .
Two conditionings. The measure of the theorem differs from the one the test actually runs under in two ways; we handle them in turn. First fix . Writing with and , we have exactly when and is reduced with , the two junctions. Indeed reads , so for every whatever is; the prefix extends to only when . Under the two forbidden last letters of are distinct.
For a reduced word of length has continuations for each prescribed last letter, so exactly words are admissible; for the only admissible is the empty word, a count of one. Distinct give distinct . For each fixed the tilt is therefore a constant, independent of : conditional on , is uniform among the reduced words of length with . The event has probability bounded away from and : the letter chain has , so .
The abelianisation conditioning. Second, as and cancel, reads . Intervals are those of the linear word , of length ; a block wrapping across the end splits into at most two, so an arc occupies at most four intervals and a two-arc path at most eight. We need the conditioning on only through its denominator. Write , where is the set of reduced words of length . A word with is with and ; as has choices this gives . By (ii), for even one has , with depending only on , not on or ; only even can have . For the finitely many even the cost is bounded by a constant depending only on , absorbed by enlarging in (6). Hence, for ,
Hence , and Bayes with from (ii) gives uniformly in .
Transferring the pattern estimate. Let be a pattern event with . Bayes and give
using (iii) in the form ; the bound is relative to the unconditional pattern probability (not to , which need not hold). With , a pattern on a union of at most eight intervals with therefore obeys with : the from the bridge convolution cancels the same factor in the denominator. An arc or a path is specified by its position set, a union of at most eight intervals (for a path ). The patterns compatible with a fixed position set define disjoint events, and does not depend on : fix the position set of a candidate arc or path first, then sum the display over all compatible with , which costs at most the same constant since . Hence the arc bound survives as soon as and the path bound as soon as .
The remaining patterns have : an arc then has , and a two-arc path has . They are counted on the residual, of length , so there are arcs and paths; each has probability at most or , and conditioning multiplies this by at most : the Bayes factor over costs , since is bounded below, and the denominator was just computed. Since is uniformly bounded in , their total contribution is . Thus the expectation in (6) is in the conditional measure too, with larger constants; here , and the strip and the full-input sweep are already covered by (6).
Higher class: recursion and completeness
Peels are exchanges
3.1. Peels are exchanges. A peel is a triple of cut points : it splits into four blocks and keeps the reduced word
as the residual. There are triples, one nested loop fewer than the four cut points of (2) require; the range is the whole one, degenerate triples with an empty block included, and the exact peel of Theorem 3.6 is among them.
Proposition 3.1 (a peel removes one commutator). With , ,
so . Free reduction never lengthens a word, so , with strict inequality as soon as one of the junctions , , cancels.
Proof. Multiplying out and cancelling gives, in the free group,
these equalities survive reduction.
For the length bound, before reduction has the length of ; reduction never increases it, and a cancelling junction makes it strictly smaller (a cascade may delete more than two letters). This proves the proposition. □
Exact peels and completeness
Proposition 3.1 gives only , while the recursion needs a peel that lowers the class by exactly one at every node.
The proof runs in four steps, one per lemma below: pairings give the orbit count that computes (Proposition 3.2); a peel left-multiplies by a 3-cycle (Lemma 3.3); the gain is two exactly for one cyclic order of its entries (Lemma 3.4); a descent count produces such entries (Lemma 3.5).
Proposition 3.2 (the cycle-count formula, after Culler and Bardakov). Let be reduced with and , let . Call a fixed-point-free involution of with a pairing. Then
the maximum over all pairings, where counts orbits.
Proof. Pairings exist because the abelianisation of vanishes. Place one dart for each letter of around a single vertex, in the order , and glue them in pairs as prescribes. The resulting map has one vertex and edges. Walking a face alternates crossing an edge, which is , with turning to the next dart, which is ; so the faces are the orbits of and . Euler’s formula with and gives , and Culler’s theorem, in the permutation form given by Bardakov [1], identifies with the least genus of such a map. Minimising the genus means maximising . □
Let be an optimal pairing, one maximising , and put
Since is a conjugacy invariant we may take cyclically reduced: if and has , then , , , give , of the same class. A pairing pairs inverse letters, so cyclically reducedness restricts twice: would force at adjacent positions, and would force across the cyclic seam; both are forbidden. Hence every pair with has and , and since , the permutation has no fixed point.
Lemma 3.3 (a peel multiplies by a 3-cycle). Let with cut points $0 \le i < j < k \le M - 1$, let , and let be a pairing of . Transporting along the block move gives a map on whose face permutation has the orbit count of , where . Moreover
Proof. The block move is the permutation
which is order preserving inside each block. Hence equals except where the successor in leaves a block. Reading the three junctions off the new boundary gives (indices in the first column modulo , so is when )
| last dart of | ||
| (or if ) | ||
Table 1.
For the last dart of still goes to , as under ; for it is the row of the table and goes to . Since , evaluating gives
Table 2.
and elsewhere, so . For a concrete check, , has , , and otherwise, hence , , . Finally .
The transport is a conjugation: carries the letter at to position , so pairs two positions of carrying inverse letters, that is, it is again a pairing. Its face permutation is conjugate to , so the two have the same orbit count.
Reading the darts around the single vertex in cyclic order gives the word (equivalently, the dual one-face map has boundary ), which need not be reduced; it spells the element , and Culler’s theorem bounds by the genus of the map — it refers to the element, not to the spelling — so nothing asks to be reduced. □
Lemma 3.4 (when a 3-cycle adds two orbits). Let and let be any permutation. Then if lie on one -orbit in the cyclic order , and otherwise.
Proof. Write , the right factor acting first. For a transposition we have if and lie on one orbit of , and if they lie on two. Set and ; we apply these two rules in turn.
If lie on distinct orbits, merges them () and adds at most , so the net change is non-positive. Otherwise splits their orbit into the arc from to and the arc from to , a gain of 1. If the cyclic order is , then lies on the first arc and cuts again, giving 2. In the other order, or if lies on another orbit, and are on distinct orbits of and merges them, giving 0. □
Call with a 132-triple if its entries lie on one -orbit in the cyclic order . With , the two lemmas give
So an exact peel exists as soon as contains a 132-triple.
Lemma 3.5 (a 132-triple exists). Let be an optimal pairing of a cyclically reduced with . Then has a 132-triple.
Proof. Descents. Put and, for an orbit , let be its number of descents. We claim
Write each pair of as with . Since and ,
so , and the pairs give because . On an orbit the increments cancel, while exceeds by exactly when ; hence . Comparing the two expressions for gives . Each , because for all fails at the largest element of . Since there are orbits, .
Reading a triple off an orbit. By the claim some orbit has ; an orbit of two points has exactly one descent, so . Let be the least element of and write its -cycle as . Since is the least element and has no fixed point, . The last step is a descent, since it returns to the least element, and it is not the only one, so the least index with satisfies . Then . Put , and . Then , because (the cycle meets only at its two ends), and the cycle read from reaches before , so is a 132-triple. □
Theorem 3.6 (exact peel). Every reduced with has cut points $0 \le i \le j \le k \le |w|$ whose residual (7) satisfies ; the cut points may be taken with .
Proof. The cyclically reduced core. Write with cyclically reduced (possibly ), and let be an optimal pairing of ; thus and, for , with . Lemma 3.5 applied to and gives a 132-triple . Cut there and set .
The genus drops by one. Lemma 3.3 and (9), now applied to , give , so the one-vertex map obtained by transporting to has genus
and the map has boundary and genus . A pair in the boundary is a spur: in this one-vertex map it bounds a monogon face, so deleting the edge lowers both the edge count and the face count by one, with fixed. Hence and the genus are unchanged. Repeating removes every spur; the boundary becomes and the genus stays , so Proposition 3.2 gives . Reduction preserves the class, so — Culler’s theorem, as noted above, refers to the element, not the spelling.
Restoring the conjugation. Now put , , , , so , the cut points , , are positions of , and . For the residual of (7), reduction and conjugation preserve the class, so .
The reverse inequality. Conversely, multiplying out as in Proposition 3.1,
so and . The two bounds agree. □
The algorithm never invokes the theorem: it enumerates the triples of (7) and recurses on the survivors, which is complete by Lemma 3.7.
Lemma 3.7 (completeness). Let be a reduced residual of even length and trivial abelianisation, held by a node with target . Then the subtree below that node contains a leaf: a node whose target is , or one whose residual is empty.
Proof. Induct on . For the node is itself such a leaf: by Lemma 2.6 and Proposition 2.4 the base test decides correctly. For Theorem 3.6 applies, because , and gives a peel whose residual is reduced, even and of trivial abelianisation, a peel only permuting blocks. The filters reject only through (4), a lower bound for , so this candidate survives. If then and the child is a leaf with empty residual; otherwise , so and, as , induction applies to the child.
The filter stack
A candidate is screened by a single pass over . The count (3) is read on the cyclically reduced core of , so it is independent of the chosen anchor; it is computed afresh from itself, never carried over from the parent, and a cancellation at a junction cannot corrupt it. The pass strips the longest mirror-inverse prefix, the largest with (the of § 2), leaving a cyclically reduced core , counts the cyclic blocks of the core and forms the differences , in time and space, the order of producing from the four blocks of the parent. The candidate is rejected as soon as : by (4) this forces , hence , one more than the child’s target, and the rejection rests on (4) alone.
So a candidate costs and the filters do not enter the exponent of Theorem 3.8, only the constant; their role is practical, discarding a child whose class provably exceeds the target before any recursion starts.
The cost of the recursion
The recursion. A node holds a reduced residual of even length and a target . If it runs the base test of § 2; otherwise it tries every triple of (7), filters the candidates, and recurses on the survivors with target ; an empty residual ends its branch.
Theorem 3.8 (higher-class worst case). For a word of length and a target , the search runs in steps in the unit-cost model, in space.
Proof. Correctness. For the answer is the empty decomposition, so assume . The filters never discard a residual whose class is within its target: they reject only through (4). Hence Lemma 3.7 applies at every node that is reached, and the search returns a decomposition into at most commutators whenever . For it reports , which is correct: a leaf at depth at most would have exhibited as a product of at most commutators. Every returned answer is a product of commutators equal to : at an internal node by (8), at a base node once the stripped conjugation is put back.
The time bound. Let bound the residual length at depth . A node with target at least 2 enumerates triples and spends on each, so such a node costs ; a node with target 1 runs the base test, at by Proposition 2.5. The target, not the length, drops by one per level, so a node at depth has target and the tree has depth at most ; with and this gives , and . Hence
when . Here is the per-node cost: triples at each. The sum is dominated by its last term: the internal layers contribute against the of the base layer, so the factor spent on each candidate by the filter does not raise the exponent.
The space bound. The base test of Proposition 2.5 keeps the arc graph, which has arcs, while a peel node keeps its residual and the candidate it is screening, letters: the enumeration is depth first, recursing on a survivor as soon as it passes the single-pass filter and resuming where it left off on return, so no list of survivors is stored. The stack of residuals holds letters per level and the depth is at most , so it is as well, and the total is . □
Nor can two cut points suffice. A two-cut peel has three blocks and exchanges the last two, so a two-cut search enumerates such exchanges per level and would give . Ten letters defeat it. The word
is cyclically reduced of length 10 and a product of two commutators, but has no arc, so by Proposition 2.4. Its five inverse pairs sit at , , , , . A two-cut exchange changes only three adjacencies; checking the 36 nondegenerate cuts shows that every child has arcs only of length at most 1. Here a child is the residual of (7); reducing it to its cyclically reduced core costs at most one inverse pair, so that core still has length at least 8. Yet a -word of length at least 8 has arcs whose lengths sum to half , hence one of length at least 2. So no child is a -word, no child has class 1, and a two-cut search never reaches a leaf. The exponent is , not .
References
- [1]V. G. Bardakov, Computing commutator length in free groups, Algebra and Logic 39, No. 4 (2000), 224–251.DOI
- [2]L. Bartholdi, D. Fialkowski, S. O. Ivanov, On commutator length in free groups, Groups Geom. D., 18 (2024), 191–202.
- [3]M. Culler, Using surfaces to solve equations in free groups, Topology 20 (1981), 133–145.DOI
- [4]G. Manacher, A new linear-time “on-line” algorithm for finding the smallest initial palindrome of a string, J. Assoc. Comput. Mach. 22 (1975), 346–351.DOI
- [5]M. J. Wicks, Commutators in free products, J. London Math. Soc. 37 (1962), 433–444.DOI