Shellsort

Shellsort is an optimization of insertion sort that allows the exchange of items that are far apart. The idea is to arrange the list of elements so that, starting anywhere, taking every hth element produces a sorted list. Such a list is said to be h-sorted. It can also be thought of as h interleaved lists, each individually sorted.[6] Beginning with large values of h allows elements to move long distances in the original list, reducing large amounts of disorder quickly, and leaving less work for smaller h-sort steps to do.[7] If the list is then k-sorted for some smaller integer k, then the list remains h-sorted. Following this idea for a decreasing sequence of h values ending in 1 is guaranteed to leave a sorted list in the end.[6]

In simplistic terms, this means if we have an array of 1024 numbers, our first gap (h) could be 512. We then run through the list comparing each element in the first half to the element in the second half. Our second gap (k) is 256, which breaks the array into four sections (starting at 0,256,512,768), and we make sure the first items in each section are sorted relative to each other, then the second item in each section, and so on. In practice the gap sequence could be anything, but the last gap is always 1 to finish the sort.

An example run of Shellsort with gaps 5, 3 and 1 is shown below.

The first pass, 5-sorting, performs insertion sort on five separate subarrays (a₁, a₆, a₁₁), (a₂, a₇, a₁₂), (a₃, a₈), (a₄, a₉), (a₅, a₁₀). For instance, it changes the subarray (a₁, a₆, a₁₁) from (62, 17, 25) to (17, 25, 62). The next pass, 3-sorting, performs insertion sort on the three subarrays (a₁, a₄, a₇, a₁₀), (a₂, a₅, a₈, a₁₁), (a₃, a₆, a₉, a₁₂). The last pass, 1-sorting, is an ordinary insertion sort of the entire array (a₁,..., a₁₂).

As the example illustrates, the subarrays that Shellsort operates on are initially short; later they are longer but almost ordered. In both cases insertion sort works efficiently.

Shellsort is not stable: it may change the relative order of elements with equal values. It is an adaptive sorting algorithm in that it executes faster when the input is partially sorted.

Using Marcin Ciura's gap sequence, with an inner insertion sort.

# Sort an array a[0...n-1].
gaps = [701, 301, 132, 57, 23, 10, 4, 1]

# Start with the largest gap and work down to a gap of 1
foreach (gap in gaps)
{
    # Do a gapped insertion sort for this gap size.
    # The first gap elements a[0..gap-1] are already in gapped order
    # keep adding one more element until the entire array is gap sorted
    for (i = gap; i < n; i += 1)
    {
        # add a[i] to the elements that have been gap sorted
        # save a[i] in temp and make a hole at position i
        temp = a[i]
        # shift earlier gap-sorted elements up until the correct location for a[i] is found
        for (j = i; j >= gap and a[j - gap] > temp; j -= gap)
        {
            a[j] = a[j - gap]
        }
        # put temp (the original a[i]) in its correct location
        a[j] = temp
    }
}

The question of deciding which gap sequence to use is difficult. Every gap sequence that contains 1 yields a correct sort (as this makes the final pass an ordinary insertion sort); however, the properties of thus obtained versions of Shellsort may be very different. Too few gaps slows down the passes, and too many gaps produces an overhead.

The table below compares most proposed gap sequences published so far. Some of them have decreasing elements that depend on the size of the sorted array (N). Others are increasing infinite sequences, whose elements less than N should be used in reverse order.

OEIS	General term (k ≥ 1)	Concrete gaps	Worst-case time complexity	Author and year of publication
	${\displaystyle \left\lfloor {\frac {N}{2^{k}}}\right\rfloor }$	${\displaystyle \left\lfloor {\frac {N}{2}}\right\rfloor ,\left\lfloor {\frac {N}{4}}\right\rfloor ,\ldots ,1}$	${\displaystyle \Theta \left(N^{2}\right)}$ [e.g. when N = 2p]	Shell, 1959[4]
	${\displaystyle 2\left\lfloor {\frac {N}{2^{k+1}}}\right\rfloor +1}$	${\displaystyle 2\left\lfloor {\frac {N}{4}}\right\rfloor +1,\ldots ,3,1}$	${\displaystyle \Theta \left(N^{\frac {3}{2}}\right)}$	Frank & Lazarus, 1960[8]
A000225	${\displaystyle 2^{k}-1}$	${\displaystyle 1,3,7,15,31,63,\ldots }$	${\displaystyle \Theta \left(N^{\frac {3}{2}}\right)}$	Hibbard, 1963[9]
A083318	${\displaystyle 2^{k}+1}$, prefixed with 1	${\displaystyle 1,3,5,9,17,33,65,\ldots }$	${\displaystyle \Theta \left(N^{\frac {3}{2}}\right)}$	Papernov & Stasevich, 1965[10]
A003586	Successive numbers of the form ${\displaystyle 2^{p}3^{q}}$ (3-smooth numbers)	${\displaystyle 1,2,3,4,6,8,9,12,\ldots }$	${\displaystyle \Theta \left(N\log ^{2}N\right)}$	Pratt, 1971[1]
A003462	${\displaystyle {\frac {3^{k}-1}{2}}}$, not greater than ${\displaystyle \left\lceil {\frac {N}{3}}\right\rceil }$	${\displaystyle 1,4,13,40,121,\ldots }$	${\displaystyle \Theta \left(N^{\frac {3}{2}}\right)}$	Knuth, 1973,[3] based on Pratt, 1971[1]
A036569	${\displaystyle {\begin{aligned}&\prod \limits _{I}a_{q},{\hbox{where}}\\a_{q}={}&\min \left\{n\in \mathbb {N} \colon n\geq \left({\frac {5}{2}}\right)^{q+1},\forall p\colon 0\leq p<q\Rightarrow \gcd(a_{p},n)=1\right\}\\I={}&\left\{0\leq q<r\mid q\neq {\frac {1}{2}}\left(r^{2}+r\right)-k\right\}\\r={}&\left\lfloor {\sqrt {2k+{\sqrt {2k}}}}\right\rfloor \end{aligned}}}$	${\displaystyle 1,3,7,21,48,112,\ldots }$	${\displaystyle O\left(N^{1+{\sqrt {\frac {8\ln \left(5/2\right)}{\ln(N)}}}}\right)}$	Incerpi & Sedgewick, 1985,[11] Knuth[3]
A036562	${\displaystyle 4^{k}+3\cdot 2^{k-1}+1}$, prefixed with 1	${\displaystyle 1,8,23,77,281,\ldots }$	${\displaystyle O\left(N^{\frac {4}{3}}\right)}$	Sedgewick, 1982[6]
A033622	${\displaystyle {\begin{cases}9\left(2^{k}-2^{\frac {k}{2}}\right)+1&k{\text{ even}},\\8\cdot 2^{k}-6\cdot 2^{(k+1)/2}+1&k{\text{ odd}}\end{cases}}}$	${\displaystyle 1,5,19,41,109,\ldots }$	${\displaystyle O\left(N^{\frac {4}{3}}\right)}$	Sedgewick, 1986[12]
	${\displaystyle h_{k}=\max \left\{\left\lfloor {\frac {5h_{k-1}}{11}}\right\rfloor ,1\right\},h_{0}=N}$	${\displaystyle \left\lfloor {\frac {5N}{11}}\right\rfloor ,\left\lfloor {\frac {5}{11}}\left\lfloor {\frac {5N}{11}}\right\rfloor \right\rfloor ,\ldots ,1}$	Unknown	Gonnet & Baeza-Yates, 1991[13]
A108870	${\displaystyle \left\lceil {\frac {1}{5}}\left(9\cdot \left({\frac {9}{4}}\right)^{k-1}-4\right)\right\rceil }$	${\displaystyle 1,4,9,20,46,103,\ldots }$	Unknown	Tokuda, 1992[14]
A102549	Unknown (experimentally derived)	${\displaystyle 1,4,10,23,57,132,301,701}$	Unknown	Ciura, 2001[15]

When the binary representation of N contains many consecutive zeroes, Shellsort using Shell's original gap sequence makes Θ(N²) comparisons in the worst case. For instance, this case occurs for N equal to a power of two when elements greater and smaller than the median occupy odd and even positions respectively, since they are compared only in the last pass.

Although it has higher complexity than the O(N log N) that is optimal for comparison sorts, Pratt's version lends itself to sorting networks and has the same asymptotic gate complexity as Batcher's bitonic sorter.

Gonnet and Baeza-Yates observed that Shellsort makes the fewest comparisons on average when the ratios of successive gaps are roughly equal to 2.2.[13] This is why their sequence with ratio 2.2 and Tokuda's sequence with ratio 2.25 prove efficient. However, it is not known why this is so. Sedgewick recommends to use gaps that have low greatest common divisors or are pairwise coprime.[16]

With respect to the average number of comparisons, Ciura's sequence[15] has the best known performance; gaps from 701 were not determined but the sequence can be further extended according to the recursive formula ${\displaystyle h_{k}=\lfloor 2.25h_{k-1}\rfloor }$.

Tokuda's sequence, defined by the simple formula ${\displaystyle h_{k}=\lceil h'_{k}\rceil }$, where ${\displaystyle h'_{k}=2.25h'_{k-1}+1}$, ${\displaystyle h'_{1}=1}$, can be recommended for practical applications.

The following property holds: after h₂-sorting of any h₁-sorted array, the array remains h₁-sorted.[17] Every h₁-sorted and h₂-sorted array is also (a₁h₁+a₂h₂)-sorted, for any nonnegative integers a₁ and a₂. The worst-case complexity of Shellsort is therefore connected with the Frobenius problem: for given integers h₁,..., h_n with gcd = 1, the Frobenius number g(h₁,..., h_n) is the greatest integer that cannot be represented as a₁h₁+ ... +a_nh_n with nonnegative integer a₁,..., a_n. Using known formulae for Frobenius numbers, we can determine the worst-case complexity of Shellsort for several classes of gap sequences.[18] Proven results are shown in the above table.

With respect to the average number of operations, none of the proven results concerns a practical gap sequence. For gaps that are powers of two, Espelid computed this average as ${\displaystyle 0.5349N{\sqrt {N}}-0.4387N-0.097{\sqrt {N}}+O(1)}$.[19] Knuth determined the average complexity of sorting an N-element array with two gaps (h, 1) to be ${\displaystyle {\frac {2N^{2}}{h}}+{\sqrt {\pi N^{3}h}}}$.[3] It follows that a two-pass Shellsort with h = Θ(N^1/3) makes on average O(N^5/3) comparisons/inversions/running time. Yao found the average complexity of a three-pass Shellsort.[20] His result was refined by Janson and Knuth:[21] the average number of comparisons/inversions/running time made during a Shellsort with three gaps (ch, cg, 1), where h and g are coprime, is ${\displaystyle {\frac {N^{2}}{4ch}}+O(N)}$ in the first pass, ${\displaystyle {\frac {1}{8g}}{\sqrt {\frac {\pi }{ch}}}(h-1)N^{3/2}+O(hN)}$ in the second pass and ${\displaystyle \psi (h,g)N+{\frac {1}{8}}{\sqrt {\frac {\pi }{c}}}(c-1)N^{3/2}+O\left((c-1)gh^{1/2}N\right)+O\left(c^{2}g^{3}h^{2}\right)}$ in the third pass. ψ(h, g) in the last formula is a complicated function asymptotically equal to ${\displaystyle {\sqrt {\frac {\pi h}{128}}}g+O\left(g^{-1/2}h^{1/2}\right)+O\left(gh^{-1/2}\right)}$. In particular, when h = Θ(N^7/15) and g = Θ(N^1/5), the average time of sorting is O(N^23/15).

Based on experiments, it is conjectured that Shellsort with Hibbard's gap sequence runs in O(N^5/4) average time,[3] and that Gonnet and Baeza-Yates's sequence requires on average 0.41NlnN(ln lnN+1/6) element moves.[13] Approximations of the average number of operations formerly put forward for other sequences fail when sorted arrays contain millions of elements.

The graph below shows the average number of element comparisons in various variants of Shellsort, divided by the theoretical lower bound, i.e. log₂N!, where the sequence 1, 4, 10, 23, 57, 132, 301, 701 has been extended according to the formula ${\displaystyle h_{k}=\lfloor 2.25h_{k-1}\rfloor }$.

Applying the theory of Kolmogorov complexity, Jiang, Li, and Vitányi proved the following lower bound for the order of the average number of operations/running time in a p-pass Shellsort: Ω(pN^1+1/p) when p≤log₂N and Ω(pN) when p>log₂N.[22] Therefore, Shellsort has prospects of running in an average time that asymptotically grows like NlogN only when using gap sequences whose number of gaps grows in proportion to the logarithm of the array size. It is, however, unknown whether Shellsort can reach this asymptotic order of average-case complexity, which is optimal for comparison sorts. The lower bound was improved by Vitányi[23] for every number of passes ${\displaystyle p}$ to ${\displaystyle \Omega (N\sum _{k=1}^{p}h_{k-1}/h_{k})}$ where ${\displaystyle h_{0}=N}$. This result implies for example the Jiang-Li-Vitányi lower bound for all ${\displaystyle p}$-pass increment sequences and improves that lower bound for particular increment sequences. In fact all bounds (lower and upper) currently known for the average case are precisely matched by this lower bound. For example, this gives the new result that the Janson-Knuth upper bound is matched by the resulting lower bound for the used increment sequence, showing that three pass Shellsort for this increment sequence uses ${\displaystyle \Theta (N^{23/15})}$ comparisons/inversions/running time. The formula allows us to search for increment sequences that yield lower bounds which are unknown; for example an increment sequence for four passes which has a lower bound greater than ${\displaystyle \Omega (pn^{1+1/p})=\Omega (n^{5/4})}$ for the increment sequence ${\displaystyle h_{1}=n^{11/16},h_{2}=n^{7/16},h_{3}=n^{3/16},h_{4}=1}$. The lower bound becomes ${\displaystyle T=\Omega (n\cdot (n^{1-11/16}+n^{11/16-7/16}+n^{7/16-3/16}+n^{3/16})=\Omega (n^{1+5/16})=\Omega (n^{21/16}).}$

The worst-case complexity of any version of Shellsort is of higher order: Plaxton, Poonen, and Suel showed that it grows at least as rapidly as ${\displaystyle \Omega \left(N\left({{\log N} \over {\log \log N}}\right)^{2}\right)}$.[24]

Shellsort performs more operations and has higher cache miss ratio than quicksort. However, since it can be implemented using little code and does not use the call stack, some implementations of the qsort function in the C standard library targeted at embedded systems use it instead of quicksort. Shellsort is, for example, used in the uClibc library.[25] For similar reasons, in the past, Shellsort was used in the Linux kernel.[26]

Shellsort can also serve as a sub-algorithm of introspective sort, to sort short subarrays and to prevent a slowdown when the recursion depth exceeds a given limit. This principle is employed, for instance, in the bzip2 compressor.[27]

• Comb sort

• Knuth, Donald E. (1997). "Shell's method". The Art of Computer Programming. Volume 3: Sorting and Searching (2nd ed.). Reading, Massachusetts: Addison-Wesley. pp. 83–95. ISBN 978-0-201-89685-5.
• Analysis of Shellsort and Related Algorithms, Robert Sedgewick, Fourth European Symposium on Algorithms, Barcelona, September 1996.

The Wikibook Algorithm implementation has a page on the topic of: Shell sort

• Animated Sorting Algorithms: Shell Sort at the Wayback Machine (archived 10 March 2015) – graphical demonstration
• Shellsort with gaps 5, 3, 1 as a Hungarian folk dance