Radix sort ($T(n,d,R)=O(d(n+R))$)

Radix sort sorts encoded sort keys a digit at a time. Its least-significant-digit version uses stable sorting on each successive digit, often by counting sort. For $d$ digits with $R$ possible values each, it takes $O(d(n+R))$ time complexity in Big O notation, assuming constant-cost digit extraction and indexing. Stable sorting retains the ordering established by earlier digits when a later digit ties. Both digit access and the direct indexing in counting sort lie outside the comparison sort model, so these bounds do not contradict the comparison sorting lower bound.^[ref]

A least-significant-digit radix sort processes numerical digits from smallest to largest place value. Fix integers $R\ge 2$ and $d\ge 1$, and take $n$ complete sorting records whose sort keys satisfy $0\le k<R^d$. Use exactly $d$ positions in positional notation with radix $R$, including leading zeros. At pass $j=0,\dots ,d-1$, apply a stable digit sort to the current array usingThe floor function selects the integer quotient, and the remainder selects the current numerical digit. Move the complete sorting record, including its associated data. There are exactly $d$ passes, so this procedure terminates even for an empty array. The radix sort suffix invariant supplies its algorithm correctness argument.

For an original input, let the letters identify the attached data:Here $R=10$, $d=3$, and leading zeros only align place values; $005$ represents the integer five. The exact pass ledger is:After the units pass, the numerical digits are $0,2,4,5,5,7,7,7$. The ties retain $C$ before $F$ and $B$ before $E$ before $G$. This array is not yet ordered by its whole integer sort keys: $314$ still precedes $305$. After the tens pass, the two-position suffixes are $05,05,14,17,27,27,40,42$. In particular, $217_E$ now precedes $027_B$: their tens numerical digits differ, so stable sorting does not require retaining their previous order. The hundreds pass produces the final sorted array, with all attached letters preserved.

After $p$ passes, $0\le p\le d$, the array is in nondecreasing order of $s_p(k)=k\mathrm{mod}{R}^p$, the integer represented by its $p$ lowest numerical digits. Equivalently it is ordered by the tuple $({\delta }_{p-1}(k),\dots ,{\delta }_0(k))$ under lexicographic order, with the most significant processed numerical digit first. Among equal suffixes it retains the original order of the complete sorting records.

The loop invariant has a simple mathematical induction proof. Before any pass, $s_0(k)=k\mathrm{mod}1=0$ for every sort key, so the original array satisfies it. Suppose it holds after $p$ passes. The next stable digit sort puts smaller ${\delta }_p$ groups first and preserves the existing $s_p$ order within each tie. Sincethe new array is ordered by $s_{p+1}$. Equal new suffixes have equal current numerical digits and equal old suffixes, so their original order is also preserved. Every pass is a permutation of complete sorting records, establishing sorting permutation preservation. At termination $p=d$, the assumption $k<R^d$ gives $s_d(k)=k$: the invariant becomes whole-key sortedness, together with stable sorting.

A stable digit sort applies stable sorting to complete sorting records using one specified numerical digit of their encoded sort keys. Counting sort can supply it; stable record placement in counting sort moves records without discarding their associated data. Stability concerns equal current numerical digits, even when the full sort keys differ. This is the information-preservation step in least-significant-digit radix sort.^[ref]

An original two-record counterexample exposes the need for the tie rule. Start with $(42_A,41_B)$ at width three. A units pass gives $(041_B,042_A)$. Both tens numerical digits are four. A deliberately unstable tens pass may reverse their tie and return $(042_A,041_B)$: it is sorted by the tens numerical digit, yet its two-position suffixes $42,41$ are decreasing. The final stable hundreds pass sees two zeros and keeps this wrong order. Sorting permutation preservation still holds, but sortedness fails. Stability in the tens pass would have retained the units information.

With counting sort as the stable digit sort, each pass initializes $R$ counters, counts $n$ current numerical digits, constructs their prefix sums, and places $n$ complete sorting records. Thus each pass costs $O(n+R)$, and $d$ passes cost $O(d(n+R))$ in Big O notation. Reuse the counters and two arrays of record references between passes: auxiliary space complexity is $O(n+R)$, rather than $O(d(n+R))$. Assume constant-cost numerical digit extraction, counter arithmetic, array indexing and record-reference movement; the bound counts these operations, not the cost of copying arbitrarily large payloads or performing arbitrary-precision integer arithmetic.

The benefit is avoiding a counter for every possible whole sort key: width $d$ has $R^d$ possibilities, while each pass needs only $R$ counters. A larger radix reduces the number of passes for a fixed key range but enlarges the counter array. The comparison sorting lower bound does not apply to these direct array indices; they reveal more than a two-way comparison. Nor does fixing $R$ make $d$ disappear: growing key width still increases work.

An encoding for radix sort must make the intended total order of sort keys agree with the lexicographic order of its fixed-width numerical digit tuples. The positional notation for bounded nonnegative integers has this property. At the most significant differing position $m$, a one-unit numerical digit difference contributes $R^m$, while all lower positions together contribute at mostThis geometric series shows why lower numerical digits cannot reverse that comparison. Leading zero padding works here because it preserves place values and the represented integer.

For signed integers in a known integer interval $L\le x\le U$, encode $e(x)=x-L$ and choose $d$ with $U-L<R^d$. This order-preserving function is also injective: $x<y$ exactly when $e(x)<e(y)$. For $-8\le x\le 7$, the four-bit encodings of $-8,-3,-1,0,4,7$ are $0000,0101,0111,1000,1100,1111$ after adding eight. Restore the original signed sort keys when displaying the complete sorting records; the encoding need not replace their data. Sorting the unsigned values of raw two's complement bit patterns would put nonnegative integers before negative ones, so their representation alone is not this ordered encoding.

For variable-length strings, first specify the alphabet order and whether a proper prefix precedes its extensions. As an original construction, use $a<b$, let every string have length at most three, encode $a$ as one and $b$ as two, and append zeros to width three. Zero is reserved for an end marker strictly below every real symbol. The strings $\epsilon ,a,aa,ab,b$ encode as $000,100,110,120,200$ in radix three, which preserves their lexicographic order. A differing real symbol decides the comparison; at a prefix boundary, the shorter string's zero comes before the extension's one or two. Left padding would instead encode $b$ as $002$ and $aa$ as $011$, reversing $aa<b$. Padding raw bytes with zero also fails if zero is an allowed data symbol: distinct strings can then receive the same encoding. A reserved end marker and an explicit symbol ordering are essential; byte order alone does not establish a desired human text order.