Jelly, 7 4 bytes

æḟÆR 

Try it online!

How it works

æḟÆR Main link. Argument: n ÆR Prime range; yield all primes in [1, ..., n]. æḟ Power floor; round n down to the nearest power of each prime in the range. 

Mathematica, 44 43 40 bytes

Saved 4 bytes thanks to miles and Martin Ender

n#^⌊#~Log~n⌋&/@Select[Range@n,PrimeQ]

(x=Prime@Range@PrimePi@#)^⌊x~Log~#⌋& 

⌊ and ⌋ are the 3 byte characters U+230A and U+230B representing \[LeftFloor] and \[RightFloor], respectively.

Explanation:

Pure function. # is short for Slot[1] which represents the first argument to the Function. PrimePi@# counts the number of primes less or equal to #, Range@PrimePi@# is the list of the first PrimePi[#] positive integers, and so Prime@Range@PrimePi@# is the list of primes less than or equal to # (this is one byte shorter than Select[Range@#,PrimeQ]). The symbol x is Set equal to this list, then raised to the Power ⌊x~Log~#⌋, which is the list of Floor[Log[n,#]] for each n in x. In Mathematica, raising a list to the Power of another list of the same length results in the list of the powers of their corresponding elements.

MATL, 13 bytes

ZqG:!^tG>~*X> 

Try it Online!

Explanation

 % Implicitly grab the input as a number (N) Zq % Get an array of all primes below N G:! % Create an array from [1...N] ^ % Raise each prime to each power in this array which creates a 2D matrix % where the powers of each prime are down the columns tG>~ % Create a logical matrix that is TRUE where the values are less than N * % Perform element-wise multiplication to force values > N to zero X> % Compute the maximum value of each column % Implicitly display the resulting array 

Jelly, 12 9 bytes

RÆEz0iṀ$€ 

Try it online! (method is too slow for the 10000 case).

How?

Builds the list of p^k in order of p.

RÆEz0iṀ$€ - Main link: n e.g. 7 R - range(n) [1 ,2 ,3 ,4 ,5 ,6 ,7] ÆE - prime factor vector (vectorises) [[],[1],[0,1],[2],[0,0,1],[1,1],[0,0,0,1]] z0 - transpose with filler 0 [[0,1,0,2,0,1,0],[0,0,1,0,0,1,0],[0,0,0,0,1,0,0],[0,0,0,0,0,0,1]] $€ - la$t two links as a monad, for €ach ^ ^ ^ ^ i - first index of 4 3 5 7 Ṁ - maximum [4,3,5,7] 

Jelly, 8 bytes

ÆR*ÆE€»/ 

Try it online!

How it works

ÆR*ÆE€»/ Main link. Argument: n ÆR Prime range; yield all primes in [1, ..., n]. ÆE€ Prime exponents each; yield the exponents of 2, 3, 5, ... of the prime factorization of each k in [1, ..., n]. For example, 28 = 2² × 3⁰ × 5⁰ × 7¹ yields [2, 0, 0, 1]. * Exponentiation; raise the elements of the prime range to the powers of each of the exponents arrays. »/ Reduce the columns by maximum. 

Pyth, 13 bytes

m^ds.lQdfP_TS 

Try it here!

 f S - filter(v, range(1, input)) P_T - is_prime(T) m - map(v, ^) .lQd - log(input, d) s - int(^) ^d - d ** ^ 

I haven't played with Pyth in a while so any golfing tips are appreciated.

I couldn't get a shorter Mathematica solution than ngenisis's solution, but I thought I'd offer a couple of (hopefully interesting) alternative approaches.

Mathematica, 65 bytes

#/#2&@@@({#,#}&/@Range@#~Select~PrimeQ//.{a_,b_}/;a<=#:>{b a,b})& 

First we use {#,#}&/@Range@#~Select~PrimeQ to build a list of all primes in the appropriate range, but with ordered pairs of each prime, like { {2,2}, {3,3}, ...}. Then we operate on that list repeatedly with the replacement rule {a_,b_}/;a<=#:>{b a,b}, which multiplies the first element of the ordered pair by the second until the first element exceeds the input. Then we apply #/#2&@@@ to output, for each ordered pair, the first element divided by the second. (They end up sorted by the underlying prime: an example output is {16, 9, 5, 7, 11, 13, 17, 19}.)

Mathematica, 44 bytes

Values@Rest@<|MangoldtLambda@#->#&~Array~#|>& 

The von Mangoldt function Λ(n) is an interesting number theory function: it equals 0 unless n is a prime power p^k, in which case it equals log p (not log n). (These are natural logs, but it won't matter.) Thus MangoldtLambda@#->#&~Array~# creates an array of rules { 0->1, Log[2]->2, Log[3]->3, Log[2]->4, Log[5]->5, 0->6, ... } whose length is the input integer.

We then turn this list of rules into an "association" using <|...|>. This has the effect of retaining only the last rule with any given left-hand value. In other words, it will throw away Log[2]->2 and Log[2]->4 and Log[2]->8 and keep only Log[2]->16 (assuming that the input is between 16 and 31 for this example). Therefore the only remaining right-hand sides are the maximal prime powers—except for the one remaining rule 0->n, where n is the largest non-prime-power up to the input integer. But Rest throws that undesired rule away, and Values extracts the right-hand sides from the rules in the association. (They end up sorted as above.)

A slightly longer (46 bytes) version, which counts the number of appearances of each log p and then exponentiates to convert to the maximal prime powers:

E^(1##)&@@@Rest@Tally[MangoldtLambda~Array~#]& 

CJam, 21 20 bytes

Saved 1 byte thanks to Martin Ender

ri_){mp},f{\1$mLi#}p 

Try it online!

Explanation

ri e# Read an integer from input (let's call it n) _ e# Duplicate it ){mp}, e# Push the array of all prime numbers up to and including n f{ e# Map the following block to each prime p: \ e# Swap the top two elements of the stack 1$ e# Copy the second element down in the stack. Stack is now [p n p] mL e# Take the base-p logatithm of n i e# Cast to int (floor) # e# Raise p to that power } e# (end of map block) p e# Print 

05AB1E, 15 12 bytes

ƒNpiN¹N.nïm, 

Try it online!

Explanation

ƒ # for N in [0 ... input] Npi # if N is prime N # push N ¹N.n # push log_N(input) ï # convert to int m # raise N to this power , # print 

Brachylog, 15 bytes

⟧{ḋ=}ˢ⊇Xhᵐ≠∧X×ᵐ 

Try it online!

This outputs the powers from biggest to smallest.

This is very inefficient.

Explanation

⟧ The Range [Input, Input - 1, ..., 1, 0] { }ˢ Select: ḋ= The elements of that range whose prime decompositions are lists of the same element (e.g. [3,3,3]); output is those prime decompositions ⊇X X is an ordered subset of that list of prime decompositions Xhᵐ≠ All first elements of the prime decompositions are different (that is, X contains prime decompositions with different primes each times) ∧ X×ᵐ Output is the result of mapping multiplication to each element of X 

This will find the biggest prime decompositions for each prime first because of the way ⊇ works: from left to right and from biggest to smallest subsets.

Brachylog, 24 21 19 bytes

3+2 bytes saved thanks to Fatalize!

This is my first time using Brachylog, and I know some things could have been done in shorter ways, but I'm happy that it even works :D

{≥.~^ℕ₁ᵐhṗ:.≜×>?∧}ᶠ 

Try it online! (Return values are ordered by their base primes)

Explanation:

{................}ᶠ #Find all possible results of what's inside. ≥. #Input is >= than the output. .~^ℕ₁ᵐ #Output can be calculated as A^B, where A and B are #Natural numbers >=1. hṗ #The first of those numbers (A) is prime :.≜×>? #That same element (A), multiplied by the output #is greater than the input - This means #that B is the maximal exponent for A. ∧ #No more restrictions on the output. 

Jelly, 9 bytes

Ræl/ÆF*/€ 

One byte longer than my other answer, but finishes for input 10,000 is a couple of seconds.

Try it online!

How it works

Ræl/ÆF*/€ Main link. Argument: n R Range; yield [1, ..., n]. æl/ Reduce by least common multiple. ÆF Factor the LCM into [prime, exponent] pairs. */€ Reduce each pair by exponentiation. 

Bash + GNU utilities, 74 bytes

seq $1|factor|sed "s@.*: \(\w*\)\$@\1;l($1);l(\1);print \"/^p\"@"|bc -l|dc 

Try it online!

The input number is passed as an argument. The output is printed to stdout. (As is customary, stderr is ignored.)

Sample output:

./maximalprimepowers 100 2>/dev/null 64 81 25 49 11 13 17 19 23 29 31 37 41 43 47 53 59 61 67 71 73 79 83 89 97 ./maximalprimepowers 10000 2>/dev/null | wc -l 1229 

Here's how it works:

Call the argument N.

seq generates all the numbers from 1 to N, and factor factors them all.

The regex in the call to sed identifies those lines where the number is a prime P, and replaces those lines with lines that are of the form `

P;l(N);l(P);print "/^p" 

(where P and N are replaced by their actual numerical values, and everything else is copied literally, even the quotes and semicolons, and the string print).

These lines are fed as input to bc -l; bc prints the values of the three indicated numbers, each followed by a newline, and then prints the characters /^p . (In bc, l(x) denotes the natural logarithm of x.) JK K

The strings that bc prints are then fed as input to dc; dc prints the value of each P^(log(N)/log(P)) using integer arithmetic (truncating); that's the greatest power of P that is <= N.

One thing that's glossed over above is what happens to lines that are produced by factors that don't correspond to primes. Those lines don't match the regex in the call to sed, so no replacement is done on those. As a result, those lines start with a number followed by a colon, which generates an error when fed as input to bc. But bc just prints to stderr then, which we ignore; it doesn't print anything to stdout. By default, stderr is ignored on PPCG.

Haskell, 73 67 66 bytes

p n=[last[x^i|i<-[1..n],x^i<=n]|x<-[2..n],all((>0).mod x)[2..x-1]] 

Try it online! Usage:

Prelude> p 50 [32,27,25,49,11,13,17,19,23,29,31,37,41,43,47] 

Edit: 6 bytes off thanks to Zgarb!

Explanation:

p n=[... x|x<-[2..n] ] -- list of all x in the range 2 to n p n=[... x|x<-[2..n], mod x<$>[2..x-1]] -- where the remainders of x mod the numbers 2 to x-1 p n=[... x|x<-[2..n],all(>0)$mod x<$>[2..x-1]] -- are all greater 0. This yields all primes in the range. p n=[ [x^i|i<-[1..n] ]|...] -- for each of those x generate the list of all x^i with i in the range 1 to n p n=[last[x^i|i<-[1..n],x^i<=n]|...] -- where x^i is smaller or equal to n p n=[last[x^i|i<-[1..n],x^i<=n]|...] -- and take the last (that is the largest) element 

Regex (ECMAScript 2018 / Python^regex / .NET), 97 bytes

(?=(?=(xx+?)\1*$)(((?=(x+)(\4+$))\5)*\1$))(?<!(?=(?=(xx+?)(\6*$))(?=\7\1$)(?=(x+)\8+$)\8\9$)(x+)) 

Try it online! - ECMAScript 2018
Try it online! - Python (with regex)
Try it online! - .NET

Like Pseudofactorial, this uses Neil's prime powers test instead of mine, because when capturing the smallest prime factor is also needed, it results in a shorter total length.

 # tail = N ≤ M = input number # Assert tail is a prime power ≥ 2 (?= # Atomic lookahead (?=(xx+?)\1*$) # \1 = smallest prime factor of tail, # asserting that tail ≥ 2 ( # \2 = tail = N ( (?=(x+)(\4+$))\5 # tail = {largest proper divisor of tail} # = tail / {smallest prime factor of tail} )* # Iterate the above as many times necessary # (minimum zero) to make the following match: \1$ # Assert that tail == \1; if N is a prime power, # the end result of the above loop will always # be this. Otherwise it will be a larger prime, # and this will fail to match. ) ) # Assert there is no prime power (of the same base prime \1) # that is greater than N but less than or equal to M. (?<! # Negative Lookbehind - evaluated right-to-left, so read it # from bottom to top, but once entering the lookahead, go # back to left-to-right (top to bottom); assert that the # following cannot match: (?= # Lookahead (?=(xx+?)(\6*$)) # \6 = smallest prime factor of tail; \7 = tail-\6 (?= # Lookahead \7 # tail = tail - (tail - \6) = \6 \1$ # Assert tail == \1 ) (?=(x+)\8+$) # \8 = {largest proper divisor of tail} # = tail / {smallest prime factor of tail} \8\9$ # Assert tail == \8 + \9, meaning \8 == N ) (x+) # tail += \9 = any positive value, in which all possibilities # are tried, until a match is found, or all have been tried # and no matches were found) as long as the resulting tail ≤ M ) 

JavaScript (ES6), 118 120 119 114 112 105 bytes

(n,r)=>(r=k=>[...Array(k-1)].map((_,i)=>i+2),r(n).filter(q=>r(q).every(d=>q%d|!(d%(q/d)*(q/d)%d)&q*d>n))) 

Suggestions welcome. This is kind of long, but it seemed worth posting because it does all the divisibility testing explicitly rather than using built-ins related to primes.

Notes:

• A natural number q is a prime power <=> all of q's divisors are powers of the same prime <=> for any d which divides q, one of d and q/d is a divisor of the other.
• If q is a power of p, q is maximal <=> q * p > n <=> q * d > n for every nontrivial divisor d of q.

Sage, 43 bytes

lambda i:[x^int(ln(i,x))for x in primes(i)] 

Maps each prime in the range primes(i) to its maximal prime power. ln is just an alias of log so it accepts alternate bases although its name suggests it can only use base e.

Haskell, 110 90 bytes

s[]=[];s(x:t)=x:s[y|y<-t,y`rem`x>0];f n=[last.fst.span(<=n).scanl1(*)$repeat x|x<-s[2..n]] 

--updated per Laikoni's feedback

Haskell, 70 bytes

f n=[k|k<-[2..n],p:q<-[[d|d<-[2..k],mod k d<1]],k==p*p^length q,p*k>n] 

Defines a function f. Try it online!

Explanation

The idea is to filter the range [2..n] for those numbers k that satisfy k == p^length(divisors k) and p*k > n, where p is the smallest prime divisor of k.

f n= -- Define f n as [k| -- the list of numbers k, where k<-[2..n], -- k is drawn from [2..n], p:q<-[ -- the list p:q is drawn from [d| -- those lists of numbers d where d<-[2..k], -- d is drawn from [2..k] and mod k d<1] -- d divides k ], -- (so p:q are the divisors of k except 1, and p is the smallest one), k==p*p^length q, -- k equals p to the power of the divisor list's length -- (so it's in particular a prime power), and p*k>n] -- p*k, the next power of p, is not in the range [2..n]. 

Vyxal, 6 bytes

~æ~•⌊e 

Try it Online!

~ # Filter (implicit) range from 1 to n by æ # Is prime? ~ # Without popping • # Take the log of the input with base each of the primes ⌊ # Floor e # Raise the primes to those powers 

PHP, 101 93 91 88 bytes

just a little bit of real maths ...

for($n=$argv[$i=1];$n>$j=$i++;$j?:$r[$p=$i**~~log($n,$i)]=$p)for(;$i%$j--;);print_r($r); 

breakdown

for($n=$argv[$i=1]; // loop $i from 2 to $n $n>$j=$i++; // 0.: init $j to $i-1 $j?: // 2. if $i is prime $r[$p=$i**~~log($n,$i)]=$p) // 3. add maximum power to results for(;$i%$j--;); // 1. prime check (if $i is prime, $j will be 0) print_r($r); // print results 

JavaScript ES7, 93 bytes

Recursively iterate i from 0 up to and including n. If i is prime raise it to the highest floored exponent that makes it <= n (i ^ floor(log(n) / log(i)))

F=(n,i=n)=>i?[...((P=j=>i%--j?P(j):1==j)(i)?[i**((l=Math.log)(n)/l(i)|0)]:[]),...F(n,--i)]:[] 

05AB1E, 8 bytes

LÓ0ζεZk> 

Port of @JonathanAllan's Jelly answer.

Try it online or verify (almost) all test cases.

Or alternatively (and faster):

ÅPIL.¿Óm 

Port of @Dennis' (9-bytes) Jelly answer.

Try it online or verify all test cases.

Explanation:

L # Push a list in the range [1, (implicit) input] Ó # Map each value to a list of its prime exponents ζ # Zip/transpose, swapping rows/columns, 0 # using a 0 as filler-item for unequal length rows ε # Map over each inner list: Z # Push the maximum (without popping the list) k # Get the (0-based) index of this maximum in the list > # Increase it by 1 to make it a 1-based index # (after which the resulting list is output implicitly) ÅP # Push a list of all primes smaller than or equal to the (implicit) input IL # Push a list in the range [1,input] .¿ # Pop and get its Least Common Multiple (lcm) Ó # Get the prime exponents of this lcm m # Take each prime to each prime exponent at the same positions # (after which the resulting list is output implicitly) 

APL(NARS), 19 chars

{a*⌊⍵⍟⍨a←k/⍨0πk←⍳⍵} 

It use the fact if p is a number >1 r=floor(log_p(InputNumber)) is the max number with p^r<InputNumber. test:

 f←{a*⌊⍵⍟⍨a←k/⍨0πk←⍳⍵} ≢f 10000 1229 f 20 16 9 5 7 11 13 17 19 

J-uby -rprime, 34 bytes

Port of Dennis' 9-byte Jelly solution; give him an upvote.

:+|:/&:lcm|:prime_division|:*&+:** 

Attempt This Online!

Explanation

:+ | :/ & :lcm | :prime_division | :* & +:**­⁡​‎‎⁡⁠⁡‏⁠‎⁡⁠⁢‏⁠‎⁡⁠⁣‏⁠‎⁡⁠⁤‏⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠‏​⁡⁠⁡‌⁢​‎‎⁡⁠⁢⁢‏⁠‎⁡⁠⁢⁣‏⁠‎⁡⁠⁢⁤‏⁠‎⁡⁠⁣⁡‏⁠‎⁡⁠⁣⁢‏⁠‎⁡⁠⁣⁣‏⁠‎⁡⁠⁣⁤‏⁠‎⁡⁠⁤⁡‏⁠‎⁡⁠⁤⁢‏⁠‎⁡⁠⁤⁣‏⁠‎⁡⁠⁤⁤‏‏​⁡⁠⁡‌⁣​‎‎⁡⁠⁢⁡⁢‏⁠‎⁡⁠⁢⁡⁣‏⁠‎⁡⁠⁢⁡⁤‏⁠‎⁡⁠⁢⁢⁡‏⁠‎⁡⁠⁢⁢⁢‏⁠‎⁡⁠⁢⁢⁣‏⁠‎⁡⁠⁢⁢⁤‏⁠‎⁡⁠⁢⁣⁡‏⁠‎⁡⁠⁢⁣⁢‏⁠‎⁡⁠⁢⁣⁣‏⁠‎⁡⁠⁢⁣⁤‏⁠‎⁡⁠⁢⁤⁡‏⁠‎⁡⁠⁢⁤⁢‏⁠‎⁡⁠⁢⁤⁣‏⁠‎⁡⁠⁢⁤⁤‏⁠‎⁡⁠⁣⁡⁡‏⁠‎⁡⁠⁣⁡⁢‏‏​⁡⁠⁡‌⁤​‎‎⁡⁠⁣⁡⁤‏⁠‎⁡⁠⁣⁢⁡‏⁠‎⁡⁠⁣⁢⁢‏⁠‎⁡⁠⁣⁢⁣‏⁠‎⁡⁠⁣⁢⁤‏⁠‎⁡⁠⁣⁣⁡‏⁠‎⁡⁠⁣⁣⁢‏⁠‎⁡⁠⁣⁣⁣‏⁠‎⁡⁠⁣⁣⁤‏⁠‎⁡⁠⁣⁤⁡‏⁠‎⁡⁠⁣⁤⁢‏⁠‎⁡⁠⁣⁤⁣‏⁠‎⁡⁠⁣⁤⁤‏⁠‎⁡⁠⁤⁡⁡‏⁠‎⁡⁠⁤⁡⁢‏⁠‎⁡⁠⁤⁡⁣‏‏​⁡⁠⁡‌­ :+ | # ‎⁡Range 1..n, then :/ & :lcm | # ‎⁢reduce by LCM, then :prime_division | # ‎⁣get prime factors as [factor, exponent] pairs, then :* & +:** # ‎⁤map each pair as exponentiation 💎 

Created with the help of Luminespire.