import macros;macro ImportExpand(s:untyped):untyped = parseStmt($s[2])
# {.checks: off.}
ImportExpand "cplib/tmpl/citrus.nim" <=== "when not declared CPLIB_TMPL_CITRUS:\n    const CPLIB_TMPL_CITRUS* = 1\n    {.warning[UnusedImport]: off.}\n    {.hint[XDeclaredButNotUsed]: off.}\n    import os\n    import algorithm\n    import sequtils\n    import tables\n    import macros\n    import std/math\n    import sets\n    import strutils\n    import strformat\n    import sugar\n    import streams\n    import deques\n    import bitops\n    import heapqueue\n    import options\n    import hashes\n    const MODINT998244353* = 998244353\n    const MODINT1000000007* = 1000000007\n    const INF* = 100100111\n    const INFL* = int(3300300300300300491)\n    type double* = float64\n    let readNext = iterator(getsChar: bool = false): string {.closure.} =\n        while true:\n            var si: string\n            try: si = stdin.readLine\n            except EOFError: yield \"\"\n            for s in si.split:\n                if getsChar:\n                    for i in 0..<s.len():\n                        yield s[i..i]\n                else:\n                    if s.isEmptyOrWhitespace: continue\n                    yield s\n    proc input*(t: typedesc[string]): string = readNext()\n    proc input*(t: typedesc[char]): char = readNext(true)[0]\n    proc input*(t: typedesc[int]): int = readNext().parseInt\n    proc input*(t: typedesc[float]): float = readNext().parseFloat\n    macro input*(t: typedesc, n: varargs[int]): untyped =\n        var repStr = \"\"\n        for arg in n:\n            repStr &= &\"({arg.repr}).newSeqWith \"\n        parseExpr(&\"{repStr}input({t})\")\n    macro input*(ts: varargs[auto]): untyped =\n        var tupStr = \"\"\n        for t in ts:\n            tupStr &= &\"input({t.repr}),\"\n        parseExpr(&\"({tupStr})\")\n    macro input*(n: int, ts: varargs[auto]): untyped =\n        for typ in ts:\n            if typ.typeKind != ntyAnything:\n                error(\"Expected typedesc, got \" & typ.repr, typ)\n        parseExpr(&\"({n.repr}).newSeqWith input({ts.repr})\")\n    proc `fmtprint`*(x: int or string or char or bool): string = return $x\n    proc `fmtprint`*(x: float or float32 or float64): string = return &\"{x:.16f}\"\n    proc `fmtprint`*[T](x: seq[T] or Deque[T] or HashSet[T] or set[T]): string = return x.toSeq.join(\" \")\n    proc `fmtprint`*[T, N](x: array[T, N]): string = return x.toSeq.join(\" \")\n    proc `fmtprint`*[T](x: HeapQueue[T]): string =\n        var q = x\n        while q.len != 0:\n            result &= &\"{q.pop()}\"\n            if q.len != 0: result &= \" \"\n    proc `fmtprint`*[T](x: CountTable[T]): string =\n        result = x.pairs.toSeq.mapIt(&\"{it[0]}: {it[1]}\").join(\" \")\n    proc `fmtprint`*[K, V](x: Table[K, V]): string =\n        result = x.pairs.toSeq.mapIt(&\"{it[0]}: {it[1]}\").join(\" \")\n    proc print*(prop: tuple[f: File, sepc: string, endc: string, flush: bool], args: varargs[string, `fmtprint`]) =\n        for i in 0..<len(args):\n            prop.f.write(&\"{args[i]}\")\n            if i != len(args) - 1: prop.f.write(prop.sepc) else: prop.f.write(prop.endc)\n        if prop.flush: prop.f.flushFile()\n    proc print*(args: varargs[string, `fmtprint`]) = print((f: stdout, sepc: \" \", endc: \"\\n\", flush: false), args)\n    const LOCAL_DEBUG{.booldefine.} = false\n    macro getSymbolName(x: typed): string = x.toStrLit\n    macro debug*(args: varargs[untyped]): untyped =\n        when LOCAL_DEBUG:\n            result = newNimNode(nnkStmtList, args)\n            template prop(e: string = \"\"): untyped = (f: stderr, sepc: \"\", endc: e, flush: true)\n            for i, arg in args:\n                if arg.kind == nnkStrLit:\n                    result.add(quote do: print(prop(), \"\\\"\", `arg`, \"\\\"\"))\n                else:\n                    result.add(quote do: print(prop(\": \"), getSymbolName(`arg`)))\n                    result.add(quote do: print(prop(), `arg`))\n                if i != args.len - 1: result.add(quote do: print(prop(), \", \"))\n                else: result.add(quote do: print(prop(), \"\\n\"))\n        else:\n            return (quote do: discard)\n    proc `%`*(x: SomeInteger, y: SomeInteger): int =\n        result = x mod y\n        if y > 0 and result < 0: result += y\n        if y < 0 and result > 0: result += y\n    proc `//`*(x: SomeInteger, y: SomeInteger): int =\n        result = x div y\n        if y > 0 and result * y > x: result -= 1\n        if y < 0 and result * y < x: result -= 1\n    proc `^`*(x: SomeInteger, y: SomeInteger): int = x xor y\n    proc `&`*(x: SomeInteger, y: SomeInteger): int = x and y\n    proc `|`*(x: SomeInteger, y: SomeInteger): int = x or y\n    proc `>>`*(x: SomeInteger, y: SomeInteger): int = x shr y\n    proc `<<`*(x: SomeInteger, y: SomeInteger): int = x shl y\n    proc `%=`*(x: var SomeInteger, y: SomeInteger): void = x = x % y\n    proc `//=`*(x: var SomeInteger, y: SomeInteger): void = x = x // y\n    proc `^=`*(x: var SomeInteger, y: SomeInteger): void = x = x ^ y\n    proc `&=`*(x: var SomeInteger, y: SomeInteger): void = x = x & y\n    proc `|=`*(x: var SomeInteger, y: SomeInteger): void = x = x | y\n    proc `>>=`*(x: var SomeInteger, y: SomeInteger): void = x = x >> y\n    proc `<<=`*(x: var SomeInteger, y: SomeInteger): void = x = x << y\n    proc `[]`*(x, n: int): bool = (x and (1 shl n)) != 0\n    proc `[]=`*(x: var int, n: int, i: bool) =\n        if i: x = x or (1 << n)\n        else: (if x[n]: x = x xor (1 << n))\n    proc pow*(a, n: int, m = INFL): int =\n        var\n            rev = 1\n            a = a\n            n = n\n        while n > 0:\n            if n % 2 != 0: rev = (rev * a) mod m\n            if n > 1: a = (a * a) mod m\n            n >>= 1\n        return rev\n    #[ include cplib/math/isqrt ]#\n    when not declared CPLIB_MATH_ISQRT:\n        const CPLIB_MATH_ISQRT* = 1\n        proc isqrt*(n: int): int =\n            var x = n\n            var y = (x + 1) shr 1\n            while y < x:\n                x = y\n                y = (x + n div x) shr 1\n            return x\n    proc chmax*[T](x: var T, y: T): bool {.discardable.} = (if x < y: (x = y; return true; ) return false)\n    proc chmin*[T](x: var T, y: T): bool {.discardable.} = (if x > y: (x = y; return true; ) return false)\n    proc `max=`*[T](x: var T, y: T) = x = max(x, y)\n    proc `min=`*[T](x: var T, y: T) = x = min(x, y)\n    proc at*(x: char, a = '0'): int = int(x) - int(a)\n    proc Yes*(b: bool = true): void = print(if b: \"Yes\" else: \"No\")\n    proc No*(b: bool = true): void = Yes(not b)\n    proc YES_upper*(b: bool = true): void = print(if b: \"YES\" else: \"NO\")\n    proc NO_upper*(b: bool = true): void = Yes_upper(not b)\n    const DXY* = [(0, -1), (0, 1), (-1, 0), (1, 0)]\n    const DDXY* = [(1, -1), (1, 0), (1, 1), (0, -1), (0, 1), (-1, -1), (-1, 0), (-1, 1)]\n    macro exit*(statement: untyped): untyped = (quote do: (`statement`; quit()))\n    proc initHashSet[T](): Hashset[T] = initHashSet[T](0)\n"
ImportExpand "cplib/collections/segtree.nim" <=== "when not declared CPLIB_COLLECTIONS_SEGTREE:\n    const CPLIB_COLLECTIONS_SEGTREE* = 1\n    import algorithm\n    type SegmentTree*[T] = ref object\n        default: T\n        merge: proc(x: T, y: T): T\n        arr*: seq[T]\n        lastnode: int\n        length: int\n    proc initSegmentTree*[T](v: seq[T], merge: proc(x: T, y: T): T, default: T): SegmentTree[T] =\n        ## セグメントツリーを生成します。\n        ## vに元となるリスト、mergeに二つの区間をマージする関数、デフォルトに単位元を与えてください。\n        var lastnode = 1\n        while lastnode < len(v):\n            lastnode*=2\n        var arr = newSeq[T](2*lastnode)\n        arr.fill(default)\n        var self = SegmentTree[T](default: default, merge: merge, arr: arr, lastnode: lastnode, length: len(v))\n        #1-indexedで作成する\n        for i in 0..<len(v):\n            self.arr[self.lastnode+i] = v[i]\n        for i in countdown(lastnode-1, 1):\n            self.arr[i] = self.merge(self.arr[2*i], self.arr[2*i+1])\n        return self\n\n    proc update*[T](self: SegmentTree[T], x: Natural, val: T) =\n        ## xの要素をvalに変更します。\n        assert x < self.length\n        var x = x\n        x += self.lastnode\n        self.arr[x] = val\n        while x > 1:\n            x = x shr 1\n            self.arr[x] = self.merge(self.arr[2*x], self.arr[2*x+1])\n    proc get*[T](self: SegmentTree[T], q_left: Natural, q_right: Natural): T =\n        ## 半解区間[q_left,q_right)についての演算結果を返します。\n        assert q_left <= q_right and 0 <= q_left and q_right <= self.length\n        var q_left = q_left\n        var q_right = q_right\n        q_left += self.lastnode\n        q_right += self.lastnode\n        var (lres, rres) = (self.default, self.default)\n        while q_left < q_right:\n            if (q_left and 1) > 0:\n                lres = self.merge(lres, self.arr[q_left])\n                q_left += 1\n            if (q_right and 1) > 0:\n                q_right -= 1\n                rres = self.merge(self.arr[q_right], rres)\n            q_left = q_left shr 1\n            q_right = q_right shr 1\n        return self.merge(lres, rres)\n    proc get*[T](self: SegmentTree[T], segment: HSlice[int, int]): T =\n        assert segment.a <= segment.b + 1 and 0 <= segment.a and segment.b+1 <= self.length\n        return self.get(segment.a, segment.b+1)\n    proc `[]`*[T](self: SegmentTree[T], index: Natural): T =\n        assert index < self.length\n        return self.arr[index+self.lastnode]\n    proc `[]=`*[T](self: SegmentTree[T], index: Natural, val: T) =\n        assert index < self.length\n        self.update(index, val)\n    proc get_all*[T](self: SegmentTree[T]): T =\n        ## [0,len(self))区間の演算結果をO(1)で返す\n        return self.arr[1]\n    proc len*[T](self: SegmentTree[T]): int =\n        return self.length\n\n    template newSegWith*(V, merge, default: untyped): untyped =\n        initSegmentTree(V, (l, r{.inject.}: typeof(default)) => merge, default)\n\n"
ImportExpand "cplib/collections/hashtable.nim" <=== "when not declared CPLIB_COLLECTIONS_HASHTABLE:\n    const CPLIB_COLLECTIONS_HASHTABLE* = 1\n    import bitops\n    import sequtils\n    import hashes\n    type State = enum\n        empty, active, inactive\n    type Node[K, V] = object\n        state: State\n        value: (K, V)\n    type HashTable*[K, V] = object\n        values*: seq[Node[K, V]]\n        len: int\n        fill: int\n        mask: int\n    proc vlen(x: int): int = (if x == 0: 4 else: 1 shl (fastLog2(x) + 2))\n    proc len*[K, V](self: HashTable[K, V]): int = self.len\n    proc initHashTable*[K, V](): HashTable[K, V] =\n        var vlen = 4\n        HashTable[K, V](values: newSeqWith(vlen, Node[K, V](state: State.empty)), len: 0, fill: 0, mask: vlen - 1)\n    iterator pairs*[K, V](self: HashTable[K, V]): (K, V) =\n        for item in self.values:\n            if item.state == State.active: yield item.value\n    iterator keys*[K, V](self: HashTable[K, V]): K =\n        for item in self.values:\n            if item.state == State.active: yield item.value[0]\n    iterator values*[K, V](self: HashTable[K, V]): V =\n        for item in self.values:\n            if item.state == State.active: yield item.value[1]\n    proc find[K, V](self: HashTable[K, V], x: K): int =\n        var sh: int = hash(x) and self.mask\n        while self.values[sh].state != State.empty and self.values[sh].value[0] != x:\n            sh = (sh + 1) and self.mask\n        return sh\n    proc add_item[K, V](self: var HashTable[K, V], key: K, val: V) =\n        var pos = self.find(key)\n        if self.values[pos].state == State.active:\n            self.values[pos].value[1] = val\n            return\n        self.len += 1\n        self.fill += 1\n        self.values[pos].value = (key, val)\n        self.values[pos].state = State.active\n    proc resize[K, V](self: var HashTable[K, V]) =\n        var vlen = self.len.vlen\n        var vi = newSeq[Node[K, V]](vlen)\n        self.mask = vlen - 1\n        self.len = 0\n        self.fill = 0\n        swap(vi, self.values)\n        for item in vi:\n            if item.state == State.empty: continue\n            var (key, val) = item.value\n            self.add_item(key, val)\n    proc incl*[K, V](self: var HashTable[K, V], val: (K, V)) =\n        self.add_item(val)\n        if self.fill.vlen > self.values.len: self.resize\n        # if self.fill > self.values.len div HASHSET_INCL_RESIZE_RATIO: self.resize\n    proc contains*[K, V](self: var HashTable[K, V], key: K): bool = (self.values[self.find(key)].state == State.active)\n    proc hasKey*[K, V](self: var HashTable[K, V], key: K): bool = self.contains(key)\n    proc `[]`*[K, V](self: HashTable[K, V], key: K): V =\n        var pos = self.find(key)\n        assert self.values[pos].state == State.active, \"Key \\\"\" & $key & \"\\\" not found\"\n        return self.values[pos].value[1]\n    proc `[]`*[K, V](self: var HashTable[K, V], key: K): var V =\n        var pos = self.find(key)\n        assert self.values[pos].state == State.active, \"Key \\\"\" & $key & \"\\\" not found\"\n        return self.values[pos].value[1]\n    proc `[]=`*[K, V](self: var HashTable[K, V], key: K, val: V) =\n        var pos = self.find(key)\n        self.values[pos].value = (key, val)\n        self.values[pos].state = State.active\n        self.len += 1\n        self.fill += 1\n        if self.fill.vlen > self.values.len: self.resize\n    proc clear*[K, V](self: var HashTable[K, V]) = self = initHashTable[K, V]()\n    proc del*[K, V](self: var HashTable[K, V], key: K) =\n        var pos = self.find(key)\n        if self.values[pos].state != State.active: return\n        self.len -= 1\n        self.values[pos].state = State.inactive\n    proc excl*[K, V](self: var HashTable[K, V], key: K) = self.del(key)\n    proc hash*[K, V](self: HashTable[K, V]): Hash =\n        for item in self.pairs:\n            result = result !& hash(item)\n"

var n, m, q = input(int)
var a, b = newSeq[int](n)
for i in 0..<n: (a[i], b[i]) = input(int, int)

proc dp(a, b: seq[int]): HashTable[(int, int), int] =
    var n = a.len
    var dp = initHashTable[(int, int), int]()
    dp[(0, 0)] = 0
    for i in 0..<n:
        var dpn = dp
        for t, v in dp:
            var (x, y) = t
            if (x + a[i], y) in dpn: dpn[(x + a[i], y)].max = v + b[i]
            else: dpn[(x + a[i], y)] = v + b[i]
            if (x, y + a[i]) in dpn: dpn[(x, y + a[i])].max = v + b[i]
            else: dpn[(x, y + a[i])] = v + b[i]
        swap(dp, dpn)
    return dp
var dp1 = dp(a[0..<(n//2)], b[0..<(n//2)])
var dp2 = dp(a[(n//2)..<n], b[(n//2)..<n])
    # debug(dp1)
    # debug(dp2)

var c = dp2.keys.toSeq.mapIt(it[0]).sorted.deduplicate(true)
var add = newSeqWith(c.len, newSeq[(int, int)](0))
for k, v in dp2:
    var (x, y) = k
    add[c.lowerBound(x)].add((y, v))
var seg = newSegWith(newSeqWith(c.len, -INFL), max(l, r), -INFL)

var qi = dp1.pairs.toseq.mapIt((it[0][0], it[0][1], it[1])).sortedByIt(-it[0])
# debug(c)
var ans = 0
var pos = 0
for (x, y, v) in qi:
    # debug(x, y, v)
    while pos < c.len and m - x >= c[pos]:
        for (yn, vn) in add[pos]:
            seg[c.lowerBound(yn)] = max(seg[c.lowerBound(yn)], vn)
        pos += 1
    var r = c.upperBound(q - y)
    ans.max = seg.get(0..<r) + v
    # var d = (0..<c.len).toseq.mapIt(seg[it])
    # debug(d)
print(ans)