KeyError: 3 while using dictionary in python

I am trying to calculate number of hydrogen bonds for water in a simulation where I have 3000 molecules of water ( 1000 oxygen and 2000 hydrogen ) . So I have written a code for it.

I have a dataframe (df1) in which I oxygen atoms at a position which is a multiple of 3 (0,3,6 ...) and at other position hydrogen atom and the x,y,z coordinates of an atom (xi,yi,zi). Now for HB calculation I need two oxygen atoms whose distance is less than 3.5 i.e. d1<3.5 and one angle which I am calculating should be less than 30 degree i.e. final < pi/6 . Everytime my both conditions are satisfied i'm increasing the count variable by one. But now here I need to keep two checks that between two molecules there can be only one hydrogen bond ( flag is used for it ) and a single molecule can form atmost four HB Bonds ( record dictionary is used for it and in checking this condition error is coming in the last if statement) .

Now I have written the entire logic , but in the ending if statement I am getting an error: KeyError: 3

import math
count=0
record={}

for i in range(0,3000,3):
  for j in range (i+3,3000,3):
    flag=0
    x1=0
    y1=0
    z1=0
    d1=0
    x1=df1['xi'][i]-df1['xi'][j]
    y1=df1['yi'][i]-df1['yi'][j]
    z1=df1['zi'][i]-df1['zi'][j]
    d1=math.sqrt((x1**2)+(y1**2)+(z1**2))
    
    if (d1<3.5):
      if(flag==0):
        for k in range(1,3,1):
            x2=0
            y2=0
            z2=0
            d2=0
            x2=df1['xi'][i+k]-df1['xi'][i]
            y2=df1['yi'][i+k]-df1['yi'][i]
            z2=df1['zi'][i+k]-df1['zi'][i]
            d2=math.sqrt((x2**2)+(y2**2)+(z2**2))

            x3=0
            y3=0
            z3=0
            d3=0
            x3=df1['xi'][i+k]-df1['xi'][j]
            y3=df1['yi'][i+k]-df1['yi'][j]
            z3=df1['zi'][i+k]-df1['zi'][j]
            d3=math.sqrt((x3**2)+(y3**2)+(z3**2))

            final=0
            final=math.acos(((d2**2)+(d3**2)-(d1**2))/(2*(d2*d3)))

            if (final<0.523):

              if j not in record:
                record.update({j:1})
              else :
                record[j]=record[j]+1

              if i not in record:
                record.update({i:1})
              else :
                record[i]=record[i]+1

              count=count+1
              flag=1
      
      if (flag==0):
        for l in range(1,3,1):
            x2=0
            y2=0
            z2=0
            d2=0
            x2=df1['xi'][i]-df1['xi'][j+l]
            y2=df1['yi'][i]-df1['yi'][j+l]
            z2=df1['zi'][i]-df1['zi'][j+l]
            d2=math.sqrt((x2**2)+(y2**2)+(z2**2))

            x3=0
            y3=0
            z3=0
            d3=0
            x3=df1['xi'][j]-df1['xi'][j+l]
            y3=df1['yi'][j]-df1['yi'][j+l]
            z3=df1['zi'][j]-df1['zi'][j+l]
            d3=math.sqrt((x3**2)+(y3**2)+(z3**2))

            final=0
            final=math.acos(((d2**2)+(d3**2)-(d1**2))/(2*(d2*d3)))

            if (final<0.523):

              if j not in record:
                record.update({j:1})
              else :
                record[j]=record[j]+1
            
              if i not in record:
                record.update({i:1})
              else :
                record[i]=record[i]+1

              count=count+1
              flag=1

    if (record[j]==4 or record[i]==4):
     break

    else:
      continue