resistor.  The solution is to shift sampled voltage negative. Placing two voltage references in series works. Now a third reference is added to establish a fixed voltage relative to ground. A better choice would be to use two different valued voltage references: one LM385-2.5 for attaching to the top of the cathode resistor and one LM385-1.2 for attaching to ground, as this would eliminate one voltage reference. 
The best choice, however, would be to use only one voltage reference. The circuit to the right uses four 1% resistors and one voltage reference to perform the same tasks as the three references in the previous example. The lone voltage reference's voltage will be mirrored across the cathode resistor, as the op-amplifier will move its output voltage up and down in an attempt to eliminate the voltage difference between its inputs. The only way the inputs could line up would be if all resistors shared the exact same voltage across them, which means the cathode of the voltage reference would have to be at ground potential. Since the voltage at the anode of the voltage reference is always 1.24 volts

Circuit for up to -40v of bias

greater than its cathode and since this anode is connected to the cathode of the tube, the voltage across the cathode resistor must equal the voltage across the voltage reference. This condition can only occur when the idle current through the tube develops the same voltage across the cathode resistor as the voltage reference. (The observant will note that the voltage reference and the two resistors in series with it will define a current path into the tube, which will serve to increase the current through the tube. True. But not by much: 0.4 mA, as 40v / 400K = 0.4 mA. )
The one potential setback to this circuit is that the feedback uses to set the bias is always in play, but the tube isn't, for at turn-on the tube will not instantly start conducting, as it cathode must come up to heat first. This lag can cause problems. The op-amplifier will sense that its positive input is more positive than its negative input and will move its output positively as a result. This positive moving output will soon reach a voltage just a volt or two from ground. The tube's cathode, which is normally somewhat protected in a fixed bias amplifier by the fixed negative voltage of

the grid and by the space charge made up of the cloud of electrons liberated by the hot cathode, in this case at turn-on, is now more dangerously exposed to the full B+ voltage of the plate. The attraction the plate will exert on the cathode can be so great as to strip away small pieces of material from the cathode (cathode stripping). It seems that in this respect the auto-bias of the grid voltage is inferior to the normal fixed bias method. The solution is shown in the circuit at the left this last and best variation solves this problem and handles a  much greater negative bias voltage.